AI infrastructure demand and broader system content support durable growth, while customer financing risks appear manageable.
Growth expectations face execution and concentration risks, while financing support and technical constraints could pressure estimates and valuation.
The stock was part of a broader GPU/ASIC bid, indicating positive sentiment in that sector.
The stock experienced a positive movement.
Downstream companies like NVDA could face more concentrated laser/DSP supply due to the potential LITE/MRVL deal.
OpenAI / GOOGL / NVDA: OpenAI plans to buy Google AI chips – Bloomberg
OpenAI plans to diversify its hardware sourcing by purchasing Google's AI chips, potentially reducing its reliance on NVIDIA.
AMKR +8.5% reversed a ~6% AH drop despite Q3 revenue and the Q4 guide missing on smartphone weakness, but the call leaned on advanced packaging prepayments (~$3B from NVDA landing in CY27) and a >$2B 2027 facilities capex step-up.
ORCL +3.5% into tomorrow’s financial analyst day after announcements today around NVDA Vera Rubin at multi-GW scale, GOOGL’s TPUs and CBRS systems coming to OCI, with CDS tightening alongside CRWV. Citi came away from Day 1 more confident on AI monetization...
Q3 annualizes to ~$20.3B, so the end-2026 $20B Springboard run rate is already in hand a quarter early (4Q guide ~$21B). Four new multibillion-dollar LTAs incl. Verizon and AT&T add carriers to the AMZN/NVDA deals from 2Q.
Barclays estimates that the company's cash flow fully covers its AI ecosystem commitments, alleviating fears about financing and circularity.
The company's cash is projected to rise significantly, and free cash flow is expected to reach substantial levels, implying that current cash plus near-term free cash flow covers commitments even after buybacks and dividends.
The company can continue using its balance sheet to accelerate AI infrastructure adoption without stressing its own liquidity, making current financing and circularity concerns manageable.
NVDA AMD AVGO interested in buying $5B in aggregate
The company's real customer-related off-balance-sheet commitments are substantial, indicating strong underlying demand for its products.
The official HBM4 pin speed has been lowered from 11.7Gbps to 9.3Gbps, impacting bandwidth.
The company is reportedly weighing a mix of 8-high HBM4 and 8-high HBM4E for Rubin Ultra, originally slated for 12-high HBM4E, due to heat and yield problems with 12-high stacks.
Both Samsung and SK Hynix need separate 8-high qualification for HBM4, which has slipped from earlier hopes for 2H26 shipments.
META -1.9% launched the first $35B of a ~$95B equity plan: $22.5B of Class A via forward sales plus ~$12.5B of mandatory convertible preferred now, then a ~$60B Class C ATM once the Class C trades in 2027, with NVDA, AMD and AVGO indicating interest in up t...
Jensen Huang described neocloud demand as "off the charts," indicating strong customer interest.
Neoclouds have strong customers and demand but need working capital.
NVDA / TSLA: Reported $2.8B investment to secure Samsung substrates – Reuters
Jensen Huang urged Nscale to accelerate delayed data-center projects that are facing equipment failures and power constraints.
The CEO noted that while neocloud demand is "off the charts," there is a scarcity of working capital to fuel that growth.
The CEO clarified that NVDA's investment in neocloud companies is different from backing them, suggesting no direct financial backstop.
The company's stock rose, benefiting from positive sentiment around increased compute capacity and bullish comments from its CEO at GTC Berlin.
The CEO stated that AI infrastructure demand remains structurally strong, driven by agents, physical AI, and industry-wide adoption.
Europe is identified as a real incremental growth market, not a lagging one, for the company.
The economics of the company's factories are improving through faster deployment and longer asset life.
The company plans a significant expansion in advanced-packaging capacity and expects 2027 test capacity to double from 2026 levels.
AMBA +4.4% spiked as much as +9.7% on a Betaville alert that NVDA may be a bidder before fading; Tae Kim in Slack: “when was the last time betaville nailed anything? 0 for last 100.”
DELL +5.3%; HPE +3.7% in sympathy with SMCI; JPM called DELL out directly as a beneficiary of easing power and financing constraints after SPCX’s guide, and both are partners on NVDA’s DSX POD.
MBLY -3.8% as RBC downgraded to Underperform, arguing carmakers’ move to central computers (NVDA, QCOM) puts its driver-assist business at risk, along with VW concentration and no new wins.
Chinese companies ByteDance, Alibaba, and Tencent placed orders for over 3 million export-compliant RTX PRO 5500 workstation GPUs, valued at over $35B at list prices, primarily for inference.
META is seeking another ~6.5GW by YE28, ORCL has resumed leasing with NVDA credit support which Cowen views as an indication of future RPO growth, and GOOGL/MSFT remain active. The bigger issue is execution: 2027 equipment slots are already oversold, build...
The company is providing credit support to Oracle, either through backstops or by leasing capacity directly and subleasing it, which is seen as evidence of the company throwing its weight behind the sector.
ORCL +1.3% (from -1.7%) on the TD Cowen checks: ORCL has returned to data-center procurement in 3Q26 with NVDA credit support and is in active vendor-financing talks, which Cowen reads as a sign of future RPO growth. That helps bulls whose FY27 case is that...
Nvidia warned partners that they might receive only 5–10% of requested quantities for Chinese orders, as total Chinese orders exceed 6 million GPUs compared to an indicated global production of no more than 3 million in 2027.
The company's stock did not react much to reports of significant orders from Chinese tech giants, and it warned partners they might receive only 5-10% of requested quantities, indicating supply constraints.
Providing credit support to Oracle is viewed as additional subordination and super negative, as it risks the company's credit and margin profile, leading to its CDS blowing out to new wides.
NVDA -29bps didn’t do much with Reuters’ report that ByteDance, BABA and Tencent ordered over 3M export-compliant RTX PRO 5500s, worth >$35B at list and mostly for inference, with NVDA warning partners they might get only 5-10% of requested quantities (tota...
SemiAnalysis argues 2027 data-center growth is materially underestimated, forecasting ~50GW of new capacity in a single year with >24GW already in advanced construction across just 40 campuses and ~60GW under construction more broadly. The key leading indic...
UBS expects another beat/raise with Sep rev/EPS of $1.27B/$2.12 and Dec guide of ~$1.32B/$2.24 vs. Street $1.27B/$2.09. Compute remains the driver: AMZN Trainium 3 shipments are modeled at ~1.8M/~2.8M units in 2026/27, NVDA networking remains strong and GOO...
We don’t disagree and think stock was helped by META underperformance today as investors shuffled their preferences into the hyperscaler prints. The CFO's payback math helps bulls who argue servers pay back in <3 years on 5-6 year lives, so negative FCF is...
Jensen Huang reiterated that AI infrastructure demand is expected to exceed component supply for years, indicating a sustained strong market for the company's products.
Jensen Huang plans to announce significant new European investments to expand support for the region's AI industry.
The stock performed better than the SOXX index, indicating relative strength within the semiconductor sector.
JPM sees networking as an increasingly critical gating factor in AI data center infrastructure, which favors Nvidia alongside other compute and foundry beneficiaries.
MS believes Nvidia screens as a beneficiary across various scenarios, including the doubling of open-weight model adoption and the repatriation of compute to on-premise, driven by security, sovereignty, and lower token costs.
Nvidia and Cisco announced a new pluggable-optics format that supports four times the bandwidth density and 1.3 times the power efficiency of existing modules, indicating innovation and potential for market leadership.
Microsoft's Scott Guthrie expects prolonged storage shortages and intense model price competition, but also indicates that Microsoft will continue to purchase chips from Nvidia, AMD, and Intel as its own Maia program expands, suggesting continued investment in AI infrastructure.
Chinese buyers are increasingly purchasing Nvidia gaming and workstation chips, with demand for over 2 million RTX Pro 5500 chips, indicating a shift in the market and continued demand for Nvidia's products despite a decline in AI-chip share.
NVDA screens as a beneficiary across scenarios, with DELL/HPE/NTAP most levered to repatriation.
NVIDIA's Vera Rubin is currently in production and reaching customer data centers, indicating successful deployment of its latest GPU architecture.
The preview of a 576-GPU Rubin Ultra NVLink domain using co-packaged optics demonstrates NVIDIA's intention to expand tightly coupled accelerator domains beyond current copper-connected racks, suggesting future optical content opportunities.
The dedicated Vera CPU rack, containing 256 Vera CPUs for agent execution and related workloads, provides NVIDIA with another avenue into data-center spending by addressing tasks not directly run on GPUs and emphasizing strong sustained per-core performance and predictable throughput.
NVIDIA's Spectrum-X Photonics is presented as a shipping-year scale-out product, while optical NVLink is part of the next-generation scale-up roadmap, indicating a comprehensive approach to addressing different networking markets and adoption schedules.
NVIDIA's MaxLPS power management discussion highlights a utilization opportunity by allowing operators to potentially install or use more compute within the same provisioned power envelope through continuous monitoring and enforcement of electrical limits.
NVIDIA is standardizing rack, cooling, power, and service interfaces through its MGX contribution, aiming to make manufacturing, installation, and servicing repeatable across partners while maintaining its system-level co-design advantage.
IDC estimates that Nvidia's China AI-chip shipments fell 44% year-over-year in the first half, resulting in a reduced market share of 34% compared to Huawei's 28%.
NVTS +4.6% after unveiling a 24kW 800V-to-12V power board built with NVDA for MGX racks Monday after the close.
The Rubin Ultra preview, while promising for future optical content, should not be considered a currently shipping 576-GPU configuration.
The power management example provided by NVIDIA contained inconsistent figures, making the precise fleet-wide uplift from the mechanism less reliable, although the mechanism itself is sound.
The comparison of assembly time for a compute tray (90 minutes vs. 1 minute) concerned only a compute tray, not an entire completed rack or commissioned cluster, which limits the broader applicability of the efficiency claim.
Management was incrementally bullish, arguing that the bottleneck for AI growth has shifted from demand to land, power, and shell availability.
The company is uniquely financeable due to GPU fungibility and secondary demand, and its concentration risk is limited as frontier labs account for only 40% of hyperscaler shipments and 20% of total revenue.
The company is exploring a deeper deal with Reflection AI, potentially an acquisition or larger investment, which could enhance its open-weight model capabilities.
Baird sees open-weight models as a major enterprise AI inflection, with PLTR positioned as the enabling layer for deployment at scale; L3Harris cited up to ~95% lower token costs using fine-tuned models like NVDA Nemotron. Vercel data shows open-model token...
Other reasons for the sell-off we heard today: NVDA’s reported Reflection AI acquisition talks raising more competition concerns. Nadella’s call for an AI “emergency brake” although investors seem increasingly desensitized to safety headlines. Elsewhere: Me...
WSJ reported AMZN is in talks to buy Decart for ~$7B; its software moves AI workloads between chips, and EBN says it could help customers migrate from NVDA to Trainium.
NVDA -1.3% held in better than the group despite MSFT’s Maia 300 (”same type of news from a Hyperscaler that pops up 2-3x a year,” per Akh in Slack here). Reminder, NVDA says its share of top-5 hyperscaler IT capex rose to 38% in 2Q26 from 11% at the s...
