NVIDIA GTC 2026 Recap: Keynote, Announcements, and Verdict

TL;DR: NVIDIA GTC 2026 ran March 16-19, 2026, in San Jose, California, with Jensen Huang’s keynote delivered from the 17,000-seat SAP Center on Monday, March 16. The headline reveal was the Vera Rubin computing platform - NVIDIA’s next-generation AI infrastructure stack - alongside a new agentic AI framework called OpenClaw, a fresh open-model coalition, and a roughly $1 trillion revenue outlook through 2027. NVIDIA’s own pre-event numbers put attendance north of 30,000 people from 190+ countries, and in-person passes sold out before the show started.

I track NVIDIA GTC every year for the same reason I test AI tools with my own money before recommending them: what gets announced on that stage in March tends to show up as shipping product inside AI tools I review six to twelve months later. Vera Rubin, the AI infrastructure platform NVIDIA unveiled at NVIDIA GTC 2026, is the kind of announcement that eventually reshapes the pricing and speed of the AI SaaS tools covered on this site, even though most readers here will never touch a GPU rack directly. GTC also sits at the top of my AI events hub for exactly that reason - the announcements made on this stage become product roadmaps a few quarters later.

This is not a “should you attend” guide. GTC 2026 already happened. This is a recap: what NVIDIA actually said, what shipped, how many people showed up, and an honest verdict on whether the trip (or the free livestream) was worth it. Every claim below is sourced, and where the numbers from NVIDIA’s own channels do not perfectly agree with each other, I say so instead of picking whichever figure sounds cleaner.

Key Takeaways

• The dates were March 16-19, 2026, not March 16-20. NVIDIA’s own March 3, 2026 press release confirms the four-day run in San Jose, matching Wikipedia’s event record. A handful of third-party event-listing sites show a fifth day; NVIDIA’s official announcement does not.
• Vera Rubin was the headline hardware reveal. The new platform packages seven chip types into five rack-scale systems and one AI supercomputer, hitting 3.6 exaflops of compute and 260 terabytes per second of NVLink bandwidth, according to NVIDIA’s own technical briefing.
• OpenClaw became NVIDIA’s agentic AI centerpiece. Huang called the open source agent framework “the most popular open source project in the history of humanity” and announced platform-wide support plus an enterprise-secure NemoClaw stack.
• NVIDIA raised its own demand outlook to roughly $1 trillion through 2027. That is Huang’s own figure from the keynote stage, not a third-party analyst estimate, and it is worth treating with the same skepticism I apply to any vendor’s own growth forecast.
• In-person passes sold out. NVIDIA confirmed this on its own social channels before the event started, which tells you more about real demand than any attendance projection does.

What Is NVIDIA GTC?

NVIDIA GTC (GPU Technology Conference) is NVIDIA’s flagship annual event, and it has moved a long way from its 2009 roots as a niche GPU-computing meetup that drew about 1,500 attendees to a Fairmont hotel ballroom. Today it functions as the industry’s closest thing to an AI State of the Union, where NVIDIA CEO Jensen Huang sets the technical agenda for the next 12 months in a single keynote.

The conference runs multiple editions a year - a flagship March edition in San Jose plus smaller regional shows in Paris, Taipei, and Washington D.C. - but San Jose is the one that matters for roadmap-watching: it is where the year’s core hardware architecture and platform partnerships get named first, per GTC’s documented event history on Wikipedia.

What used to be a hardware-and-graphics show is now, in Huang’s own words from a prior keynote, the “Super Bowl of AI.” Attendees skew heavily toward business leaders and enterprise decision-makers rather than pure developers, and multi-year GPU procurement conversations now happen on the show floor alongside the technical sessions.

What Happened at NVIDIA GTC 2026: Dates, Venue, and Scale

NVIDIA GTC 2026 ran March 16-19, 2026, in San Jose, California. NVIDIA turned downtown San Jose into what its own press materials called an “AI campus,” spanning 10 venues across the city, anchored by the San Jose McEnery Convention Center. Huang’s keynote, however, was delivered from a different location entirely: the SAP Center, San Jose’s 17,000-seat hockey arena, the same venue GTC used in 2025 to accommodate keynote demand that had outgrown the convention center’s own auditorium space.

The keynote opened Monday, March 16, at 11 a.m. PT, preceded by a GTC Live pregame show starting at 8 a.m. PT featuring venture investors Sarah Guo (Conviction), Gavin Baker (Atreides Management), and Alfred Lin (Sequoia Capital) interviewing a rotating cast of AI-native CEOs, including Perplexity’s Aravind Srinivas, LangChain’s Harrison Chase, and Mistral AI’s Arthur Mensch. NVIDIA also held a separate financial analyst Q&A for investors on Tuesday, March 17, at 9 a.m. PT, a sign of how much the event now doubles as an investor-relations moment, not just a developer conference.

