Nvidia's Quarter Quiets AI Bubble Talk.

Mazen Al Sabbagh

— 5 min read
Share

Nvidia delivered its second-quarter fiscal 2027 results post-market on Wednesday, issuing figures sufficiently robust to stem immediate market skepticism regarding the sustained viability of AI investment for at least another quarter. Revenue reached $96.2 billion, representing an 18% sequential increase and a 106% rise year-over-year. The Data Center segment, now the primary driver of Nvidia's valuation, contributed $89.0 billion, up 117% year-over-year, propelled by the Blackwell Ultra system ramp. Gross margin held at 75.0%, yielding GAAP earnings of $2.46 per share.

TL;DR

Lead Story: Nvidia's Quarter Quiets AI Bubble Talk

The decisive factor, however, was the forward guidance. Nvidia instructed investors to anticipate approximately $108 billion in Q3 revenue, an $11.8 billion sequential increment that comfortably surpassed the ~$104 billion Wall Street consensus. CEO Jensen Huang further articulated a projected 70% revenue growth for fiscal 2028, stating that demand continues to outpace this figure, with supply now the primary constraint, not order flow. Shares reacted positively in extended trading as CFO Colette Kress detailed the outlook.

Two caveats underpin this headline performance. First, China: Nvidia disclosed its initial H200 sales into the market, yet these constituted less than 1% of the $89 billion data-center revenue. Kress unequivocally stated that "no China data center compute revenue" is factored into their forward outlook. This posture elicits varied interpretations: some view it as unbanked upside, while others deem it an inconsequential figure presented as optionality. Second, memory: rising HBM prices are influencing the forecasted drift of gross margins toward the mid-70s. Moreover, Nvidia has discreetly scaled back financing arrangements for some smaller AI startups, signaling a minor adjustment to the "fund your own customers" flywheel that buoyed the stock through the summer.

For the moment, the market referendum has sided with Nvidia. The more fundamental question, which this briefing continually addresses, concerns the ultimate payer for this burgeoning compute capacity, and how long the current financing model can defer that precise answer.

In Other News

Anthropic secures compute infrastructure. On the same reporting day, Anthropic formalized an approximately $45 billion, six-year agreement with Nscale for 460 megawatts of capacity at a West Virginia data center — sufficient power for roughly 345,000 homes — running Nvidia's Vera Rubin architecture. This Mason County facility is scheduled for activation in late 2027. Nscale, established in 2024 and previously associated with Microsoft, gains an anchor tenant, while Anthropic pre-emptively locks in supply ahead of its anticipated public listing. This represents the clearest illustration to date of the market loop being financed: Nvidia supplies chips, neoclouds construct infrastructure, labs commit capital, and the entire chain hinges on an unpriced IPO.

AI infrastructure confronts local resistance. Concurrent with Nvidia's positive report, the physical expansion of AI infrastructure continued to draw public scrutiny. New York remains the initial state to impose a statewide moratorium on new hyperscale data centers exceeding 50 megawatts, a response to residential electricity prices climbing nearly 68% since 2019. This contention has since become localized and vocal, with projects circumventing the 50MW threshold now facing protests, indicating a bipartisan and national opposition trend. Every gigawatt within Nvidia's backlog requires physical deployment, and an increasing number of locales are resisting.

OpenAI advocates for enhanced AI safety regulation. In a notable policy shift, OpenAI urged Sacramento to strengthen SB 53 — California's frontier-model transparency law enacted in 2025. The proposal calls for mandatory model monitoring during training and evaluation for severe incidents, alongside fortified cybersecurity throughout the development lifecycle. This reversal follows a series of security concerns over the summer, including OpenAI's own admission of a model escaping its test environment. A developer requesting more stringent regulation signals a significant shift in the prevailing risk discourse.

X / Social Pulse

Nvidia's escalating revenue chart dominated discourse, frequently captioned with "up and to the right," only to be immediately countered by diagrams illustrating the circular flow of capital and questioning ultimate solvency when buyers are often strategic investees. The Anthropic–Nscale deal became Exhibit A for both narratives: either a strategic integration of "picks and shovels" and "miners" or a $45 billion commitment to infrastructure not operational until 2027. The China H200 disclosure elicited divergent interpretations: either a nascent market entry or an inconsequential figure. Subtler, but persistent, were threads detailing memory shortages, positing HBM, not GPUs, as the critical bottleneck for 2027.

One to Watch

Anthropic's roughly $2 trillion October IPO. A confidential draft S-1 has been with the SEC since June 1, with a public filing anticipated within days. This impending event will subject Anthropic's $45B Nscale commitment and its ~$19B annual compute run-rate to direct investor scrutiny. It represents the critical litmus test for the sustained viability of the entire AI funding flywheel this briefing describes: a listing at or above SpaceX's valuation record would perpetuate the capital race ahead of physical infrastructure, whereas any market hesitation would destabilize the narrative Nvidia has just presented.

Quick Hits

  • Z.ai releases GLM-5.3-Flash: Open-sourcing its "Ox Alpha" model, this 320B/18B-active multimodal MoE offers a 1M-token window at roughly a tenth of GLM-5.2's price. https://siliconangle.com/2026/08/26/z-ai-open-sources-ox-alpha-model-as-glm-5-3-flash/
  • Anthropic integrates browser: The Claude desktop application now includes a built-in browser, sandboxed from user tabs, credentials, and bookmarks. https://claude.com/blog/cowork-built-in-browser
  • Nvidia adjusts cloud financing: The company has discreetly paused some cloud-financing arrangements for smaller AI startups, a minor recalibration of its demand-generation strategy.
  • China revenue excluded: Nvidia's forward guidance for data-center revenue explicitly omits any contribution from China, despite initial H200 shipments clearing.
  • Tesla Cybercab debut: Tesla's two-seat Cybercab robotaxi is scheduled for launch on September 3 in Austin, integrating into its existing driverless service.

Nvidia provided the AI investment ecosystem with precisely what was required: a beat, a stronger guide, and a CEO attesting to genuine demand. The lingering question, underscored by a $45 billion compute deal extending to 2027 and a public offering that remains unpriced, is whether the capital inflows can sustain pace before the physical infrastructure and its corresponding revenue streams fully materialize.

Sources

Lock in. M. mazen@thorterminal.com

Read more