September Taiwan ADP exports surged +25.9% month-over-month, implying approximately $84.3B of F3Q DC revenue, a +13.9% quarter-over-quarter increase, which is supportive despite ADP being an imperfect proxy.
The company is reportedly in talks for $40 billion in Nvidia-chip financing for SpaceX, indicating significant demand for its chips in AI infrastructure.
The launch of Nvidia-powered Windows laptops, available for preorder and shipping soon, demonstrates the company's expanding presence in the consumer and developer markets.
The IPO of Nvidia-backed Firmus faltered due to insufficient demand at its proposed valuation, raising concerns about investor appetite for certain AI-related ventures.
The stock is underperforming due to its status as an OpenAI counterparty and technical weakness, with charts suggesting a potential false breakout that often precedes further downside.
SPCX / NVDA: SpaceX seeks $40B to buy chips led by Apollo – FT
The company is considered shielded from the AI power crunch impacting the chip supply chain.
Semis +10bps lagged with megacap AI up and most everything else down: AVGO +2.1%; NVDA +2.1% worked after Morgan Stanley said both are shielded from the AI power crunch, the latter hitting new all-time highs while the neocloud/BTM complex reversed early gains (NBIS -...
AVGO +2.1%; NVDA +2.1% worked after MS said both are shielded as the AI power crunch hits the chip supply chain.
CRDO -2.8% faded an early +1% to as much as -5.8% with no news (members were asking here). We heard Bluefin was out today saying CRDO working closely with NVDA to qualify 800G ZF module for neoclouds. Bulls say that the CPO-cannibalization worry is overdone...
Morgan Stanley believes the company is shielded from the AI power crunch impacting the chip supply chain.
Semis +10bps lagged with megacap AI up and most everything else down: AVGO +2.1%; NVDA +2.1% worked after Morgan Stanley said both are shielded from the AI power crunch, the latter hitting new all-time highs while the neocloud/BTM complex reversed early gains (NBIS -...
AVGO +2.1%; NVDA +2.1% worked after MS said both are shielded as the AI power crunch hits the chip supply chain.
CRDO -2.8% faded an early +1% to as much as -5.8% with no news (members were asking here). We heard Bluefin was out today saying CRDO working closely with NVDA to qualify 800G ZF module for neoclouds. Bulls say that the CPO-cannibalization worry is overdone...
The company is a top pick due to customer breadth, hyperscaler share gains, and agentic demand reinforcing the upside case, with management expecting revenue per GW to rise from ~$40B towards >$50B with Feynman as land/power/shell become the constraint, favoring the company's higher token density.
Agentic workloads are expected to lift CPU demand but drive significantly more GPU spend, potentially ~10x the token usage of human users.
The company anticipates CPU revenue ramping from ~$20B in 2026 to ~$40B in 2027.
In Tech, Semis leading the way higher as NVDA +1.7% close to all-time highs this morning on the back of MS making it a top pick after NDRs this week; STX/WDC ~-10% on news Toshiba is investing ¥60bn to double its production capacity for hard disk drives while targeti...
MS says after hosting Jensen Huang, CFO Colette Kress and IR Toshiya Hari, NVDA’s customer breadth, hyperscaler share gains and agentic demand reinforce the upside case. Management sees revenue per GW rising from ~$40B toward >$50B with Feynman as land/powe...
NVDA also sees CPU revenue ramping from ~$20B in 2026 to ~$40B in 2027; MS reinstates NVDA as Top Pick, OW/$300.
CBRS: Freedom upgrades to Buy/$209 after -17% selloff, sees OpenAI/NVDA fears overdone
Freedom says CBRS shares fell ~17% on reports OpenAI is using NVDA GPUs for its “Ultrafast” GPT-1 Sol model, but expects plenty of room for CBRS alongside NVDA, AMD and AWS, including potential architectures where GPUs handle prefill and CBRS WSE handles de...
Lenders are pushing for stronger guarantees on GPU-backed loans, questioning the long-term collateral value of GPUs, as banks typically underwrite chips over three to four years, while the company argues they can generate revenue for a decade.
Some loan structures may require investment-grade customer revenue rather than relying primarily on chip resale values.
Management sounded incrementally more bullish on demand, with potential upside to the guided 70%+ CY27 growth.
The company's share of hyperscaler IT capex could rise to mid-40s% from ~40% today, indicating underappreciated market share gains.
Gross margins are expected to hold better than feared, contributing to a positive financial outlook.
The NDR deck implies significant upside to hyperscaler revenue in both CY26 and CY27, exceeding current estimates.
Analyst recaps from the NDR indicated increased bullishness from management, with potential for over 70% CY27 growth and an increased share of hyperscaler IT capex.
Some good discussion in Slack here around LITE’s outsized strength. n asked if it was rumors of a new ELS customer, pointing to LITE saying at ECOC that one emerging ELS customer could ultimately be a bigger opportunity than NVDA given the optical inte...
NVDA +1.1% on the NDR recaps: Cantor says mgmt sounded incrementally more bullish, with upside to 70%+ CY27 growth and NVDA’s share of hyperscaler IT capex potentially rising to the mid-40s% from ~40%, while Barclays’ read of the NDR deck implies $30B+ of u...
CRDO +7.9% still riding Jefferies' NYC NDR (SPCX potentially adding another 660K GB300s; OmniConnect ramping in 2H27). Irrational Analysis also added the name to his portfolio and had some positive comments, saying CRDO’s SerDes portfolio is second only to...
Supply remains the key constraint, with a large portion of tracked GW still under construction, limiting immediate growth potential.
Cantor says after hosting Jensen Huang, CFO Colette Kress and IR Toshiya Hari, management sounded incrementally more bullish on demand, with upside to guided 70%+ CY27 growth and NVDA’s share of hyperscaler IT capex potentially rising to mid-40s% from ~40%...
GLW: MS says AT&T LTA reinforces optical demand, with capacity ramp the key swing factor MS says AT&T’s >$3B multiyear deal, Verizon’s >80M fiber-mile commitment through 2032 and an NVDA-backed build extend demand visibility and support a return toward MSD...
CBRS -8.9% after SemiAnalysis said last night that OpenAI’s GPT-6.1 Sol Ultrafast “is NOT running on Cerebras but… at a low batch size on NVIDIA GPUs” (@SemiAnalysis_ on X), a day after most read Ultrafast as a CBRS win. Isaac S. in Slack here reread a...
Nvidia is working with Howden Re on potential insurance against losses when neocloud borrowers default and their GPU collateral is worth less than the outstanding loans, which could unlock more AI financing.
Stifel says AMD is effectively buying a seat at the model frontier, with World Labs adding spatial/world-model expertise for 3D simulation, robotics and physical AI while Dr. Fei-Fei Li joins as Chief Scientist. The strategic value is tighter model-to-hardw...
The company's stock was one of only three to be green over the past month among SOXX or SMH components, indicating relative strength.
Nvidia is lobbying the White House to oppose proposed restrictions on advanced chip exports to China, indicating a potential negative impact from such restrictions.
The company's stock remained a laggard, underperforming peers over the past month.
The company added $150 billion to its existing share buyback authorization, indicating confidence in its financial position and future prospects.
Chinese officials are considering approving purchases of Nvidia’s RTX Pro 5500 chips by Alibaba and ByteDance, with ByteDance potentially ordering 1 million units, suggesting strong demand from key Chinese tech companies.
Nvidia aims to begin shipments of RTX Pro 5500 chips to China in late December and supply about 500,000 chips per quarter, indicating a significant new revenue stream.
The company introduced the NVIDIA Open Agent Safety Platform, combining open-source security tools with its hardware, which enhances security for AI agents and could prevent breaches, allowing for safer testing of advanced models.
Citi says ECOC/management meetings reinforced that OCS is evolving from a Google-centric niche into a multi-customer AI optics market, sizing merchant TAM at ~$11B by 2030: ~$5B scale-up, ~$2B spine/other applications and another ~$4B longer-term reconfigur...
NVDA / DELL / SMCI / Wistron: Vera Rubin deliveries build toward Q4 ramp – Commercial Times NVDA / DELL / SMCI / Wistron Vera Rubin
The company announced a significant $150 billion share buyback, the largest in U.S. history, and launched an Open Agent Safety Platform.
NVDA was the standout +1.7% in semis after announced $150B buyback on top of their existing authorization. Software -55bps held in far better than the headline of the day would suggest (META going after enterprise AI with MDB’s -18% CEO) as cyber ripped and...
NVDA +1.7% after adding $150B to its buyback (to $235B, the largest-ever US authorization), launching an Open Agent Safety Platform on a day AI-agent breaches were in the headlines, and The Information reported China is weighing approvals for BABA/ByteDance...
Cyber ripped off the open: PANW +4.6%; OKTA +3.6%; ZS +3.3%; CRWD +2.8%; S +1.6%; FTNT +1.7%; NET +1.5%. No bid pre-market despite a weekend of AI-agent breach headlines, but NVDA named CRWD and PANW as security partners on its agent safety platform.
The stock is performing well, remaining close to its highs within the AI infrastructure trade.
Conversations with various companies consistently indicated strong AI infrastructure demand despite ongoing debates about AI regulation.
Memory and storage demand continues to exceed supply.
Agentic CPU demand is expected to broaden from head nodes and traditional servers in 2027 to sandbox workloads in 2028, benefiting next-generation Vera, Venice, and Diamond Rapids CPUs.
CXL is gaining traction, with CXL 2.0 ramping at Meta and Microsoft, and CXL 3.0 being crucial for broader adoption.
Management pushed back on custom-ASIC concerns, stating that FY27 growth remains approximately 70% supply-constrained across various components, indicating robust demand.
Management expects open models to expand compute demand rather than weaken the company's competitive advantage.
BNP Paribas says conversations across AMD, ANET, AMZN, AVGO, NVDA and others consistently pointed to AI infrastructure demand remaining strong despite calls for tighter AI regulation.
BNP says broader 200G/lane 102T switch ASIC support from MRVL/AVGO/NVDA/CSCO should enable the 1.6T optical transceiver ramp beginning CQ4’26 across Vera Rubin, MI455x Helios and Google TPUv8.
Agentic AI workloads are significantly more demanding than previous chat models, reinforcing a major step-up in compute intensity.
Vera Rubin offers significant improvements in token throughput and AI-factory throughput on agentic workloads through full-stack co-design.
Efficiency gains from new technologies should expand usable AI workloads rather than reduce infrastructure demand.
The company is part of an emerging "Wintel" model for enterprise AI, owning the application/control layer.
The company's residual value guarantee for vendor financing is capped at around 25%, which is lower than competitors, suggesting a more conservative and potentially less risky approach to financing.
The company is being very selective with its revenue share agreements, using Firmus as a template where it captures about 50% of the delta between agreed baseline and higher market rates, indicating a strategic approach to maximizing profitability while managing risk.
The company's 2028 guide was aimed at securing better wafer allocation and maintaining gross margins from TSMC, suggesting a proactive strategy to manage supply chain and input costs.
NVDA +1%: IR was out talking to sell-side and investors. One of the key takeaways passed along to me from the calls was that IR was more specific on the vendor financing mechanics: NVDA’s residual value guarantee is capped around 25% — vs AVGO’s first tranc...