Quick facts:

Detail	Information
Event	NVIDIA GTC 2026
Dates	March 16-19, 2026
Keynote venue	SAP Center, 525 W. Santa Clara St., San Jose, CA (17,000-seat capacity)
Conference footprint	San Jose McEnery Convention Center plus 10 total venues across downtown San Jose
Format	Hybrid - in-person plus free virtual keynote livestream
Sessions	1,000+ technical sessions, 2,000+ speakers, 450+ sponsoring companies
Startups showcased	240+ NVIDIA Inception startups
Keynote speaker	Jensen Huang, NVIDIA founder and CEO
On-demand access	Free at nvidia.com/gtc/keynote, no registration required
Official recap source	NVIDIA’s official GTC 2026 live blog

The financial-analyst-Q&A detail matters more than it looks. A developer conference does not usually need a dedicated investor session the day after the keynote. GTC has become a hybrid event that serves NVIDIA’s stock story and its developer ecosystem simultaneously, and that dual purpose shapes what gets announced on stage - big, headline-friendly platform names, trillion-dollar demand figures, and splashy demos alongside the technical substance.

Keynote Recap: What Jensen Huang Actually Said

Huang’s keynote ran just over two hours, opening with a video framing the “token” as the fundamental unit of modern AI, then moving into what NVIDIA’s own team has branded the “five-layer cake” of AI: energy, chips, infrastructure, models, and applications. Huang used this framing to argue that every layer of that stack is advancing simultaneously, which is the underlying thesis for nearly everything else announced that day.

He opened by marking CUDA’s 20th anniversary, calling it the “flywheel” behind accelerated computing, and noted the scale of the show itself: 450+ sponsors, 1,000 sessions, and 2,000 speakers, according to NVIDIA’s own live coverage of the keynote. From there, Huang walked through NVIDIA’s data-processing partnerships with IBM, Dell, Google Cloud, AWS, Microsoft Azure, Oracle, and CoreWeave before turning to the year’s two biggest reveals: the Vera Rubin computing platform and NVIDIA’s push into agentic AI through OpenClaw.

One quote worth sitting with, direct from NVIDIA’s official press release announcing the event: “AI is no longer a single breakthrough or application - it is essential infrastructure. Every company will use it. Every nation will build it,” Huang said. That framing, decided weeks before the keynote even happened, previews exactly the ground the keynote covered: infrastructure at planetary scale, not a single flashy product demo.

The technical press that covered the keynote independently backs up NVIDIA’s own account of the scale involved. Atlan’s technical recap of the GTC 2026 keynote notes that Huang delivered the address “to a sold-out arena,” and that third-party performance analysis from SemiAnalysis independently confirmed - and in some cases exceeded - NVIDIA’s own claimed performance numbers for the new hardware, finding roughly 50x more tokens per watt compared to the prior-generation Hopper H200 chips. That kind of independent verification is worth flagging specifically because vendor keynote claims should never be taken at face value, and in this case a third party checked the math and the numbers held up.

The keynote closed on a lighter note: Disney’s Olaf, from Frozen, walked on stage as a live physical AI demo powered by NVIDIA’s Newton physics engine and Omniverse simulation platform, followed by an AI-generated musical sing-along video recapping the day’s announcements. It is a strange closing beat for a hardware keynote, but it doubles as a live demo of the same simulation technology NVIDIA is selling to robotics and manufacturing customers - the entertainment bit and the sales pitch are the same demo.

Major Announcements and Product News From GTC 2026

This is the part search traffic actually wants, so here is what shipped or got confirmed on stage, in order of how much it matters for anyone actually building on NVIDIA’s stack.

Vera Rubin platform. The headline hardware announcement. Vera Rubin packages seven chip types - a new Vera CPU, Rubin GPUs, NVLink 6 switches, ConnectX-9 NICs, BlueField-4 DPUs, Spectrum-X optical NICs, and a Groq 3 LPU - into five rack-scale systems that function as a single AI supercomputer. NVIDIA’s own numbers: 3.6 exaflops of compute, 260 terabytes per second of all-to-all NVLink bandwidth, fully liquid-cooled at 45 degrees Celsius, with rack installation time cut from two days to two hours. The platform is in production now, not a future promise, per NVIDIA’s official Vera Rubin platform announcement.

The Groq acquisition pays off. NVIDIA quietly acquired the Groq team and its LPU (language processing unit) technology in late 2025, and GTC 2026 was where that acquisition showed up in shipping hardware. The Groq 3 LPU handles the decode phase of AI inference - where token-generation speed is the bottleneck - while Vera Rubin handles the prefill phase, with a software layer called NVIDIA Dynamo splitting the work between the two chip architectures. Independent analysis cited in the Atlan recap put the combined throughput gain at roughly 35x per megawatt compared to running Blackwell alone.

