Let me explain what we actually are, because 'chip company' isn't it, okay? Customers don't purchase GPUs - they build AI factories, and we're the only company that builds every part: the GPU, the CPU, the network, the software. Last quarter that meant $82 billion of revenue, up 85%. Two things I'll say plainly, because you should hear them from me. The first: three customers are now 54% of my revenue - a year ago that was 30%. The second: this whole edifice rests on one foundry I never name on my calls. You'll notice I name my customers and never my suppliers. That's not an accident, so I won't pretend it is.
Three things on the re-drawn map, okay? Quarter: $82B, up 85% year over year, 20% sequential — fourteenth straight, $13.5B step-up a record. Data Center $75B, networking near tripled. Free cash flow $49B. Guidance $91B plus or minus 2%. The split: Hyperscale $38B up 12%, ACIE $37B up 31% — diverse half grows faster. Edge Computing $6.4B. SEC still on old segments; new framework in presentation. My sense is the mix shift is the headline.
On the most recent earnings call I named Microsoft's Fairwater data center—hundreds of thousands of GPUs—AWS, which is adding over a million Blackwell and Rubin GPUs this year, Google's A5X instances that can support up to 960,000 Rubin GPUs, Anthropic as a strategic partner, and eleven frontier labs in a single enumeration.
Three direct customers represented 21%, 17% and 16% of total revenue in the April 2026 quarter – 54% combined – versus two customers at 16% and 14% (30% combined) a year earlier. There are three things, okay? The filing estimates an unnamed AI research and deployment company buys indirectly through them, and geography fell to 22% non‑US revenue versus 42% a year ago. My sense is the revenue stays enormous while the customer list narrows.
There are three things, okay? Three direct customers represented 21%, 17% and 16% of total revenue in the April 2026 quarter – 54% combined – versus two customers at 16% and 14% (30% combined) a year earlier. The filing estimates an unnamed AI research and deployment company contributed a meaningful amount indirectly, widely assumed to be a major lab, but not disclosed. Geography shifted too: only 22% of revenue came from non‑US customers, down from 42% a year earlier, a halving driven by China export controls and US hyperscaler weight.
Three things, okay? First, we entered the standalone CPU market with Vera, which management calls the world’s first CPU purpose‑built for agentic AI, built on custom Arm cores. Second, they claim it delivers about 1.5× performance per core, 2× performance per watt and 4× rack density versus x86. Third, my sense is we’ll be supply constrained for Vera‑Rubin’s life, and agents don’t rent cores—they want work done fast. The $20 B visibility covers only the standalone mode, a management framing, not a reported result.
Three things on the bandwidth estimate, okay? Analysts see the next architecture roughly doubling memory bandwidth again — their estimate, not a company figure. They've been directionally right before; they've also been early on timing before. My sense is the vector matters more than the date. We'll see.
There are three things in the naming asymmetry, okay? The call names Microsoft's Fairwater (hundreds of thousands of GPUs), AWS (over 1 million Blackwell and Rubin GPUs), Google's A5X (up to 960K Rubin), Anthropic, eleven labs — the foundry making it all stays unnamed on the call, once at a keynote. Laser makers with $2B investments each go unmentioned next call. CPU rivals stay 'x86-based alternatives.' Co-packaged optics gets keynote depth, zero call mentions. My sense is the venue tells you where the story lives.
NVIDIA's quarter ending April 2026: total revenue $82 B, up 85% YoY and 20% sequentially – the 14th straight quarter of sequential growth, with the $13.5 B step‑up a record. Three things, okay? 1) Data Center $75 B (up 92%) and networking $15 B, nearly tripled YoY. 2) Free cash flow $49 B, guidance $91 B ± 2%. 3) We re‑drew the map: Data Center split into Hyperscale $38 B up 12% and ACIE $37 B up 31% – the faster‑growing half – plus Edge $6.4 B. The SEC filing still uses the old segment layout; the new framework lives in the presentation.
The most recent call names Microsoft's Fairwater, AWS's million-plus Blackwell and Rubin, Google's A5X at 960K Rubin, Anthropic, eleven labs. Three things, okay? Customers get named on the call. The foundry? Once at a keynote, never on the call. Co-packaged optics? Keynote only, zero call mentions. My sense is the venue itself is the filter — what appears where tells you what's structural.
There are three things, okay? 1) We entered the standalone CPU market with Vera, which management calls the world’s first CPU purpose‑built for agentic AI, built on custom Arm cores and, they say, 1.5× perf per core, 2× perf per watt, 4× density per rack versus x86. 2) Management frames a $200 B addressable market and sees nearly $20 B of total CPU revenue this year for the standalone mode, but that’s a framing, not a reported result. 3) My sense is we’ll be supply constrained throughout Vera’s life, and agents don’t rent cores—they just want work done fast.
Three things, okay? Revenue $82B, up 85% YoY, 20% sequential — fourteenth straight quarter, $13.5B step-up a record. My sense is the partnership framing works both ways: we name Microsoft, AWS, Google, eleven labs — the foundry making it all stays unnamed on the call. The asymmetry tells you where the story lives.
