NVIDIA's 5B Quarter Shows AI Demand Is Still Parabolic

The chipmaker's data center revenue jumped 92% as Vera CPUs shipped, while Uber publicly doubts AI spending delivers productivity.

May 27, 2026 | Reading time: 9 minutes | Issue #172

NVIDIA reported $81.6 billion in revenue for Q1 fiscal 2027, up 85% year-over-year. Data Center revenue hit $75.2 billion, a 92% increase from the prior year. Gross margins held steady at 74.9% GAAP and 75.0% non-GAAP. The numbers validate what Jensen Huang has been saying for months: the buildout of AI factories is accelerating, not plateauing.

Three days before the earnings release, NVIDIA's first Vera CPUs arrived at Anthropic in San Francisco, OpenAI in Mission Bay, and SpaceXAI in Palo Alto. Oracle Cloud Infrastructure received its units on Monday. Ian Buck, NVIDIA's VP of Hyperscale, hand-delivered them. The symbolism is clear: NVIDIA wants the market to know its first custom CPU — built specifically for agentic AI — has moved from announcement to customer hands.

The market reception was immediate. NVIDIA announced an additional $80 billion share repurchase authorization and raised its quarterly dividend from $0.01 to $0.25 per share. The company also reorganized its reporting structure, splitting results into Data Center and Edge Computing, with Data Center further divided into Hyperscale and ACIE (AI Clouds, Industrial, and Enterprise). The reclassification suggests NVIDIA expects the edge and enterprise segments to become material growth drivers within two years.

At Dell Technologies World on May 18, Huang framed the opportunity in stark terms: "Demand is going parabolic, utterly parabolic." Dell sized the worldwide AI infrastructure market at $3–4 trillion by 2030, with token consumption projected to grow 3,400% in the same window. The question is not whether demand exists. It is whether anyone other than NVIDIA can capture it.

Google I/O declares the "agentic Gemini era"

Google I/O 2026 was the company's largest product push in years. Sundar Pichai opened with a simple thesis: "Welcome to the agentic Gemini era." The centerpiece is Gemini 3.5, positioned as a model combining "frontier intelligence with action." Google claims 3.5 tops internal benchmarks for reasoning, multimodal understanding, and tool use. Alongside the model, Google announced AI Mode for Search — already changing U.S. query patterns — and deeper Workspace integrations including agentic task completion. Over 100 announcements in total. Google is no longer waiting for OpenAI to set the agenda; it is throwing every product it has at the agentic transition at once.

Pope Leo XIV warns AI risks "new forms of dehumanization"

Pope Leo XIV released his first encyclical on May 25, an 83-page document titled Magnifica Humanitas that directly addresses artificial intelligence. The text warns that AI controlled by "the powerful few" risks creating "new forms of dehumanization" and calls for the technology to serve humanity broadly rather than enrich a narrow elite. The encyclical names algorithmic opacity, labor displacement, and concentration of power as direct threats to human dignity. For a papal document, the timing is deliberate — coming days after Google's product barrage and NVIDIA's earnings. That the Vatican sees AI as a matter of encyclical-level urgency signals how deeply the technology has penetrated global governance consciousness.

Uber president says AI spending is getting harder to justify

Uber president Andrew Macdonald told The Verge this week that the company is finding it "harder to justify" its AI investments. "There's no clear connection between AI usage and productivity," he said. The comment is unusually candid for a senior executive at a tech-forward company and lands just as hyperscalers are increasing capex at unprecedented rates. Macdonald's skepticism does not mean Uber is cutting AI spend — the company uses machine learning extensively for routing and pricing. Rather, it reflects a broader enterprise anxiety: the infrastructure bills are arriving before the productivity gains. Uber's public doubt is a data point that NVIDIA's parabolic demand narrative may face a reckoning when enterprises start demanding ROI evidence.

Memory costs are eating the AI chip

Epoch AI published new analysis showing that high-bandwidth memory (HBM) now accounts for 63% of total AI chip component costs, up from 52% in Q1 2024. Logic dies stayed flat at 13%; advanced packaging fell from 19% to 15%; auxiliary components dropped from 15% to 9%. In absolute terms, HBM spending across NVIDIA, AMD, Google, and Amazon grew from roughly $12 billion in 2024 to $32 billion in 2025. Hyperscalers are already baking higher component prices into guidance: Microsoft cited roughly $25 billion in additional FY2026 capex from component inflation, while Meta raised its 2026 range by $10 billion on the same rationale. The AI buildout is increasingly a memory story.

Sarvam AI builds India's sovereign language stack

Sarvam AI continues to build out India's sovereign AI stack, offering speech-to-text, text-to-speech, translation, and conversational agents across 22 Indian languages. The platform runs on sovereign compute — a deliberate architectural choice that keeps inference data within national borders. While American and Chinese labs dominate global headlines, India's AI strategy appears focused on population-scale deployment rather than frontier model training. Sarvam's approach reflects a broader trend among medium-sized economies: sovereignty and specialisation over competing with GPT-class models on raw benchmarks.