OpenClaw and NemoClaw. NVIDIA’s biggest software-side bet of the keynote. OpenClaw is an open source agentic AI framework built by independent developer Peter Steinberger, and Huang called it “the most popular open source project in the history of humanity” on stage. NVIDIA announced platform-wide support for it, plus a new enterprise-hardened stack called NemoClaw that adds policy enforcement, network guardrails, and privacy routing so companies can deploy always-on AI agents without losing control of what those agents can touch.

Nemotron Coalition - six open model families. NVIDIA expanded its open-model ecosystem into a formal coalition spanning six frontier model families: Nemotron (language and reasoning), Cosmos (world and vision models), Isaac GR00T (general-purpose robotics), Alpamayo (autonomous driving), BioNeMo (biology and chemistry), and Earth-2 (weather and climate modeling). This is NVIDIA positioning itself as an open-model backer across nearly every AI application category at once, not just a chip vendor.

Physical AI and robotaxis. NVIDIA’s robotaxi-ready DRIVE Hyperion platform picked up new automaker partners - BYD, Hyundai, Nissan, and Geely - plus a deployment partnership with Uber to put these vehicles into its ride-hailing network. On the industrial side, NVIDIA’s IGX Thor edge-AI platform went generally available, with named enterprise adopters including Caterpillar, Hitachi Rail, Johnson & Johnson, and Medtronic building real-time physical AI into construction equipment, rail inspection, and surgical robotics.

The Feynman roadmap. Looking past Vera Rubin, Huang previewed NVIDIA’s next architecture generation, Feynman, which includes a new GPU, an LP40 LPU built jointly with the former Groq team, a new Rosa CPU (named for Rosalind Franklin), and a next-generation BlueField-5 networking chip. NVIDIA also announced it is building toward orbital AI data centers under the name NVIDIA Space-1 Vera Rubin - a genuinely unusual reveal that got less press coverage than it probably deserved.

DLSS 5. On the graphics side - GTC’s original reason for existing before AI took over - NVIDIA announced DLSS 5, an AI-powered neural rendering upgrade that uses 3D-guided techniques to hit real-time, photoreal 4K rendering on local consumer hardware.

The $1 trillion figure. Huang told the keynote audience he now expects roughly $1 trillion in cumulative NVIDIA revenue from 2025 through 2027, driven by what he framed as an “inference inflection” - a shift from training-heavy AI workloads to inference-heavy ones, per NVIDIA’s own live coverage of the keynote. Treat this the same way you would treat any company’s own multi-year revenue projection from its own CEO on its own stage: directionally informative, not a guarantee.

Attendance and Scale: How Big Was GTC 2026, Really?

NVIDIA’s own numbers here are worth reading carefully, because two different official figures do not fully agree. NVIDIA’s March 3, 2026 pre-event press release projected “more than 30,000 attendees … from over 190 countries,” spanning developers, researchers, business leaders, and AI-native companies. That figure covers the full event footprint, including virtual attendance and the broader global audience NVIDIA counts across its regional programming.

Separately, NVIDIA’s own social media announcement confirming that in-person conference passes had sold out cited “over 20,000 attendees” specifically for the in-person San Jose show. Both numbers come from NVIDIA directly - a smaller in-person crowd inside a larger global registrant count is a normal pattern for a hybrid event. What is genuinely confirmed, rather than projected, is that passes sold out before the event started.

For scale context, NVIDIA’s own numbers put the event at 450+ sponsoring companies, 1,000+ technical sessions, 2,000+ speakers, 240+ NVIDIA Inception startups, and 150+ research poster presentations - closer to a trade show crossed with an academic conference than a typical vendor developer day.

One honest caveat: I could not find an official, single, final post-event attendance number that NVIDIA confirmed after the show closed. The 30,000+ figure is a pre-event projection; the 20,000+ figure is tied specifically to the sold-out-passes announcement. I would rather say that plainly than round one of these figures into something more precise-sounding than it actually is.

Who Should Have Attended NVIDIA GTC 2026

Looking back at what actually got covered, GTC 2026 was worth the trip (or worth blocking calendar time for the free livestream) for a specific set of roles, and a clear miss for others.

AI infrastructure and MLOps engineers got the most direct value. Vera Rubin’s architecture details, the Groq-powered inference split, and the Dynamo software layer are the kind of specifics that directly change capacity-planning decisions for anyone running GPU fleets, and that level of detail was not available anywhere else before this keynote.

Robotics, physical AI, and manufacturing teams had a genuinely strong reason to attend, given the IGX Thor general-availability announcement, named enterprise deployments across surgical robotics and rail infrastructure, and the robotaxi partnership news. If your roadmap touches physical automation, this was the single most information-dense day of the year for that specific niche.