The pre-release Vera Rubin is already delivering significantly higher token throughput per MW compared to Blackwell, exceeding NVIDIA's initial claims, with further improvements expected as software matures.
Rubin's higher upfront and rental costs are more than offset by superior inference economics, which supports rapid upgrade adoption and continued pricing power for NVIDIA.
Semianalysis touted Rubing Economics, stating that Vera Rubin is already delivering up to 7x Blackwell's token throughput per MW, compared to NVIDIA's claim of 3x, with further upside as software and kernels mature.
Frontier labs continue to make major infrastructure commitments, suggesting near-term compute procurement should remain strong even if the new safety debate pressures multiples.
Current semi estimates effectively require ~380GW of AI capacity additions from 2026–30, making any slowdown in powered-shell deployment a meaningful risk to out-year numbers.
The estimates already assume a bullish conversion of unpermitted grid-connected and off-grid capacity, leaving limited room for execution slippage.
Piper ranks MRVL alongside NVDA and AVGO as a long-term winner in XPU compute and says Marvell leads in custom attach. Piper also flags the upcoming analyst day as a potential catalyst to sharpen the growth framework and custom-silicon opportunity.
The company reaffirmed strong growth expectations and highlighted increasing per-system content value in its next-generation hardware architectures.
Regulators are investigating Nvidia's licensing deal with Groq, with the DOJ probing whether the $17B deal was structured to avoid antitrust review, putting the increasingly common AI "acqui-hire without an acquisition" structure under scrutiny.
The company has reportedly raised prices to bundle and test interface capacity, which could impact customers.
Huawei has unveiled a new optical technology standard, near-package optics (NPO), which it claims is more feasible than Nvidia's co-packaged optics (CPO) method and leverages existing chipmaking supply chains, posing a challenge to Nvidia in AI computing.
Morgan Stanley prefers beneficiaries where stronger inference meets limited supply or rising complexity, including NVDA.
Earnings revisions remain strongest in AI compute, networking, and non-AI sectors.
Iren's CEO, an Nvidia partner, argues that AI compute supply may never meet demand, with new capacity potentially creating more demand and power/permitting constraints limiting oversupply risk.
Nvidia is increasingly intertwined with Iren as a supplier, customer, and potential shareholder, highlighting the aggressive mobilization of capital behind neocloud build-outs.
MS prefers beneficiaries where stronger inference meets limited supply or rising complexity, including NVDA, MediaTek/GUC, GLW/LITE/COHR and CXMT, while favoring STM/NXP/Renesas for non-AI diversification.
SemiAnalysis's benchmarks for TPUv7 Ironwood inference show up to ~50% better performance/$ compared to B200/B300 on FP8 aggregated serving, driven by lower TCO rather than superior raw chip performance.
Google is externalizing the TPU software stack, with native TorchTPU support for vLLM/SGLang targeted to open-source around mid-October, which will materially lower the CUDA/software-moat barrier for outside customers.
NVIDIA still leads in some important cases, particularly FP4 and optimized disaggregated serving, but SemiAnalysis expects those gaps to narrow as Google adds disaggregation/spec decoding, and TPUv8i adds native FP4 and an inference-specific network architecture.
Nvidia is expanding its role as a financier and credit backstop for AI infrastructure, enabling non-investment-grade customers and neoclouds to secure significant compute resources.
The company's FY28 revenue growth guidance of approximately 70% significantly exceeded Street expectations, implying higher EPS for C27 and C28.
The company's decision to absorb some memory inflation rather than passing it all through, despite a gross margin step-down, was viewed positively as it addresses a potential negative head-on.
The stock is considered undervalued at 10x C28 earnings for a leading AI factory platform, suggesting it does not fully account for growth beyond 2028.
The company's strong demand, with customer forecasts pointing to approximately 100% growth, indicates that supply, not demand, is the primary bottleneck, alleviating near-term concerns about AI demand rolling over.
The F3Q27 revenue guidance of $108.0B at the midpoint exceeded Street expectations, and the revenue upside more than compensates for the lower margin framework at the EPS level.
The Vera Rubin product ramp is expected to be the fastest in the company's history, with production shipments already started and purchase orders from all major hyperscalers, AI clouds, and system OEMs.
The full-stack revenue opportunity per gigawatt is significantly increasing with new generations of chips, providing a major content tailwind.
The broadening customer mix, with ACIE now representing approximately 45% of Data Center revenue and expected to reach roughly half, diversifies the company beyond concentrated hyperscaler CapEx cycles.
The company's use of its balance sheet to invest in frontier AI labs and provide financing platforms supports demand and accelerates infrastructure deployment, with management arguing the risk is limited due to fungible compute.
The neocloud revenue-sharing structure is strategically important, potentially producing billions of recurring, usage-linked revenue over time.
The company's competitive differentiation lies in CUDA, networking, system design, and global availability, with most major open models running on its platform, making open-model share gains additive.
Grace CPU revenue exceeded $5B on a trailing-12-month basis, and preliminary FY28 CPU revenue is expected to more than double, indicating strong growth in its CPU offerings.
The company returned a record $26B to shareholders, including approximately $20B of buybacks, demonstrating a commitment to shareholder returns.
The company is transitioning from selling discrete accelerators to providing a complete AI-factory platform, including GPUs, CPUs, networking, systems, and software, which strengthens its market position.
Agentic AI significantly increases compute demand, and the company's fungible architecture can be used across various AI tasks, further strengthening its thesis.
The FY28 growth guide of approximately 70% exceeded even bullish buyside expectations by roughly 10 points, indicating stronger-than-anticipated growth.
The company can support extraordinary unit and content growth even after absorbing part of the memory inflation, demonstrating its ability to manage costs while expanding.
The apparent ACIE shortfall was clarified as a customer moving to hyperscale, and both ACIE and hyperscale showed strong growth, indicating broad demand.
The balance sheet strengthens the company's moat by allowing it to reserve scarce components, secure land and power, and help infrastructure partners obtain capital.
The stock is considered too cheap at approximately 13x FY28 earnings and potentially below 10x FY29 earnings, especially with a potential 10% post-FY28 earnings CAGR.
The acquisition of Hugging Face for $12.9B is strategically important for controlling a key open-source AI distribution hub and strengthening open models as a counterweight to competitors.
The Hugging Face acquisition could revive the company's cloud ambitions by providing a channel to monetize excess compute tied to its growing credit-support commitments.
The acquisition aligns with the CEO's aggressive push to invest across the AI ecosystem to lock in demand for the company's hardware.
Bulls will say you are getting NVDA at 10x C28 earnings for the marquee AI factory platform which means you aren’t paying for any growth past ‘28. Bears will say some of the debates heading in still aren’t solved (TPU/AMD competition, GM pressure, SPCX/OAI/...
The stock remains attractive due to its dominant AI platform and growth potential, with bear arguments regarding competition and margin pressure being overstated.
Large cap names up in sympathy with NVDA: TSM +2.3% and AVGO +4.5%
The gross margin step-down to 72-73% was a negative, and could be even lower when excluding neocloud revenue-share flows.
Some bear debates, such as TPU/AMD competition, GM pressure, and customer concentration, remain unresolved.
The gross-margin reset was a primary negative, with F3Q GM falling to 74.0%, F4Q expected to trough at 71%-72%, and FY28 settling at 72%-73%, below prior frameworks.
Memory pricing has risen more than expected, and the company is deliberately absorbing part of the inflation, which impacts margins.
Hyperscaler and frontier-lab ASIC efforts represent the clearest structural share risk, particularly for high-volume, narrowly defined inference workloads.
Free Cash Flow fell significantly due to increased inventory and higher Days Sales Outstanding, raising concerns about cash flow management.
The bear case has shifted to concerns about who funds AI demand and bears the residual risk, as the company is increasingly taking on risk through various commitments.
A significant portion of future growth, potentially 25% of FY28 revenue, is expected to come from AI labs where the company leverages its balance sheet, making this growth less 'clean' than hyperscaler CapEx.
The 70% growth guide may be inflated by pricing and mix, with underlying unit and workload growth potentially closer to 50%-55%.
The incremental growth is disproportionately tied to specific customers like OpenAI, SPCX, and newer clouds, rather than the existing hyperscaler base, indicating potential concentration risk.
High-margin revenue-share income could flatter consolidated gross margins, making underlying hardware economics weaker than the headline figures suggest.
Extreme HBM inflation may accelerate de-speccing, flash substitution, and more memory-efficient custom silicon, potentially turning a current shortage into a future architectural shift.
The bear case suggests that growth may become more financed, price-driven, and less durable, while hardware margins and competitive positioning quietly weaken.
JPM's OpenRouter data shows that while AI usage is strong, rapidly falling model pricing and the first cracks in leading-edge GPU rental economics are worth watching.
B200 rentals declined sequentially for the first time in over six months, and the premium versus H100 narrowed, hinting at some rotation toward older capacity.
Cleveland is positive on the company, anticipating CY27 ASPs to increase by over 30% and potential for higher shipments due to memory and bill of materials changes.
The Rubin Ultra will ship with a significantly reduced HBM content of 192GB of HBM4 8-hi, a substantial downgrade from the originally previewed 1TB and even lower than the regular Rubin's 288GB, due to cuts in the number of cubes and stack height.
An Nvidia senior manager and Super Micro employees were charged with smuggling 130 Nvidia B300 AI servers to China, with 74 ultimately reaching Chinese customers, by disguising a Taiwan data center as the end customer to evade checks.
Semianalysis's note on OpenAI's Jalapeno performance is incrementally negative for NVDA, as it suggests OpenAI is no longer solely reliant on GPUs, weakening the CUDA moat.
The competitive performance of Jalapeno could pressure NVDA's long-term margins by shifting negotiating leverage, even if Rubin remains the benchmark.
Consensus suggests that the three debates holding back the multiple—share loss to TPU and AMD, circular financing, and the gross margin trajectory beyond this year—are 2027 issues and unlikely to change with the upcoming earnings print.
There is significant debate regarding the gross margin trajectory due to headwinds from memory and recent de-spec'ing.
Cantor reiterated an Overweight rating and a $350 price target, arguing that investors are wrongly treating AI compute as a zero-sum game, as Nvidia retains resilient revenue share by monetizing networking, systems, software, and increasingly financing around the GPU.
Nvidia's balance sheet, equity stakes, and financing arrangements with neoclouds/labs are creating a new competitive moat by making GPUs a financeable asset class and funding customers who then buy more Nvidia infrastructure.
Cantor sees a path to over $1 trillion in Data Center revenue and $25-30 of EPS by 2030, making Nvidia a Top Pick, with stretch CY27/28 EPS of approximately $16-17/$23-25 leaving the stock at only approximately 13x/9x.
Nvidia is discussing investing billions in Perplexity at a valuation over $30 billion, up over 50% from its last round, as Perplexity's ARR has tripled year-to-date to over $750 million.
Nvidia and Perplexity have deepened ties through Nvidia's Nemotron Coalition, meeting multiple times weekly on hardware/software development, as Nvidia aggressively builds out its own open-model ecosystem.