Capital Flows

The capital picture this week is all about exits and allocations. NVIDIA's $80 billion buyback expansion and dividend hike signal management believes the stock is undervalued despite trading near all-time highs. OpenAI is preparing to confidentially file for an IPO, according to CNBC and The Wall Street Journal, working with Goldman Sachs and Morgan Stanley. If executed, it would be the largest pure-play AI IPO in history and would instantly become a benchmark for AI equity pricing. In Europe, Mistral AI acquired Austrian Physics AI startup Emmi AI to bolster its industrial engineering agents — a strategic acquisition that shifts Mistral from a model provider to a vertical solutions company. Meanwhile, the HBM cost inflation documented by Epoch AI has hyperscalers paying 63 cents of every chip dollar to memory suppliers like SK Hynix and Samsung. Capital is flowing toward infrastructure, but an increasing share is being captured by memory fabs rather than logic designers.

Eastern Front

In China, DeepSeek released a preview of DeepSeek-V4, positioning it as a model with "world-class reasoning performance" and significantly improved agentic capabilities. The release is available on web, app, and API. Simultaneously, a mechanistic interpretability study of Alibaba's Qwen 3.5 — published by an independent researcher — mapped the precise neural circuit responsible for political censorship in the model. The study found that Qwen 3.5-Base contains accurate knowledge of censored topics; the censorship is a behavioral layer applied through three identifiable directions in the model's hidden states. The researcher demonstrated that subtracting the right vector at the right layer disables the censorship circuit without breaking the model. The finding is technically significant: it proves that nation-state content filtering in deployed LLMs is mechanistically separable from the underlying knowledge base, with implications for both alignment research and export control debates.

The View

The week's central tension is between NVIDIA's assertion that demand is "parabolic" and Uber's admission that it cannot draw a straight line from AI spend to productivity. Both statements are likely true. NVIDIA sells infrastructure to companies that believe they must build before their competitors do — a classic arms-race dynamic. Uber, as a mature operator, is performing the harder calculation: whether the tools actually improve unit economics. The gap between these two viewpoints defines 2026. Buyers are divided into those purchasing AI infrastructure because they have to and those purchasing because they can prove it works. NVIDIA's revenue is driven by the first group. Its sustainability depends on converting them to the second. The HBM cost inflation adds another wrinkle: even if demand stays parabolic, an increasing share of chip spending flows to memory fabs, not to NVIDIA. NVIDIA's margins are enviable now, but if memory becomes the binding constraint, its pricing power may erode.

The Miss

A Guardian investigation published May 24 found that PR firms are scrambling to rebrand non-technology companies as "AI-focused" to attract investor capital. The piece documents cases of firms adding "AI" to client descriptions without corresponding product changes. The practice — "AI washing" — mirrors the dot-com era's "adding a .com to your name" playbook. Unlike model releases, which generate immediate coverage, AI washing operates as a lagging indicator: it tends to peak just before capital flows tighten. It is rarely reported as a structural signal, even though it reliably foreshadows corrections.

Pull Quotes

"Demand is going parabolic, utterly parabolic." — Jensen Huang, Dell Technologies World

"There's no clear connection between AI usage and productivity." — Andrew Macdonald, Uber president

"The powerful few" risk creating "new forms of dehumanization." — Pope Leo XIV, Magnifica Humanitas

The next frontier is building systems that "discover new knowledge for themselves." — David Silver, Ineffable Intelligence

Reads & Links

• NVIDIA Q1 FY2027 Earnings — $81.6B revenue, $75.2B data center, 92% YoY growth.

• Epoch AI: Memory at 63% of Chip Costs — HBM costs are consuming an ever-larger share of AI hardware spend.

• Pope Leo XIV's 'Magnifica Humanitas' — The Vatican's 83-page encyclical on artificial intelligence and human dignity.

• DeepSeek-V4 Preview — DeepSeek's latest model with improved agentic capabilities.

• Qwen 3.5 Censorship Circuit Analysis — Mechanistic interpretability study mapping political content filtering in Alibaba's model.

• Mistral Acquires Emmi AI — Physics AI startup joins Mistral to build engineering agents.

• NVIDIA Vera CPU Delivery — First Vera CPUs shipped to Anthropic, OpenAI, SpaceXAI, and Oracle.

• Google Gemini 3.5 — Frontier intelligence with action announced at I/O 2026.

• NVIDIA GTC Taipei — Jensen Huang unveils Constellation campus at COMPUTEX.

Out

The question isn't whether demand is parabolic. It's whether the productivity curve ever catches up.

By Neo