The company delivered better execution than expected, addressing investor concerns around execution, new-deal unit economics, and GPU obsolescence, which led to a modest revenue beat and significant operating income outperformance.
Active power additions significantly exceeded Street expectations, reaching 1.5GW after a record ~500MW q/q addition, including >300MW in June, which de-risks the Q4 profit inflection as data center sites ramp up.
Revenue backlog was substantially better than expectations, totaling $104.2B, with an additional >$25B in commitments signed in early Q3.
The company's Q2 adjusted operating income of $128M/5.0% nearly doubled Street expectations of $67M/2.6%.
Management successfully addressed concerns about GPU obsolescence and residual value by highlighting an A100 contract extending through 2029 at an attractive price, indicating longer utilization at higher prices.
New customer contracts signed in Q2 came with expected contribution margins 5 to 10 percentage points above those added in recent quarters, indicating improving unit economics.
The company's ability to diversify financing terms through DDTL 5.5 has opened up new markets for clients seeking compute for two to three years.
The significant active power additions in Q2, particularly in June, were well ahead of Street expectations and materially de-risk the second-half revenue and margin ramp.
The company raised its year-end active power guidance to >1.85GW from >1.7GW.
Contracts signed in Q2 carry expected contribution margins 5–10 points above recent deals, indicating improving fundamentals beyond the prior narrative.
Near-term capacity remains effectively sold out, with multiple customers competing for each GPU, and demand is broadening beyond frontier labs and hyperscalers.
The company implemented an approximately 25% increase across SKUs in July, which was not merely a pass-through of higher component costs, and Blackwell and Vera Rubin pricing and margins are reaching new highs.
The A100 contract extending through 2029 at an attractive price provides clear evidence that older GPU residual values may be better than bears assume.
Managed inference booked ARR increased significantly from roughly $1M to >$100M within a few months, with a target of at least $250M exiting 2026, and other products like storage, CPU, networking, and software already exceed $400M of ARR.
These platform products carry higher margins, diversify the company beyond raw GPU rental, and contribute to a more credible full-service AI-cloud narrative.
The company secured more than $32B of debt and equity capital and reduced its weighted-average cost of debt by almost 300bps, resulting in roughly $1.1B of annualized interest savings.
The company is transitioning from a story investors must trust to one it is beginning to prove, with strong execution evidence in adding ~500MW of active power in one quarter.
The combination of 4.2GW of contracted power, >1.5GW of identified options, and >$129B of backlog creates much better visibility into the next several years, with over 75% of the prior >$30B 2027 exit ARR target already contracted.
Unit economics are improving while capacity remains scarce, with new contracts coming in 5–10 points above prior contribution margins and a 25% price increase in July.
The company's ability to monetize both new and maturing fleets through managed inference, software, storage, networking, CPU, and Omni strengthens its position as a full-stack AI cloud.
The Q2 revenue beat was modest at 0.5%, and the FY revenue guide was only slightly higher than Street expectations despite increased active power.
The upside in Q2 operating income was primarily driven by capex/capacity timing leading to lower depreciation burden, rather than improving fundamentals like pricing or contribution margins.
Q3 operating profit guidance was below Street expectations, implying heavy lifting for Q4 to meet sequential margin expansion targets.
The quarter was not completely clean, with a modest revenue beat, gross margin below Street, operating cash flow miss, and decelerated short-term cRPO growth.
Q3 operating-profit guidance missed Street, leaving meaningful execution embedded in Q4.
The company continues to consume extraordinary amounts of capital for growth, with Q2 CapEx nearly four times quarterly revenue and FY26 CapEx guidance increasing significantly.
The Q2 operating-profit beat did not fully flow through to FY guidance, requiring a large Q4 profit to meet targets.
Gross margin missed, operating cash flow missed, and free cash flow remains deeply negative, indicating that growth requires more debt, equity, leases, or prepayments.
Managed inference, while having better gross economics, is characterized as more spot-like and exposed to utilization risk.
Today’s pricing power is a product of severe supply scarcity and may not be a durable moat, as industry capacity could catch demand, leading to normalized pricing and lower gross margins.
The software, inference, and enterprise businesses, while growing rapidly, remain small relative to the company’s core infrastructure commitments.
Shorter contracts improve pricing but reduce duration, and the residual-value thesis could reverse quickly if future GPU generations produce larger performance-per-dollar improvements.
Customer concentration, financing dependence, regulatory delays, and the possibility of hyperscalers attacking the enterprise market remain material risks.