Nvidia plans to use its $6 billion Poolside licensing/hiring deal to build one of the world's most powerful open-weight AI models, aiming to make Nemotron competitive with DeepSeek, Kimi K3, and eventually frontier U.S. models.
Nvidia will invest $1 billion in Poolside at a $12 billion pre-money valuation and hire over 100 engineers, while Poolside's founders remain independent.
The push into open-model ecosystems reflects Jensen Huang's view that a strong U.S. open-model ecosystem is critical to counter China and ultimately drive more demand for Nvidia hardware.
The company's price increases are seen as a bullish demand signal, indicating strong underlying demand for its products.
Nvidia is raising prices on AI server systems due to surging DRAM costs from Samsung, SK Hynix, and Micron, indicating that even the industry's most profitable chip company is passing higher costs to customers.
The price increases could add further pressure to already-expensive AI infrastructure buildouts and potentially create openings for hyperscalers' in-house chips, though those efforts remain constrained by memory shortages.
The company is increasingly competing with major chip customers like OpenAI and Anthropic.
The company's price increases are viewed as a cost burden on an ecosystem increasingly reliant on debt financing, suggesting potential financial strain for customers.
The stock has fallen 11.9% from its high since the "Leopold Low," indicating a significant pullback.
Sentiment has deteriorated enough to surprise even us, and we can't help but feel we're approaching a psychological low, especially with NVDA earnings, Jackson Hole, the September tech conference circuit and Anthropic's S-1 all ahead. Our base case since th...
| Basket | Since Leopold Low | Semis | Ticker | % Off High | | :-------------------- | :---------------- | :------ | :----- | :--------- | | Global Memory | 20.7% | | NVDA | -11.9% | | AMZN Trainium Infra | 19.8% | | ANET | -12.4% | | Neoclouds/Miners | 19....
BMO initiated coverage with an Outperform rating and a $340 price target, highlighting over 12 months of largely sold-out capacity, a differentiated full-stack AI platform, and strong margins and free cash flow.
ADI/NVDA/MU/MCHP/SWKS: BMO initiates ADI/NVDA/MU/MCHP at Outperform, SWKS at Market Perform ADI/NVDA/MU/MCHP/SWKS BMO ADI NVDA
BMO starts NVDA at Outperform/$340, highlighting 12+ months of largely sold-out capacity, a differentiated full-stack AI platform and strong margins/FCF; MU at Outperform/$1,300 on a prolonged memory supercycle driven by tight supply and broad-based demand....
Jefferies expects a beat and raise, forecasting July revenue of approximately $95 billion (vs. Street $91.9 billion) and October guidance of approximately $108 billion (vs. $103.7 billion).
The Rubin ramp is expected to be faster, becoming more material in December and potentially overtaking Blackwell as the dominant contributor by F1Q28, with Jefferies modeling approximately 2 million Rubin units in F1Q28.
OpenAI’s existing and planned commitments represent approximately 12GW of NVDA compute, expandable to approximately 16GW or ~$600 billion through 2030, providing a concrete demand anchor.
Edge is positive on NVDA, noting that Rubin demand forecasts continue to increase.
Jefferies flagged potential CPO delays and power risks.
NVDA is expected to raise GB2/300 prices by as much as 20-30% to offset rising memory costs.
China has begun allowing small batches of Nvidia H200 chips into the mainland, with ByteDance and Tencent each receiving approximately 10,000 chips, indicating a recognition of the need for Nvidia GPUs for cutting-edge AI training.
Citi raised Jul/Oct-Q sales estimates by approximately $1 billion/$1.5 billion above Street expectations due to stronger transceiver shipments.
The B300 remains strong, and the Rubin platform is ramping faster than expected.
AI financing risk is viewed as manageable, as Nvidia is using its balance sheet to alleviate bottlenecks for labs and neoclouds.
Citi raised FY27-29 EPS by 1%/2%/2% and expects the stock to trade higher post-earnings, maintaining a Buy rating and a $300 price target.
Beijing is still limiting most H200 imports to support domestic suppliers like Huawei, indicating ongoing restrictions.
Citi trimmed Rubin unit assumptions due to memory supply constraints.
The company's strategy to secure AI inputs is valuable, and its SOTP valuation implies a significant discount to peer-adjusted value even after accounting for financing risks.
Increasing capital returns, specifically by lifting the share of FCF directed to buybacks/dividends, could improve perceived earnings quality and the stock's multiple.
The strategy of securing all AI inputs increases balance-sheet risk if demand slows.
The company is expected to deliver another beat-and-raise quarter, with strong revenue projections for July and October.
The Rubin platform is expected to preserve NVIDIA’s performance leadership, and compute demand continues to outstrip supply.
Blackwell demand is expected to remain strong into next year, with additional upside from networking, new products, and revenue-sharing streams.
The stock offers a compelling risk/reward at approximately 17x FY28 EPS.
The company is a positive read-through from strong 2027 AI-server demand, with Vera Rubin potentially exceeding 50K racks and production ramping from late 2026.
Mizuho says 2027 AI-server demand remains strong, with Vera Rubin potentially exceeding 50K racks and production ramping from late 2026, led initially by xAI and Meta. Mizuho sees the biggest incremental content shifts in networking and memory: NPO begins r...
Morgan Stanley expects another beat-and-raise quarter, modeling ~$91.2B of July revenue and ~$102.3B for October, but says the print itself may not be the catalyst given NVDA has sold off the next day after each of the last four quarters despite strong numb...
The stock has sold off the day after each of the last four quarters despite strong numbers, suggesting the earnings print itself may not be a catalyst.
The company has reworked a deal with OpenAI to materially reduce its upfront risk for a data-center campus, initially guaranteeing less than $120 billion of financing, down from a previously discussed $250 billion backstop.
UBS reiterated a Buy rating and a price target, raising estimates again, expecting the company to beat FQ2 revenue and guide FQ3 higher due to continued Blackwell strength and Rubin contributions.
UBS lifted CY26/CY27 revenue estimates, noting improved CoWoS/HBM supply visibility supporting a more constructive unit outlook, and rising hyperscaler capex expectations.
UBS stated that the company's new large third-party financing platform could support significant additional AI infrastructure capacity while shifting funding risk off its balance sheet, reinforcing confidence in future EPS targets.
Morgan Stanley reiterated Overweight/Top Pick and a price target, suggesting that the company's neocloud model could create a meaningful recurring revenue stream with high margins and limited downside.
Morgan Stanley sees significant EPS upside even with moderate deployments, with much of the revenue near 100% gross margin.
Morgan Stanley acknowledged circularity concerns but noted that GPU-backed economics and the new third-party financing vehicle significantly reduce direct balance-sheet risk.
The new $500B AI-factory financing structure is a net positive because six sophisticated third-party investors provide most of the capital and underwriting, while NVIDIA retains optionality to co-invest up to roughly 25%. This reduces circularity concerns and accelerates sovereign and neocloud deployments.
Potential revenue-sharing economics from the financing structure could create a 100%-gross-margin annuity stream and drive more than 10% upside to FY29 EPS.
The $500B+ financing platform with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR is a cleaner structure than direct vendor financing because third-party investors bear most of the underwriting, while NVIDIA supports asset value through fungible, transferable CUDA GPUs with strong residual economics.
The financing structure should help labs, neoclouds, and sovereigns finance GPUs, power, and data centers, extending the runway toward a $1.7T+ CY30 AI systems TAM, and the setup remains structurally bullish into Vera Rubin.
The financing mechanism, where NVIDIA backstops loans, unlocks chip sales by enabling smaller players to build data centers like larger giants, effectively providing "free sales juice" as long as the AI economy remains robust.
The company's financing deal helps customers overcome the high upfront cost of GPUs, making NVIDIA's solutions more accessible and strengthening its market position.
The financing strategy allows NVIDIA to maintain market share without compromising its high gross margins.
The $500 billion financing is likely to benefit neoclouds and other emerging AI players, helping them scale against larger hyperscalers.
NVIDIA is essentially co-signing loans that customers use to buy its own product, booking revenue today but quietly retaining some risk. If the AI economy stumbles, past "sales" could become losses, similar to Lucent in the dot-com era.
The entire financing model rests on the assumption that used GPUs will retain sufficient value to be seized and resold in case of loan defaults.
The company's upfront GPU costs are significantly higher than alternatives like ASICs, which can be a barrier for customers despite potentially better payback.
The financing structure involves taking on residual cost risk and could be seen as a move to preserve market share amid increasing competition and potential erosion of the CUDA moat.
The increasing use of aggressive financing structures to fund AI infrastructure raises concerns about risk building in the credit markets, especially if the underlying revenue base is not yet proven.
Some believe the financing deal suggests NVIDIA sees its market share topping out due to the CUDA moat and system limits, indicating a defensive move rather than pure demand.
unclebobs Today at 5:43 AM Alts financing this project doesn't seem like a bad place to be * Debt financing @ +HSD% yield (loosely referencing CRWV's recent DDTL facility) * Ultimately taking on NVDA's credit risk * Simplistically assuming each equally cont...
The glass-half-full interpretation is that private capital continues to support AI infrastructure build-out, easing concerns about funding availability constraining data center deployment.
TSEM flat bucked the optical weakness as Jukan was out saying on X over the weekend saying it was his favorite idea as co. benefits from NVDA NPO usage. Irrational analysis was also positive.
The stock lagged, even though some flows were expected to go to it, indicating a broader shift in investor preference outside AI semis.
The glass-half-empty interpretation is that the scale of financing required increasingly highlights the AI capex cycle's dependence on creative financing, leverage, and healthy credit markets.
In early action, software -3% leading the way lower while Semis 0.5% as NVDA +1% helping prop up the space. App sw names down 4-5% on the back of HUBS and FIG disappointing prints. Some hyperscalers like AMZN/META helping prop up the QQQs as well as AAPL +1%
Musk with interesting datapoints during this call. He said SPCX is an NVDA exclusive shop as he thinks Rubin Ultra is the best machine — no ASIC build = higher priority from Jensen. He also said he is ramping FY27 capacity to 10GW, implying 5+GW of new capa...
Elon Musk's comments about xAI's compute ramp and exclusive reliance on NVDA for its AI machines are seen as a positive catalyst for the stock.
Post-close ANET/ALAB delivering the goods while fell AMD short of bogeys. These comments from Elon have NVDA +1% in the post:
Wells Fargo reiterated a Buy rating and a price target of $315, raising Data Center estimates for FY27/FY28/FY29 due to Vera Rubin capacity and monetization visibility, which could reaccelerate sentiment.
NVDA/AMD/AVGO: Jefferies Fubon Raises 2027 CoWoS Outlook as AI Packaging Demand Broadens
NVDA/ORCL/AVGO: Morgan Stanley Credit Says NVIDIA Backstop Validates Ecosystem Financing but Is Not a Direct ORCL Positive
Jefferies Fubon expects TSMC CoWoS capacity to reach roughly 180kwpm by end- 2027 and 220kwpm by end-2028, with NVDA remaining the largest customer, AMD allocation rising toward 300k wafers in 2027 and AVGO/MediaTek also gaining capacity. Jefferies Fubon fo...
Hard looking at the GPL log chart on Bloomberg, but SNDK now down close to about 50% from the peak at sitting at 61.8% fib retracement on the chart and just blew past what we thought was a decent r/r near $1,500. Memory trade continues to suffer from downti...
