Nvidia
NVDA · NASDAQEvery AI model on Earth eventually needs to run on a GPU — and for most of them, that GPU says Nvidia.
Who they are
Nvidia started life in 1993 as a maker of graphics chips for PC gaming, and for over a decade that's exactly what it was. The shift began around 2006, when Nvidia built CUDA, a programming platform that let developers use its graphics chips for general-purpose computing, not just rendering pixels. Almost twenty years later, that decision looks less like a side project and more like the reason Nvidia now sits at the center of the entire AI industry: more than five million developers build on CUDA today, and most AI frameworks are optimized for Nvidia hardware first. That software moat is what separates Nvidia from rivals who can build a fast chip but can't replicate two decades of tooling built around it.
What they do
Nvidia doesn't really sell chips anymore — it sells rack-scale systems. The current Blackwell platform bundles GPUs with high-speed NVLink networking and a full software stack into what the company calls "AI factories," and it has reportedly been sold out through mid-2026. The next platform, Vera Rubin, pairs a new custom Vera CPU with a Rubin GPU and, according to Nvidia, ramped into full production on May 31, 2026. Where Blackwell was framed around raw training power, Vera Rubin is explicitly built for agentic AI — workloads where a single user request can trigger many internal model calls, retrieval steps, and tool invocations, all of which need to run fast and cheap at scale.
How Nvidia makes money
Data center systems are now almost the entire business. Nvidia's fiscal 2026 (ended January 2026) revenue reached $215.9 billion, up 65% year over year, and the first quarter of fiscal 2027 (reported May 2026) posted a record $81.6 billion, up 85%, with data center revenue alone hitting $75.2 billion. Gaming and professional visualization still exist, but they're now a rounding error next to the scale of the AI buildout. CEO Jensen Huang told Nvidia's GTC 2026 conference that Blackwell and Vera Rubin combined already have visibility into "at least" $1 trillion in cumulative revenue through 2027 — and added that he expects real demand to run higher than that.
The bigger trend
The frontier is shifting from "build the model" to "run the model everywhere, all the time, cheaply enough that nobody notices the meter running." That's why Nvidia's roadmap now emphasizes platform throughput and cost per token as much as raw performance. It's also why the company's own words about supply are worth noting: at GTC 2026, Huang was blunt that "we are going to be short," framing the constraint on Nvidia's growth as manufacturing and packaging capacity, not customer demand. Two risks sit underneath the growth story. First, competition: hyperscalers including Google and Amazon design their own custom AI chips (with help from Broadcom and Marvell), which compete for the same capital budgets as Nvidia GPUs. Second, geopolitics: US export controls have forced Nvidia to exclude China data-center compute revenue from certain forecasts, closing off part of one of the world's largest technology markets.
Whether Vera Rubin's ramp keeps pace with the $1 trillion cumulative-revenue assumption Huang laid out; how much future hyperscaler AI capex shifts toward custom in-house chips instead of Nvidia GPUs; and whether US-China export policy tightens further or eases, since it directly gates how much of the China market Nvidia can address at all.
Related companies
Frequently asked questions
Not anymore. Nvidia increasingly sells complete rack-scale systems that bundle GPUs, a custom CPU (Vera), high-speed networking, and software into a single deployable "AI factory," rather than standalone chips sold individually.
Blackwell is Nvidia's current-generation platform, shipping in volume since 2024–2025 and reportedly sold out through mid-2026. Vera Rubin is the next platform, pairing a new Vera CPU with a Rubin GPU, which Nvidia said ramped into full production on May 31, 2026, built specifically for agentic AI workloads.
US rules restrict Nvidia's ability to sell its most advanced AI chips into China. Nvidia's recent outlooks have explicitly excluded data-center compute revenue from China in certain forecasts, reflecting how tightening export rules limit its addressable market in one of the world's largest technology economies.
This page describes public value-chain positioning for informational purposes only. It is not investment advice, and inclusion here is not a recommendation to buy or sell any security. Figures reflect public reporting as of mid-2026 and may have changed since.