Nvidia Logs $96.2B Quarter as Vera Rubin Ramp and AWS Commitment Frame the Next AI Infrastructure Cycle
A blowout fiscal second quarter, the fastest product ramp in Nvidia's history, and a cascade of hyperscaler commitments drove shares 8.7% higher, though export-control investigations and emerging custom-silicon competition ensure the road ahead is not one-sided.
Nvidia shares surged 8.7% to close at $227.98 on August 28, 2026, as fiscal second-quarter results cleared a high bar set by a market that had spent months pricing in extraordinary growth. Revenue reached $96.2 billion, up 18% sequentially and 106% year-over-year — a rate of expansion that, measured against the company's $26.9 billion in fiscal 2022 revenue, illustrates how profoundly the generative-AI buildout has reconstructed Nvidia's financial profile in four years. The day's move reflected not just the reported numbers but the guidance attached to them: the company's CFO has stated publicly that demand is tracking toward 140% growth, even as constrained supply can currently satisfy only roughly half of that trajectory — a ratio that simultaneously explains the stock's premium and the anxiety embedded in it.
The most consequential disclosure from this earnings cycle was the Vera Rubin ramp. Nvidia says it expects to book $20 billion in Vera Rubin system sales in the fiscal third quarter alone — which the company characterizes as the fastest product ramp in its history and which would represent approximately 20% of data center revenue in a single quarter. Demand from hyperscalers has been concrete rather than directional: AWS announced a commitment to deploy two million additional Nvidia GPUs alongside next-generation infrastructure for agentic and physical AI workloads. IREN has signed a $3.4 billion, five-year cloud contract structured around Nvidia hardware; Nscale is reportedly in a $45 billion arrangement to deploy Vera Rubin chips specifically for Anthropic's infrastructure needs; and SpaceX's chief executive stated the company will rely exclusively on Nvidia compute, with ambitions to reach 10 gigawatts of capacity by end of 2027. India has already deployed its first Vera Rubin cluster — 9,000 GPUs running on green power — having navigated export-control constraints in the process.
The supply chain underpinning all of this is being assembled under visible pressure. SK Hynix broke ground this week on a high-bandwidth memory packaging facility in Indiana, targeting mass production by 2029 and explicitly positioned to serve Nvidia. Simultaneously, Nvidia extended NVLink Fusion to support NVHBM, a custom high-bandwidth memory specification developed with Amazon's Annapurna Labs as the first formal partner — a move that deepens ecosystem integration while broadening the available supply base for next-generation memory. Nvidia is separately pressing Samsung, SK Hynix, and Micron for 16-layer HBM4 deliveries in Q4 2026. The scarcity dynamic is already visible in pricing: the company is reportedly raising server system prices by approximately 15%, offering the projection that memory costs will exceed GPU costs for cloud operators by 2027 as partial justification. Nvidia has also disclosed a $105 billion liability cap associated with a 4.25-gigawatt Ohio data center project, a figure that conveys the capital-intensive scale of commitments now embedded in its infrastructure ambitions.
Two structural risks warrant equal attention alongside the bullish narrative. On the regulatory front, U.S. authorities are actively investigating Apex Logistics over suspected shipments of Nvidia accelerators to China, and the administration is separately preparing to close loopholes that have enabled indirect export of restricted AI chips to Chinese buyers; a reportedly pending $13 billion Nvidia AI deal carries a geopolitical dimension that remains unresolved. On the competitive front, OpenAI's in-house Jalapeño chip claims benchmark superiority over Nvidia's Blackwell architecture on AI inference — a claim that, if validated at production scale, would pressure Nvidia's position in the fastest-growing workload segment. Amazon's custom silicon is now running at a reported $25 billion annualized rate, a figure suggesting that in-house chip programs have matured well beyond experimental status. The market's reaction to the earnings was itself instructive: AMD fell while Intel edged higher, a divergence the investing community read as consolidating Nvidia's training dominance while acknowledging possible structural changes in inference economics and foundry demand.
Two longer-dated strategic moves add texture to the picture. Nvidia has announced the formation of an employee-funded political action committee — an unusual step for a hardware company and a signal that Washington's decisions on export licensing, antitrust, and defense procurement have become genuinely central to the company's competitive calculus. Reports — as yet unconfirmed by Nvidia — suggest the company is in discussions to acquire Hugging Face for approximately $12.9 billion; if completed, such a transaction would represent a meaningful extension from hardware into the open-model software layer that sits atop its accelerator stack. For investors and analysts tracking the next phase of the Nvidia story, three signals warrant close attention in the coming quarters: whether Vera Rubin shipments validate the $20 billion Q3 guidance figure; whether tightened China export controls materially reduce Nvidia's addressable revenue; and whether hyperscaler custom silicon — particularly Amazon's Trainium and Inferentia programs — begins capturing net-new GPU orders rather than merely supplementing Nvidia deployments. The answers will determine whether today's 8.7% move marked the opening of the next leg or a temporary clearing of an unusually crowded anxiety list.