The potential $250 billion guarantee to support OpenAI's data center lease and separate financing of up to $350 billion for GPU purchases raises concerns about financial risk if AI growth or investor appetite weakens.
The stock declined after outperforming leading into AI day, with some investors selling the news post-event, especially given INTC's poor price action after its earnings.
Clev is positive on the company, noting that ASP increases are driving further upside through CY27.
The company's Groq 3 LPX cabinets are expected to ship in the second half of 2026, introducing Q-Glass quartz fiber fabrics to drive high-end CCL, copper foil, and PCB stockpiling.
Japan plans to purchase 27,500 next-generation Rubin chips to build a homegrown foundational AI model for robots.
Blackwell momentum is accelerating, and the company has confirmed that next-generation Vera Rubin hardware is on track for production.
KeyBanc reiterates Overweight, raises estimates and lifts its price target to $155 from $110, citing outsized server CPU demand from agentic AI, with Intel capacity supporting 25%–30% unit growth this year and more than 50% next year. The firm says 18A yiel...
The company's CoWoS supply in 2027 is projected to reach 1.1 million units, indicating strong future capacity.
The CEO's recent investor meetings conveyed a particularly bullish outlook, with one prominent frontier model shifting from ASIC to nearly 50% NVIDIA exposure, implying increased adoption by companies like Anthropic.
Recent developments from META and xAI have countered the bear case that Anthropic's rise challenged NVIDIA's need for viable frontier lab customers and customer lock-in.
The dynamic of NVIDIA being a funding short for high-beta semi trades is reversing, suggesting a potential upward movement for the stock.
The stock may be poised for an upward 'pain trade' as investor resignation about its choppiness sets in.
The narrative is shifting with incremental positive tailwinds after a period of negative news regarding margins, TPU competition, and moat fears.
The company is increasingly recognized as a major beneficiary of the open-source trend, as open-source models are expected to run optimally on its hardware due to its established platform ecosystem.
The stock's chart is becoming more appealing if it can surpass the $212-$213 price level.
The company has significantly reduced the number of Asian customers authorized to buy its AI chips due to stricter compliance checks aimed at preventing sales to China.
The renewed vetting process excluded more than half of the company's previous customers, indicating a substantial impact on its customer base in Asia.
The stock has been flat for nine months, and the opportunity cost of investing in it compared to other stocks like SNDK has likely resulted in net negative returns.
It is difficult to envision the stock having a strong run while the semiconductor sector remains volatile.
The company is identified as a primary GPU/ASIC beneficiary in the AI data center market.
The stock experienced a decline.
The company remains a top semiconductor pick, with management meetings reinforcing accelerating growth across diverse AI segments, including AI labs, hyperscalers, neoclouds, enterprises, sovereigns, networking, and CPUs.
The company's demand is diversified, with AI labs representing about 20%, and traditional hyperscalers and the broader enterprise/sovereign segment each accounting for about 40%, reducing dependence on any single customer group.
Management is confident in the Vera Rubin roadmap, views memory shortages as manageable, and believes the company's full-stack economics and broadening end-market exposure should ultimately close its valuation gap versus peers.
TD Cowen is incrementally confident in the durability of AI spending, viewing the company as materially undervalued at less than 20x 2027 EPS and maintaining it as a Top Pick.
The company's opportunity is framed too narrowly, as only about 20% of its current business supports frontier models, with a low-to-mid teens percentage coming from diversifying enterprise verticals building domain-specific intelligence.
The shift from selling chips to operating 'AI factories' with a new revenue-sharing model for startups and regional players, along with Nemotron's vertical open-source models, is an underappreciated competitive edge.
Compute remains in shortage, with legacy GPU rental prices rising and cloud deals signing at premiums.
The Kyber Rubin Ultra roadmap and Computex NVLink domains are fully intact despite delay chatter, indicating continued strong product development.
Jensen's $100B-per-1GW comment reflects future energy efficiency, with revenue intensity per GW rising as ROI increases, suggesting improved profitability.
The company sees the non-hyperscaler market (neoclouds, sovereigns, enterprise, physical AI) ultimately becoming bigger than hyperscalers.
Mid-70s gross margin guidance holds despite memory long-term agreements, indicating strong profitability.
Buybacks are likely to increase given the current stock valuation.
China plans to allow top AI firms to purchase a limited amount of H200 chips, indicating continued demand from a key market.
The stock's approximately 30–35% P/E discount to Amazon, Meta, Google, Microsoft, and Apple implicitly prices in an unwarranted EPS haircut.
HBM content per rack rises only ~$0.2–0.3M from Blackwell to Rubin versus $2–3M of rack pricing uplift, which will keep gross margins in the mid-70s.
Vendor financing (~$65B) consumes less than 35% of free cash flow, leaving ample buyback capacity.
The stock's valuation is implicitly pricing in an unwarranted EPS haircut, as its P/E discount to peers suggests a significant earnings reduction that the firm does not believe is justified.
Gross margins are expected to remain strong in the mid-70s, as the increase in HBM content per rack is significantly less than the rack pricing uplift, mitigating margin pressure concerns.
NVDA +3.6% finally getting some love after Vivek was out defending this morning pushing back on the four dominant bear debates — HBM-driven margin pressure, ASIC competition, crowded ownership, and vendor financing — arguing the stock’s ~30–35% P/E discount...
The company's CEO expressed a very bullish outlook during meetings hosted by Cowen.
The CEO's bullish stance was also perceived as defensive, which is consistent with recent observations.
The company stated its AI chip roadmap remains intact despite news of Kyber NVL144 delays.
NVDA +37bps underperformed all day while AMD +6.6% seemingly the big winner from the news as MI500 now lands against a dual -die Rubin Ultra in a legacy-style rack instead of the quad-die Kyber.
Semianalysis reports that the Kyber NVL144 has been delayed by more than 12 months.
Jefferies cut its AI PCB/CCL TAM due to potential delays to Nvidia Kyber/backplane PCB adoption for Rubin Ultra, which could create approximately 5-8% downside to its 2027 AI PCB/CCL TAM and another 11-16% downside in 2028 if Kyber is ultimately cancelled.
Jefferies now expects Kyber may slip out of Rubin Ultra in 2027 and possibly into a new Rubin Ultra variant in 2028, with the Oberon structure likely used as a fallback.
This delay is seen as negative for PCB supply-chain sentiment.
AVGO leading the way higher after renewing with AAPL through 2031, the usual suspects on a semi up day (HDDs/Memory/semicap) outperforming early, and AMD catching a bid on the NVDA Kyber delay news getting passed around.
The stock underperformed throughout the day, gaining only 37 basis points.
The delay of Kyber and optical NVLink, which were expected to widen NVDA's moat in 2027, could make competitors' offerings more comparable.
Semianalysis estimates data center compute revenue will be 20% higher than street expectations.
The company is identified as a key compute beneficiary in humanoid robotics due to its AI processors and full-stack robotics capabilities.
The company is building a massive global AI network, creating a potential step-function demand opportunity for networking infrastructure.
GF Securities argues SpaceX is transforming into a neocloud provider, with compute capacity potentially expanding from ~1GW today to 4.5-9.5GW by 2027/28 and requiring ~2.2-2.5M Blackwell-equivalent GPUs. The firm sees NVDA as the primary GPU beneficiary.
The stock continues to perform poorly.
NVIDIA's Vera Rubin platform will adopt 800V high-voltage direct current (HVDC) architecture, with small-volume shipments expected to begin in Q3.
The stock saw a bid despite underperforming, suggesting underlying support.
Asian press reported that the company will adopt 800V technology, with volume shipments beginning in Q3, indicating a significant technological advancement and market adoption.
AVGO/NVDA +3%: even the funding shorts got a bid even though they underperformed.
Nvidia is quickly pivoting to its new "Vera" central processors for AI data centers, with availability expected as soon as August, and has begun sales pitches to Chinese clients, allowing them to place orders.
The company's fortunes in China are rapidly declining, as shipments of its second-most powerful AI chip, the H200, to the country have stalled for months.
BofA continues to view NVDA as the top overall AI beneficiary.
Give lack of positive news flow/raise, the competitive/margin fears continue to pop up on AVGO despite the LT story being very much intact. Continues to be in funding short bucket with NVDA for now.
The company continues to be a popular funding short.
Nvidia could provide financial backing for OpenAI's proposed 10-gigawatt data center campus in Ohio.
China's plan to spend $295 billion on AI infrastructure over five years aims to rely on local suppliers like Huawei for at least 80% of technology, effectively squeezing out Nvidia.
Taiwan is considering stricter export controls on AI chip sales to China to align with US measures, addressing semiconductor smuggling that risks drawing a rebuke from Beijing. This effort aims to prevent the diversion of advanced hardware, like AI servers with Nvidia Corp. chips, from Taiwan to China, as such sales are already banned under US regulations unless companies get Washington’s permission.
Power names were also weak as Semianalysis said NVDA’s single-ended 800VDC architecture is getting pushed to 2028+ as hyperscalers push back (it's not needed for Rubin) — main read-through here is its largely neg for wide bandgap pure plays like WOLF (-12%)...
Morgan Stanley came away from Computex more constructive on cloud semis and memory, arguing recent server memory optimization moves reflect supply constraints rather than weakening demand. The firm continues to see strong demand for cloud CPUs, GPUs, and related connectivity components into 2027, highlighting Nvidia’s Vera CPU strategy as a positive for memory interface vendors and CPU-related content as a Vera CPU can accommodate 1.5TB of LPDDR5X, triple of Grace CPU, which means Vera alone will need more than 10% of last year’s total DRAM bit demand.
UBS came away from Computex with a constructive view on AI infrastructure demand, citing continued strength across CSPs, neoclouds, and enterprises, with tight supply conditions persisting across memory, advanced packaging, foundry, and networking. The firm highlighted Rubin Ultra as a major step-up in rack-scale architectures, driving higher content across CPUs, memory, storage, networking, and power.
Nvidia is running early trials on 18A, indicating potential for future collaboration or adoption of Intel's advanced packaging technology. Additionally, there's an implied indication from meetings that NVDA will be gaining share with Anthropic 'soon,' which could precede a big announcement.
Memory bouncing back nicely with MU +9% and SNDK +5% as street was out defending what was obvious late last week given the SoCAMM reductions at NVDA. Citi anod others said they expect no meaningful reduction in overall SoCAMM demand as memory suppliers rema...
Bulls here still at $140-$150B here for CY27, and in low 30s EPS for 2028; if you believe that, stock trading well under 15x but debate I’ve heard from investors over the past week is whether you want to buy AVGO or NVDA at that multiple (the right answer i...
The company has stated that SOCAMM represents a $300B total addressable market.
The company is reducing Rubin NVL72 SOCAMM DRAM capacity per rack by approximately 50%, which lowers the estimated rack cost and total cost of ownership.
Deutsche Bank came away from Computex with increased confidence in NVDA’s technology leadership, highlighting Vera Rubin’s on-time ramp, growing CPU ambitions, and expansion into AI PCs through RTX Spark.