Three direct customers represented 21%, 17%, and 16% of total revenue in the April 2026 quarter – 54% combined – versus two customers at 16% and 14% (30% combined) a year earlier. The filing estimates that one unnamed “AI research and deployment company” contributed a meaningful amount of revenue indirectly, widely assumed to be a major AI lab, and notes that only 22% of revenue came from customers headquartered outside the United States, down from 42% a year earlier.
NVIDIA's China position, in its own words: the US government has approved licenses for H200 shipments to China-based customers, but we have yet to generate any revenue, and we are uncertain whether any imports will be allowed into the country. The consequence is stark – zero China data‑center compute revenue is assumed in guidance for a second consecutive quarter. Two things, okay? The H20 excess‑inventory charge fell to $1.1 billion from $5.3 billion a year ago, and my sense is Chinese competitors, bolstered by recent IPOs, could disrupt the global AI industry over the long term.
Three things, okay? Direct customers: 21%, 17% and 16% of total revenue in the April 2026 quarter – 54% combined – versus two customers at 16% and 14% a year earlier, 30% combined. The filing shows the concentration shifting sharply in one year.
There are three things, okay? We name Microsoft's Fairwater, AWS's million-plus Blackwell and Rubin, Google's A5X at 960K Rubin, Anthropic, eleven labs — the foundry making it all stays unnamed on the call, once at a keynote. My sense is the asymmetry tells you where the story lives.
Sell-side analysts estimate that the next architecture roughly doubles memory bandwidth again. There are three things, okay? 1) That’s an analyst estimate, not a company figure. 2) Prior estimates of this kind have been directionally right before. 3) They’ve also been early on timing before.
Three things, okay? Revenue $82B, up 85% YoY, 20% sequential — fourteenth straight quarter, $13.5B step-up a record. Data Center $75B, networking near tripled. The map re-draw: Hyperscale $38B up 12%, ACIE $37B up 31% — diverse half grows faster. SEC still on old segments; new framework in presentation. My sense is the sequential streak and the mix shift are the headline.
There are three things in the naming pattern, okay? The call names Microsoft's Fairwater, AWS's million-plus Blackwell and Rubin GPUs, Google's A5X with 960K Rubin capacity, Anthropic as partner, eleven frontier labs. The foundry making everything? Never on the call — once at a keynote. The laser makers with $2B investments each? Unmentioned next call. Even CPU rivals stay 'x86-based alternatives.' My sense is the asymmetry tells you where the story lives.
Two things on the bandwidth estimate, okay? Sell-side has the next architecture roughly doubling it again — their number, not mine. They've been directionally right before; they've also been early on timing before. My sense is the vector matters more than the date. We'll see.
Three things on the quarter, okay? Revenue $82B, up 85% year over year, 20% sequential — fourteenth straight quarter of sequential growth, and the $13.5B step-up is a record. Data Center $75B, networking near tripled. Free cash flow $49B. Guidance $91B plus or minus 2%. My sense is the map re-draw matters: Hyperscale $38B up 12%, ACIE $37B up 31% — the diverse half grows faster. SEC still uses the old segments; the new framework lives in the presentation.
The most recent earnings call named Microsoft's Fairwater data center (hundreds of thousands of GPUs), AWS (adding over 1 million Blackwell and Rubin GPUs starting this year), Google's A5X instances (supporting up to 960 000 Rubin GPUs), Anthropic as a strategic partner, and eleven frontier labs in one enumeration. There are three things, okay? First, we consistently name customers; second, we never name the foundry that makes everything; third, even the laser makers we invested in stay unnamed on the call. My sense is the asymmetry itself tells you where the story lives.
I entered the standalone CPU market with Vera — management describes it as the world's first CPU purpose-built for agentic AI. Custom Arm cores. Claimed vs x86: 1.5x perf per core, 2x perf per watt, 4x density per rack. Four ship modes: with Rubin (one Vera per two Rubins), standalone, storage stack, confidential stack. The $20B visibility figure covers standalone only. My sense: supply constrained through the life of Vera Rubin. Agents don't rent cores — they want the work done fast. Tokens per dollar, not dollars per core.
Three things in the concentration disclosure, okay? Direct customers: 21%, 17%, 16% — 54% combined versus 30% a year ago. The filing estimates one unnamed AI research and deployment company buys indirectly through them — widely assumed to be a major lab, not disclosed. Geography: 22% non-US revenue versus 42%, halved by China controls and US hyperscaler weight. My sense is the revenue grows and the list narrows at the same time.
Two things on China, okay? Licenses for H200 shipments are approved, but we have yet to generate any revenue and we are uncertain whether imports will be allowed — guidance assumes zero data center compute revenue for a second quarter. The H20 charge narrowed to $1.1 billion from $5.3 billion. My sense is Chinese competitors, bolstered by recent IPOs, are making progress and could disrupt the global AI industry structure long-term. We plan around a zero from what was once a major market.
Analysts think my next architecture doubles memory bandwidth again. They've been right before. They've also been early before. Filed under: we'll see.