Founders and product teams building AI agents should care most about the OpenClaw and NemoClaw announcements specifically, even if they skipped the rest of the show. NVIDIA effectively endorsed an entire category of always-on agent tooling this year, and that endorsement tends to accelerate enterprise adoption of the underlying open source project. If you build or evaluate AI agent tooling, this is worth a closer look - my breakdown of the best MCP servers covers the adjacent agent-infrastructure ecosystem this kind of announcement tends to feed into.

Enterprise IT decision-makers weighing multi-year GPU spend had real reason to attend given the $1 trillion demand framing and the Vera Rubin production timeline, but should treat Huang’s own revenue projection with appropriate skepticism rather than as an independent forecast.

Pure application-layer AI builders with no infrastructure decisions on their plate got comparatively little from this specific keynote beyond the Nemotron Coalition and DLSS 5 news. The free virtual keynote covers the headlines in two hours; the in-person trip mostly pays off for people making hardware or infrastructure decisions.

Was NVIDIA GTC 2026 Worth It? My Honest Take

Here is the part where I would normally hedge, and I am not going to. For the audience this event is actually built for - infrastructure engineers, robotics and physical-AI teams, and enterprise buyers making multi-year GPU commitments - GTC 2026 delivered. Vera Rubin’s specs are real and independently verified by a third party that had every incentive to poke holes in NVIDIA’s own claims and did not find enough to dispute them. The OpenClaw endorsement is the kind of platform-level bet that tends to actually move enterprise adoption, not just generate a press cycle.

For everyone else - the much larger group of people who read about GTC after the fact rather than sitting in the SAP Center - the honest verdict is that the free on-demand keynote replay covers essentially everything that mattered. NVIDIA does not gate the keynote behind registration, and there was no in-person-only announcement in this recap that changes that calculus. If your job does not involve GPU procurement, chip architecture, or physical AI deployment, watching the free replay gets you the same information the $2,000+ in-person pass did, minus the sold-out-arena atmosphere and the Denny’s pancakes NVIDIA has historically handed out before the keynote.

The one thing I would flag as a real limitation of this year’s show: the dual-purpose nature of GTC - developer conference and investor-relations event happening on the same stage - means some of the framing (the $1 trillion figure especially) is doing double duty as a stock-story talking point. That does not make the technical announcements less real, but it is worth reading the keynote with that lens rather than treating every number as pure engineering disclosure.

Frequently Asked Questions

When did NVIDIA GTC 2026 take place?

NVIDIA GTC 2026 ran March 16-19, 2026, in San Jose, California. NVIDIA’s own March 3, 2026 press release confirms this four-day window; some third-party event-listing pages show a fifth day, but that does not match NVIDIA’s official announcement.

What was the biggest announcement at GTC 2026?

The Vera Rubin computing platform was the headline hardware reveal - a full-stack AI infrastructure platform combining seven chip types into five rack-scale systems, now in production. NVIDIA’s agentic AI push through OpenClaw and NemoClaw was the biggest software-side announcement.

Where was the NVIDIA GTC 2026 keynote held?

Jensen Huang delivered the keynote from the SAP Center, a 17,000-seat arena in San Jose, on Monday, March 16, at 11 a.m. PT. The broader conference spanned 10 venues across downtown San Jose, anchored by the San Jose McEnery Convention Center.

How many people attended NVIDIA GTC 2026?

NVIDIA’s pre-event figures projected more than 30,000 attendees from over 190 countries across the full hybrid event; a separate NVIDIA announcement confirming sold-out in-person passes cited over 20,000 attendees specifically for the San Jose show. NVIDIA has not published a single final confirmed post-event headcount.

Can I still watch the NVIDIA GTC 2026 keynote?

Yes. NVIDIA hosts the full keynote and session recordings on demand at nvidia.com/gtc/keynote, free and with no registration required, per NVIDIA’s own on-demand keynote page.

The Bottom Line

NVIDIA GTC 2026 delivered exactly what a conference of its scale should: a genuinely significant hardware platform in Vera Rubin, a real strategic bet on agentic AI through OpenClaw, and enough independently verified technical substance that the keynote’s claims held up under third-party scrutiny. The dates were March 16-19, not March 16-20 as some listings suggest, and in-person passes sold out before the show started, which is the clearest signal of real demand NVIDIA offered all week.

If you build on NVIDIA’s infrastructure stack or make GPU procurement decisions, this was worth attending or at minimum worth blocking the calendar for the free keynote livestream. If you sit purely at the application layer, the two-hour on-demand replay covers the parts that matter to you.

NVIDIA GTC 2027 is already confirmed for March 14-18, 2027, back at the San Jose McEnery Convention Center - see our full NVIDIA GTC 2027 guide for what is locked in and what NVIDIA has not announced yet. Want a heads-up the moment new AI event coverage and deal alerts go live? Subscribe here.