While ARM should benefit from higher CPU adoption through royalties, the firm cautions that Nvidia’s growing CPU presence could create some double-counting risk around ARM’s standalone AI CPU opportunity. The analyst views the updates as supportive of NVDA’...
MRVL +32%(up another 6% in the postmarket) as Jensen said they were the next trillion dollar company. Pays to have Jensen as a shareholder. Don’t think this was an offhand comment — Jensen knows what he’s doing: keep supply chain happy, confirm that if you...
The company is entering the market with a new RTX chip, which is a positive development.
The company's Vera Rubin is in full production, indicating strong manufacturing capabilities.
The company's Vera CPU offers 1.8x agentic sandbox performance compared to X86, highlighting a significant performance advantage.
The company unveiled an RTX Spark 3nm chip with Mediatek, offering up to 1 petaFLOP of local AI performance and 128GB of unified memory, showcasing innovation in AI hardware.
The company announced a new open-source model, Nemotron 3 Ultra, which is 5x faster and 30% cheaper than other leading open-source models, demonstrating a competitive edge in AI software.
The company highlighted that tokens have become profitable, driving significant demand for compute.
GitHub commits have increased nearly 3x in 2026 due to AI Coding tools improving the productivity of software engineers, which the company argues will result in hiring more software engineers as output per engineer increases.
The company's revenue density is approaching $40B per gigawatt, suggesting that Street datacenter assumptions remain too conservative.
The CEO's visit to Korea highlights the company's deepening reliance on Korea for both physical AI development and memory supply, indicating strategic partnerships and supply chain strengthening.
The company is expected to source more than 80% of its HBM4 demand from Samsung and SK Hynix, reinforcing Korea’s critical role in next-generation AI infrastructure.
Robotics, humanoids, autonomous systems, and industrial AI are seen as key areas of collaboration for the company.
The stock experienced one of its best days in months, indicating strong positive momentum.
Still, broadly speaking, AI semis price action continues to be full HOT AI SUMMER, with right tail outcomes more prevalent than left tail. Just check out ARM +15% despite many expecting the NVDA RTX Spark announcement at Computex. Even NVDA +6% beginning to...
The company is expected to have the largest notional sell in today's MSCI rebalance, with an estimated $11 billion to trade.
ARM is also seen as a major winner through both higher CPU share and royalty leverage, while NVDA is expected to capture the largest absolute AI CPU share but with a more modest EPS impact given the scale of its GPU business.
The chart indicates a ripe opportunity for another short-to-medium term trade, as the stock has historically performed well after breaking below previous resistance/support in a "false breakdown/breakout" pattern, which occurred again today.
The upcoming Computex event next week serves as a potential catalyst.
There is little reason to invest new capital into the stock as other semiconductor and data center plays offer more potential for movement, and investors are performance chasing.
The stock continues to frustrate long positions, but it serves as an easy funding short when other AI semiconductor stocks are performing strongly.
The transition to a 'token economy' where token generation is a key profit center favors Nvidia's superior economics, and concerns over peak demand are misplaced.
The firm highlights NVDA’s NVL SuperPOD scaling to ~17 exabytes of memory (~240x density vs. NVL72), underscoring memory as a key bottleneck and value driver. Aletheia adds the evolving balance between compute and memory will define AI factory economics, re...
AMD +4% as Lisa Su was in Taiwan sounding good around CPU demand. They also announced >$10B in Taiwan ecosystem investments ahead of Computex, as unlike NVDA< AMD is continuing to build a 2nd/3rd packaging supply path.
AMD +4% as Lisa Su was in Taiwan sounding good around CPU demand. They also announced >$10B in Taiwan ecosystem investments ahead of Computex, as unlike NVDA< AMD is continuing to build a 2nd/3rd packaging supply path.
The company was noted as not hitting new highs along with the broader index, despite being an AI semi beta name that led earlier rallies.
The company's Q1 revenue and Q2 guidance were largely in line with expectations, indicating stable performance.
The company disclosed significant visibility for its Vera CPU, projecting a substantial market opportunity and positioning itself as a major CPU maker.
Data Center revenue showed strong year-over-year and quarter-over-quarter growth, driven by compute and networking segments.
Management indicated that the Vera Rubin product will begin production shipments soon, with major customers already secured.
The company is gaining market share in inference, supported by new customer acquisitions and improved product performance.
The firm also highlighted rising confidence that upcoming Zoox/Lucid/Nuro launches — followed by NVDA in 2027 — could create more proof points that AV supply will not consolidate around just TSLA/Waymo, supporting a gradual removal of the AV
We’ll hit on NVDA and INTU -14% first then get to the usual, where we have a lot of good stuff to cover today.
INTC +5% leading the way higher following Lip Bu Tan’s bullish JPM Conference appearance yesterday (we sent out notes here in case you missed it). The rest of AI semis strong as well with MRVL +4%, MU +4%, ALAB +3% among others as we head into NVDA earnings...
The company's July guidance for revenue was in line with buyside expectations, indicating solid performance.
The company announced a substantial share buyback program, signaling confidence in its financial position.
The company's performance has been consistently strong and predictable over recent quarters.
The company is expected to ramp CPO-enabled switches, with significant volume increases projected for 2027.
Blackwell capacity is stable, and the second quarter of 2026 is anticipated to be the largest quarter for the product.
The Rubin product is expected to ramp up significantly in the fourth quarter of 2026, with substantial planned capacity.
* **$AMD**: Inventec is expected to ramp the MI450 for MSFT with mass production starting in Q4, with more meaningful revenue in C27. * TPU volume checks are mixed, with some indicating 3M units for C26, while others suggest 3.6-3.8M units for C26 (+150% Y/...
Concerns exist regarding the delivery of HBM4 memory and a thermal issue with PCBAs.
AI networking checks remain strong, with scale-out demand still supply constrained and hyperscaler architectures increasingly converging toward optical interconnects/CPO over time.
The company is viewed as the dominant scale-up ecosystem via NVLink/NVLink Fusion, with Spectrum CPO deployments potentially reaching 150K–300K units by 2027 and broader CPO adoption accelerating into 2027–28.
The company is still viewed as the ecosystem leader with durable training share.
The company's software moat remains important even as hyperscalers/frontier labs become increasingly capable of optimizing around competing architectures.
Server builds are biased higher for CY26 on supply shifts and the outlook is strengthening through CY27+.
Rubin is still expected to materially contribute to Q2 despite delays.
The company is beginning to ship Vera GPUs, with 3-4M standalone Vera units potentially shipped in CY27.
The company is expected to have another beat-and-raise quarter.
Emerging themes like neocloud demand, non-hyperscaler AI infrastructure adoption, AI server CPUs, optics/networking opportunities, and broader TAM expansion could become more important drivers of future multiple expansion.
The firm expects sustained Blackwell/Rubin momentum alongside additional FY28 earnings upside.
AI Networking/ALAB/MRVL/NVDA: ISI Checks Remain Positive on NVDA, MRVL, and ALAB
Edgewater/ISI had positive checks, leading them to raise their numbers and expect the Rubin platform to materially contribute to Q2 despite delays.
The shift to agentic AI represents a massive, multi-year structural demand wave that requires billions of digital workers, driving exponential computing and memory needs.
The AI buildout is in its infancy and represents a decade-long runway, transitioning from digital agents to physical AI.
Nvidia has built a comprehensive, integrated agentic stack (including Grace Blackwell, Vera CPU, and Nemo) that turns LLMs into functional enterprise agents.
The company's moat faces more pressure at the high-end as inference share could fall toward ~50% by 2028 due to improvements from competitors like AMD, TPU, Trainium, Maia, and SRAM-centric architectures.
AMD -1.6% as heard some very fast money types not wanting to be long into the NVDA earnings.
Severe supply chain constraints in memory and advanced semiconductors persist as demand continues to outpace global capacity.
The supply chain will struggle to keep pace with the rapid buildout for at least the next decade.
Jensen Huang, the CEO, sounded bullish on Agentic AI at a Dell event.
NVDA -1.3% as bogeys for a $3B beat vs street at $78B and guide of ~$90B-$91B vs street at $86B. Jensen was at DELL today sounding bulled up on Agentic AI (we’ll send out some quotes shortly)
Unprecedented demand visibility stretching into 2027 with compute capacity effectively sold out.
Strong product pipeline momentum with Blackwell and Rubin platforms driving continued hyperscaler capex.
The stock is experiencing strong relative performance and catching up after lagging.
Regulatory clearance for chip sales to China provides a positive catalyst.
Concerns persist regarding the sustainability of high margins and elevated valuation multiples.
The stock is finally receiving attention as an AI laggard, showing strong performance and reaching four-month relative highs against the SOX index after touching two-year relative lows.
The stock has seen significant gains recently, outperforming the SOX index, driven by news of US clearance for H200 chip sales to Chinese firms and the availability of H100s for sale.
The firm highlights raised long-term server CPU TAM forecasts (~$120B by 2030), strong OpenAI/META/NVDA-related demand, and better-than-expected FY26/FY27 outlook supported by supply finally catching up to demand.
The company reached new all-time highs, indicating strong market performance.
MS said AI data center racks are outgrowing today’s 480V AC/54V DC architecture, with next-gen systems like NVDA Rubin Ultra/Kyber requiring higher-voltage designs (including 800V power sidecars) that materially increase power semiconductor content.
800V DC rack architecture is their highest conviction theme to play the AI infrastructure moving beyond first order winners like NVDA, memory, optics, etc. They called out various ways to play it, with WOLF (+6% today) their main pick (their 200mm SiC fav o...
The CEO's absence from the official delegation list for a China trip signals to the Chinese government that obtaining top-performing chips like those made by Nvidia will be difficult for Chinese AI labs.
The company's strongest AI platform, Vera Rubin, is rumored to enter trial production in June and ship to North American cloud service providers from July, benefiting ODM partners.
The firm remains constructive on the company.
The firm remains constructive on NVDA, AVGO, and semicap, while recently warming on AMD as AI and CPU demand broaden beyond core accelerators.
The shift towards agentic inference, which prioritizes memory capacity and cost over raw compute speed, could diminish the company's premium valuation advantage.
The stock was a lone standout.
The stock gained, playing catch-up after underperforming previously.
The analyst re-bought the stock for a medium-term trade, indicating continued confidence.
The company's shift towards hard supply agreements indicates increasing demand and potential shortages, which can be seen as a positive for its market position.
The stock is one of the few in the AI sector not performing well, which ironically benefits the rest of the space from a flow perspective.
The stock was down 3% following news about OpenAI missing internal targets and spending scrutiny, indicating market concern about its exposure to the AI complex.
BofA believes a shift toward higher dividends and buybacks could drive a re-rating for Nvidia by expanding ownership and signaling the durability of its cash flows.
Nvidia trades at a significant discount compared to Mag-7 peers despite generating more free cash flow than Apple and Microsoft combined.
Nvidia's approximately 70% AI share, ecosystem strength, and backlog visibility support sustained leadership.
BofA maintains a Buy rating, citing capital return acceleration as a key catalyst to close the valuation gap.
Advanced node capacity is moving into a structural shortage by 2027, with demand from AI customers like Nvidia far exceeding supply, leading to pricing power.
The firm says advanced node capacity is moving into structural shortage by 2027 (15– 20% gap), with demand far exceeding supply as AI customers (NVDA, AVGO, AMD) compete for allocation. The analyst notes pricing power is set to inflect higher, with N2/A16 w...
The stock experienced a significant rally, finally breaking through to 52-week highs after a prolonged period of range-bound trading, indicating a positive shift in investor sentiment.
The stock's recent strength was supported by positive sentiment surrounding OpenAI, specifically following developments with ChatGPT 5.5.
The company is playing catch-up after other semiconductor stocks have rallied significantly, suggesting its recent breakout is a normalization of its performance relative to the sector.
News from Asia indicating an increase in TSMC's 3nm capacity to 180k wafers per month from 160k bodes well for the company's ability to meet demand for its products.
A shift towards higher dividends and share buybacks could drive a re-rating of the stock higher.
The stock has performed well since the analyst's previous call for it to break out to new highs.
Lastly, the POET blow up probably helped on the margins as there was a bid to larger cap names in semis. News out of Asia that TSM’s 3nm capacity now seen at 180k wpm vs 160k previously, which bodes well for NVDA to meet demand. BofA was also out saying a s...
Increasing competitive pressure from Nvidia is likely to constrain AMD’s share gains and margin expansion.
Elevated R&D spend and supply advantages favoring Nvidia limit near-term upside for AMD despite solid execution.
DeepSeek V4's co-optimization for non-Nvidia hardware and its lower pricing signal a shift towards cost-efficient infrastructure and a weakening of CUDA lock-in over time.
The development of Deepseek's V4, which co-optimizes with non-NVDA AI hardware, suggests a potential weakening of CUDA lock-in over time as models are redesigned for Chinese chips.
The firm highlights a shift toward consumption-and outcome-based pricing (credits, usage models) expanding monetization, alongside incremental ARR drivers from GenStudio, AEP Apps, and AEM (+20%+ growth), while noting NVDA’s endorsement reinforces Adobe’s r...
Net, Barclays sees multiple headwinds—including competitive risk (NVDA/AMD), high power consumption, and potential Samsung insourcing—keeping risk/reward skewed to the downside.
The overall narrative is shifting from concerns about margin issues and Rubin delays to the understanding that compute resources are in short supply, which could be further activated by OpenAI's Spud.
The stock was negatively impacted by the Anthropic/AMZN news, and its short-term performance needs to hold the 10-day moving average and stay above the previous range of $196/$197 to avoid a trim of the short-term/medium-term trade.
The narrative for NVDA is more ambiguous compared to AMD, which benefits from clearer dual tailwinds.
NVDA -1% dinged presumably by the Anthropic/AMZN news as we head into Cloud Next tomorrow where we are likely to hear about next-gen TPUs. On a short-term basis, want to see this hold the 10d and stay above the previous range $196/197; if not would be a sig...
The company is identified as a leading beneficiary of the broader AI data center interconnect TAM, which could reach ~$22B by 2029.
Wolfe initiates CRWV at Outperform with a $150 PT, highlighting its position as a leading neocloud with strong access to GPUs via deep relationships with NVDA and hyperscalers, and a differentiated financing model (customer prepayments + asset-backed “box”...
Wolfe initiates NBIS at Peer Perform, citing strong hyperscaler demand (MSFT/META commitments, NVDA backing) and improving GPU pricing as validating the demand story, but flags key open questions around execution of large-scale capacity builds and the durab...
Clev is positive on the company, citing higher CY26 GPU shipments compared to previous expectations, VR on track, and strong demand.
Supply chain reports indicate that Vera Rubin shipments are ramping on track for Q2 and the second half of the year, and any potential HBM4 delays would simply be offset by increased Blackwell shipments.
A bearish analyst report suggested that Hynix cutting HBM4 shipments could create issues for Nvidia's Rubin platform.
Favorable bandwidth deployment trends support upside for data center revenues.
NVDA: Hearing 3p positive saying favorable BW deployment trends support upside for DC revs DELL/HPQ -2% early as NVDA denies any PC related acquisition Data Centers: TD Cowen Channel Checks Flag Record ~9.4GW of 1Q26 U.S. Hyperscaler/AI Lab DC Leasing, but...
CEO Jensen Huang urged business leaders to prioritize long-term AI breakthroughs over near-term profitability, comparing the AI revolution to the Industrial Revolution and advocating for patient, experimental investment without immediate ROI expectations, a position made easier by the company's dominance in AI chips and massive profitability.
The firm maintains its 2027 outlook, seeing upside as supply bottlenecks ease, despite near-term trims due to Blackwell timing and HBM4 constraints.
TSM/NVDA/AVGO/AMD: JPM Raises CoWoS/SoIC Outlook; Near-Term AMD&NVDA Trim, ’27 Upside Intact TSM/NVDA/AVGO/AMD JPM CoWoS/SoIC
Near-term CoWoS capacity for 2026 was trimmed due to Blackwell timing and HBM4 constraints.
UBS supply chain analysis suggests Rubin Ultra will retain a 2-GPU die CoWoS package with 8 HBM stacks, which has neutral-to-positive supply chain implications as N3/HBM silicon demand should be similar or higher than a 4-die design.
The 4-die migration is now targeted for Feynman (H2’28), contingent on TSMC’s CoPoS reaching mass production by 2028 to compete with Intel’s EMIB-T.
The stock underperformed compared to other semiconductor companies, despite being up.
UBS indicated that the upcoming Rubin Ultra (H2'27) will likely retain a 2-GPU die CoWoS package with 8 HBM stacks, possibly due to CoWoS inefficiencies at larger sizes.
The company is strategically partnering with Marvell Technology, Inc. to expand its AI ecosystem through NVIDIA NVLink Fusion, offering customers greater choice and flexibility.
The company will collaborate with Marvell on silicon photonics technology.
Reports of reduced Rubin wafer orders and lower CoWoS capacity for Rubin are not new information and are not expected to materially impact the company's short-term outlook.
The Rubin Ultra's 4-die variant has not been cancelled, despite production inefficiencies, indicating the company is aggressively pushing its supply chain.
Early price action shows continuing of yesterday: Memory names weaker while large cap tech is outperforming (AVGO /GOOGL/NVDA/MTEA/AMZN/AMD +1.5% while memory/HDDs flat to down), the perception being what is bad for memory is good for hyperscaler ROI. There...
AVGO/GOOGL/NVDA/MTEA/AMZN/AMD +1.5% /HDDs
The company's stock was bid higher as rumors of Rubin reductions were dispelled by Jefferies' Sherman.
MRVL +11% as they announced a partnership with NVDA to connect MRVL to the NVDA AI Factor and AI-RAN ecosystem through NVIDIA NVLink Fusion as well as a $2Bn investment from NVDA.
TD Cowen noted that NVIDIA is shifting to directly lease data center capacity for GPU-as-a-Service providers, moving from credit backstops to a more active role in securing supply.
Checks indicate multiple requirements for over 200MW in the market, which is significantly higher than historical levels and suggests accelerating demand.
NVIDIA could assign or sublease this capacity to partners, leveraging its balance sheet to facilitate deployments.
The stock held in relatively well despite the broader market weakness, especially given its underperformance compared to leading names since October.
Reflection, an Nvidia-backed startup developing open-source AI models, is raising $2.5 billion at a $25 billion valuation, positioning itself as a Western alternative to Chinese AI offerings.
Reflection plans to build customized models for allied nations' "sovereign AI" infrastructure and has secured multi-billion dollar commitments, including a Korean language project powered by Nvidia chips.
Reflection is part of Nvidia's broader effort to create an open-source AI ecosystem that runs on its hardware and counters Chinese AI competition.
GPU spot rental rates are increasing significantly, with B200 on-demand up 50% in a month and H100s up 20% since November, indicating strong demand.
Cantor reiterated Overweight and a $300 price target, stating that AI is entering a “token economy” phase where token generation becomes a key profit center, with NVDA best positioned given superior “token per second per watt” economics.
Cantor highlighted agentic AI (OpenClaw/Nemo), extreme co-design, and ecosystem scale as structural advantages driving continued share gains and AI factory buildout.
Cantor added that investor concerns around peak demand are misplaced, with upside catalysts from hyperscaler growth, pricing, and capital returns.
Clev is positive, saying Rubin is expected to account for > 1/3 of CY26 GPU shipments, higher than prior expectations.
The firm highlights NVDA’s NVL SuperPOD scaling to ~17 exabytes of memory (~240x density vs. NVL72), underscoring memory as a key bottleneck and value driver. Aletheia adds the evolving balance between compute and memory will define AI factory economics, re...
The company is expected to sell 1 million chips to Amazon by the end of 2027 in a cloud deal.
GTC takeaways highlight effectively insatiable demand for inference compute, with utilization near 100% and agentic AI driving step-function increases in token demand.
The company is pushing toward system-level architectures beyond GPUs as complexity rises, with customers increasingly prioritizing end-to-end performance and orchestration.
The stock has a good risk/reward.
The stock continues to be range-bound.
The company received approval from Beijing to sell its H200 AI chips in China, expanding its market access.
The company is preparing a version of the Groq AI chip for the Chinese market, indicating strategic adaptation to regional demand.
The forthcoming Vera Rubin chips cannot be sold in China, limiting the company's access to that market for its most advanced products.
Nvidia will unveil a new server rack at GTC that integrates Groq's inference-focused "language processing unit," marking the first time it has built another company's AI chip into its systems.
Nvidia is exploring the eventual fusion of the LPU with its next-gen Feynman GPU into a single chip.
The new server rack pairs Groq chips with Intel processors for inter-chip communication because Nvidia's own technology does not yet integrate cleanly with the LPU.
Fubon's checks suggest GPU server rack shipments are expected to reach approximately 75K units, up from a prior estimate of 70K, driven by improved ODM yields and strong CSP demand.
The company released Nemotron 3 Super, an architecture praised for its efficiency, achieving strong reasoning performance with minimal active parameters and high throughput.
Matt Dratch @DratchCap · 7h Jensen laying out more capital ($2B behind $NBIS) to elbow out competition from securing gigawatts. First $CRWV, now $NBIS. Both are effectively now captive $NVDA clouds. Also, 5 GW target by 2030 is bananas. I recently ran the '...
The company experienced continued downside movement.
The company needs a significant increase to its Blackwell/Rubin revenue outlook to overcome its recent trading performance, but supply chain issues with Rubin interposer and packaging yields may limit upside.
Concerns exist regarding incremental AI demand, funding strain among neoclouds, Microsoft's internal silicon efforts, China restrictions, and softer OpenAI ambitions.
Any post-earnings bounce is expected to fade.
A Chinese AI startup reportedly trained its latest AI model on Nvidia's most advanced AI chip, the Blackwell, despite US export bans, which could represent a violation of controls.
Channel checks reinforce robust GPU demand, with Blackwell lead times running 12–26 weeks and Vera Rubin tracking ahead of schedule, potentially shipping by the end of Q2 2026.
OpenAI is expected to maintain the majority of its infrastructure on NVIDIA through 2027.
The near-term setup suggests a strong beat and raise, with CY26/27 EPS tracking toward ~$9/$12, driven by sustained compute demand and no easing supply constraints.
The company is at the cusp of a rack-scale supercycle as Blackwell ramps and Rubin follows, with AI demand accelerating into CY27/28 and backlog extending through CY26.
Concerns about hyperscaler capex peaking and TPU competition are misplaced, as the 'zero-sum' narrative ignores exploding AI workloads and NVIDIA's ecosystem lead.
The new GB300 NVL72 system runs AI models up to 100x faster than the previous H100 chips, significantly exceeding initial promises.
Despite its high price, the GB300 NVL72 system delivers up to 65x more tokens per dollar spent compared to the older Hopper generation, offering better value.
The company released PersonaPlex-7B, a 100% open-source, free, full-duplex voice model that enables real-time conversation without pauses or turn-taking.
AMD/NVDA: Hearing Clev positive citing strong Server CPU (AMD) & GPU Demand
The company may relax HBM4 specifications due to reported capacity and yield limitations faced by Samsung and SK Hynix.
If NVIDIA maintains its 11.7Gbps specification, the availability of sufficient HBM4 volumes for Rubin's mass production could be constrained.
The company is considered one of the cleanest names among those relevant to improving OpenAI sentiment.
The company is seen as a relevant name benefiting from improving OpenAI sentiment.
The stock experienced weakness along with other big year-to-date winners.
The company is a beneficiary of the rotation from internet mega-cap to semi mega-cap stocks.
AVGO +3.5% / NVDA +2.5% as Megacap had a better bid today…2 names the beneficiaries if rotation from internet mega cap to semi meg cap.
MU -3% after Yonhap reported Samsung plans to restart mass production of HBM4 as early as this month for NVDA’s Vera Rubin. We pitched Samsung instead of MU back in Sept as we were worried about MU’s lack of HBM progress and those fears finally seem to be w...
NVIDIA is exploring collaboration with Intel for Feynman GPUs, with the I/O die potentially being partially produced using Intel's 18A or 14A process, and advanced packaging handled on Intel's EMIB platform, which appears to be an incremental positive development.
The potential collaboration with Intel for Feynman GPUs, including the use of Intel's 18A or 14A nodes for I/O dies starting in 2028 and EMIB packaging with a ~25/75 split with TSMC, represents new and positive details compared to prior reports.
China has approved three of its largest tech companies (ByteDance, Alibaba, and Tencent) to purchase over 400,000 H200 AI chips, indicating a shift in Beijing's stance and strong demand for NVIDIA's advanced AI hardware.
The company is expanding its collaboration with CoreWeave to accelerate the buildout of AI factories, representing a significant new business avenue.
Chinese tech firms are preparing orders for Nvidia's H200 AI chips, indicating potential formal approval for imports and strong demand from a key market.
NVDA +1%/AMD +3% up early on positive China H200 news.
There is positive sentiment from analysts regarding the company.
AMD/NVDA: Hearing both Clev and Edgew out positive on both (CPU server strength for AMD)
The stock finally received a bid following positive checks on rack shipments from analysts.
We are up over 100% in 13 days and nearly 1,000% in 5 months, right up into the big $500 century mark. Parabolic moves in the super-liquid momentum stocks often peak right at, or overshoot, these obvious round numbers: $NVDA 1000, $SMCI 1000, $MSTR 500, $TS...
The company retains an advantage on time-to-market and CUDA as a durable enterprise moat, despite Google/Broadcom's TPU closing the compute cost gap.
The company's valuation is considered reasonable, and it is expected to show renewed leadership if cloud capex visibility improves and GTC re-energizes sentiment.
Tight memory chip supplies will constrain the number of US export licenses for the company to sell its H200 AI processors to Chinese customers.
Chinese AI developers believe they cannot compete with the U.S. without better chips, indicating a potential challenge for the company's market in China.
Wolfe Research reiterates Outperform with a $250 price target, viewing the newly announced 25% tariff on advanced semiconductors as a formalization of an existing payment mechanism, not an incremental risk.
The company is currently not shipping AI chips to China, and the situation is not reflected in consensus, with potential upside of $2–5 billion per quarter if shipments resume and approvals are granted.
The tariff provides legal structure for prior arrangements and does not change the medium-term earnings outlook, with resolution likely tied to broader US-China trade negotiations.
Chinese customs authorities have informed agents that Nvidia's H200 artificial intelligence chips are not permitted to enter China, and Chinese government officials have instructed domestic technology companies not to purchase these chips unless necessary.
AI Semis / AMD / INTC: KeyBanc raises NVDA/AVGO CoWoS estimates and sees CPU Pricing Firming / Upgrades AMD and INTC on CPU refresh/pricing
The firm notes NVDA’s 2026 CoWoS supply has been lifted to ~650K interposers (+23% vs. prior_.. AVGO’s 2026 CoWoS allocation was raised to ~250K (+30% vs. prior) with OpenAI’s custom ASIC delayed to 1Q27 but lifetime units increased to ~1.5– 2M, adding $8B–...
AI server shipments from Taiwan's Big 6 makers surged in December, with combined monthly revenue rising 35.9% year-on-year to a record NT$1.59 trillion, driven by booming demand for AI servers.
AI server shipments from Taiwan's Big 6 makers continued to surge in December, bucking seasonality as combined monthly revenue rose 35.9% year-on-year to NT$1.59 trillion, a fresh record high, amid booming demand. 1/10 $NVDA $DELL $HPE #Alservers #Foxconn #...
The company is expected to benefit from persistent AI strength in GPUs and ASICs in 2026E.
AI strength is expected to persist across GPUs and ASICs (positive NVDA, DELL, AVGO), AI interconnect and optical with scale-up/out/across and 800G/1.6T (positive LITE, CRDO), and continued tailwinds for WFE (positive LRCX) alongside a memory supercycle wit...
China plans to approve some imports of the company's H200 chips as soon as this quarter, providing renewed access to a critical market.
The company is seeing robust demand and incremental upside from physical and agentic AI, with a strong Rubin ramp expected in 2H26.
Bloomberg reported that China is expected to approve H200 buying this quarter.
The company is requiring full upfront payment for H200 chips from Chinese customers, with no options to cancel, ask for refunds, or change configurations, due to uncertainty over Beijing's approval of shipments.
Morgan Stanley raised its CoWoS bookings for NVIDIA, with a significant portion tied to Rubin, indicating strong demand for their advanced packaging solutions.
CES confirmation that Rubin is in full production underscores a major manufacturability step-up, supporting a 2H26 ramp.
NVIDIA's CES commentary on Rubin operating with 45°C inlet water reduces the need for traditional chillers but still drives demand for dry coolers and high-temperature liquid-to-liquid systems.
Rubin’s fully liquid-cooled design significantly expands demand for liquid-centric infrastructure, reinforcing the shift towards tightly integrated power and cooling architectures.
AI Semis/NVDA/AVGO/AMD/CoWoS: Morgan Stanley Raises CoWoS Bookings for NVDA/AVGO, Trims AMD/MSFT Maia; Supply Constraints Still Drive ’26
Morgan Stanley’s Charlie Chan update sCoWoS assumptions, raising NVDA bookings to ~875k (from ~800k) with incremental Non-TSMC allocation and ~40% tied to Rubin, while lifting AVGO to ~280k (from ~235k) and trimming AMD to ~110k (from ~130k), including ~65k...
Management indicated that Vera Rubin is in full production and remains on track for a C2H26 ramp, with the ecosystem being more advanced compared to Blackwell.
The company is strongly emphasizing "physical AI" as the next growth driver, exemplified by Alpamayo for AV reasoning stacks, with the AV opportunity projected to exceed $10 billion by 2030.
Co-design across compute, networking, and storage is seen as a durable competitive advantage, even as AI ASICs continue to gain market share based on total cost of ownership.
The H200 chip offers significantly faster performance and more memory at over double the average selling price compared to the H20.
Beijing regulators are considering limiting access to the H200 chips, requiring buyers to go through a strict approval process.
Morgan Stanley’s Charlie Chan notes his latest checks point to 3nm capacity running tighter than expected, with TSMC’s 2025 outlook still at 110–120kwpm and 2026 rising to 140–150kwpm, supported by AZ phase-2 output and a 4/5nm conversion, yet still insuffi...
CEO Jensen Huang is actively working to protect market share by enticing major customers to make big commitments to Nvidia GPUs.
Google's aggressive TPU push and potential multi-billion dollar deal with Meta threaten Nvidia's market dominance.
Futures flat but lots of movement underneath the surface with NVDA -4% and AMD -6%. The sell OpenAI levered names vs Buy GOOGL/TPU exposed names was in full force yesterday and likely continues today after TheInformation calling out META possibly buying TPU...
KeyBanc expects a strong October quarter and higher January guidance driven by rising B300 volumes and smoother GB300 rack production.
Morgan Stanley notes accelerating demand signals, strong Vera Rubin momentum, and resolved rack issues.
Capacity constraints could limit the upside and impact gaming GPU availability.
Major customers Amazon and Microsoft are backing legislation that could restrict Nvidia's exports to China.
Citi says DRAM is poised to become the next key component to secure long-term contracts across the AI supply chain, mirroring the trajectory seen with NVDA, AMD, and AVGO. The firm believes Micron will benefit from a tightening supply backdrop and sustainab...
Buyside was about $20 in FY30 while sell side at 18.50 and most view this as conservative. The one nit being that FY28 guide of 10.65 is actually lower than the street at 11.21. Infra revs were raised from 18/32/73/114/144 to 18/34/77/129/166 from FY 26 - F...
Demand from hyperscalers, tier-2 clouds, and sovereign buyers remains insatiable, driving significant revenue and EPS estimate upgrades.
Supply chain checks indicate limited risk of unsold Blackwell inventory, easing previous market concerns about order slowdowns.
NVDA / AMD: Mizuho More Positive on Asia ODM Checks; NVL72 Forecasts Raised
NVDA: MS Raises Estimates, PT $206 on Hyperscaler and Tier-2 Cloud Strength in Quarter preview
MU/Memory: Goldman Sachs cut its HBM TAM forecast for 2026 by 13%, from $51B to $45B, as Nvidia’s NVDA next-gen Rubin GPU is expected to keep the same HBM content as the Blackwell Ultra at 288GB, instead of a significant increase
AVGO Eik 1.4% B NVDA Eik 2%; IB, 'kit In other factors, the bid to shorted names was lacking today as Megacap and HF favorites led the way higher relieving some L/S HF pain from the last couple of weeks: It was the quiet before the start of the big earnings...
NVDA +1.9% JT MS Asia analyst outjust now saying Wistron seeing ~7% raise to their GBZOO rack orders TSM -1% likely on the SamsungITSLA foundry deal although we think that says more about TSM being supply constrained than any mfg advantage Sammy might have...
Semis/IT Spending: JPM CIO Survey Shows Strong AI and Cloud Spending Intentions; NVDA, AMD, AVGO, MRVL Among Top Beneficiaries
Research through 2026-11-03 · 2025-06-27 to 2026-11-03 · 138 source reports.
| Range | Prices through |
|---|---|
| 1W | |
| 1M | |
| 3M | |
| 6M | |
| 1Y | |
| 2Y | |
| 3Y |
Research through 2026-11-03. Artifact validity does not establish factual accuracy or research freshness.