Where to watch Jensen Huang keynote as Nvidia’s GTC 2026 opens

The world’s most-watched semiconductor executive will outline his next AI roadmap on 17 March.

Sarah Mills | GlobalBeat

A single two-hour presentation next Monday will decide whether Wall Street’s $3 trillion AI darling can keep its halo, as tens of thousands tune in to watch Jensen Huang keynote the chipmaker’s flagship conference in downtown San Jose.

GTC—originally the GPU Technology Conference—has become the hardware sector’s equivalent of Apple’s iPhone launch. Last year Huang unveiled the Blackwell architecture, triggering a record one-day share jump of 16 percent. With rival AMD sharpening its MI-series accelerators and cloud giants designing their own silicon, investors will parse every slide for proof that Nvidia can stay two process nodes ahead.

The stage is bigger—and so are the bets

This year’s gathering sprawls across the new South Hall extension, adding 185,000 square feet of demo floor. Engineers have already trucked in a partial replica of the company’s 15-exaflop Eos supercomputer; liquid-cooled racks hum behind plexiglass for photo ops. Badge pickup opened at noon Sunday; the queue snaked around Santa Clara Street by 12:07.

Analysts say the production values matter. “Huang doesn’t just ship chips, he sells a story,” Stacy Rasgon of Bernstein Research told clients Friday. “If that narrative wavers, the whole AI food chain feels it.” Nvidia has booked satellite viewing lounges in Tokyo, Tel Aviv, and Bangalore for local media, a nod to the 72 percent of last year’s online viewers who tuned in from Asia.

Clockwork leaks: what is already known

Railway billboards south of the venue tease a silhouette stamped “Rubin,” the presumed codename for Nvidia’s post-Blackwell GPU family. Regulatory filings in South Korea last month listed a “R100” part number drawing 1,000 watts, 200 more than today’s top B200.

Three Taiwanese contract manufacturers—Foxconn, Quanta, and Wistron—confirmed they have secured “initial modules” for liquid-cooled chassis, implying volume production by Q4. None would comment on core counts, but one executive paraphrased Huang’s internal mantra: “We don’t launch samples, we launch fleets.”

Cloud titans wait for price lists

Amazon Web Services, Microsoft Azure, and Google Cloud collectively ordered 85 percent of last year’s high-end H100 and B200 shipments, according to third-party estimates by TechInsights. All three have booked closed-door sessions at GTC, seeking clarity on 2027 contract tiers.

Buying cycles now start two generations out. “If Rubin pricing is even five percent above Street projections, CapEx budgets climb into the tens of billions,” a senior cloud procurement manager said, requesting anonymity because negotiations are live. Investors will watch Jensen Huang keynote for any mention of “annual cadence,” code for shorter gaps between refreshes.

Software, not silicon, may steal the show

Nvidia has quietly shifted hundreds of engineers onto its cuAgent middleware, a library designed to let large-language-models spawn mini-programs on the fly. A beta scheduled for release on keynote day would mark the company’s deepest push into so-called agentic AI, moving beyond training chips to controlling how models act.

Competitors argue that tying software to proprietary GPUs risks lock-in. “Nobody wants CUDA to morph into Windows for AI,” said a marketing VP at Intel’s programmable-solutions group. Huang is expected to counter that open-source “reference stacks” will accompany cuAgent, though licensing terms remain under wraps.

Start-ups scramble for seats

Priority passcodes to watch Jensen Huang keynote from the main ballroom now trade for $950 on secondary markets, triple last year. More than 1,400 fledgling firms booked “Inception Pavilion” booths, hoping VC scouts roam the aisles.

One hopeful, Paris-based Dyvo.ai, burned its last cash on a 3x3m carpeted plot. “If we miss this shot, the Series A dies,” founder Lucia Martín admitted while assembling a demo that uses generative models to storyboard ad campaigns. Nvidia claims alumni of its Inception accelerator have raised $50 billion in follow-on funding since 2015; skeptics note the denominator includesldcooked figures from companies that later folded.

Energy fears creep onto the agenda

The numbers tell a different story than last year’s triumphant charts: a single Rubin-class server node could draw 120 kilowatts, enough to power 30 suburban homes. U.S. energy regulator FERC has invited utilities to a side panel titled “Feeding the Beast,” its first GTC appearance.

California’s grid operator CAISO quietly warned data-center developers that Silicon Valley substations will hit summer import limits by 2027. Analysts expect Huang to tout a joint venture with Brookfield Renewable for dedicated geothermal PPA contracts, pre-empting criticism that AI growth collides with climate pledges.

What governments want to hear

Beyond investors, diplomats are angling to watch Jensen Huang keynote for any mention of export controls. The U.S. Commerce Department tightened shipment rules twice last year, capping total processing power for China-bound cards. Nvidia responded with cut-down “C” variants that stay under the ceiling; shipments still fell 14 percent.

European Commission officials have signalled they may open an antitrust probe into CUDA licensing bundles if competitors file formal complaints. Huang could pre-empt scrutiny by announcing a European “AI Factory” anchored in the Netherlands, continuing a playbook that saw a similar €200 million pledge to France in 2025.

Your front-row seat from the sofa

The keynote will simulcast on Nvidia’s website, YouTube, and LinkedIn starting 09:00 Pacific Monday. No registration is required for the basic feed; developers who want to replay sessions must sign up for a free “digital pass” by Sunday night.

For Mandarin speakers, ByteDance’s Volcano引擎 will carry a separately interpreted stream that last year peaked at 1.2 million concurrent viewers. Nvidia’s own app offers real-time captions in 14 languages; the company claims glass-to-glass latency is below two seconds, fast enough for day traders to parse any price-moving nuance.

The after-hours circuit

Huang traditionally exits the stage to a DJ set and neon graphics, then spends four hours glad-handing partners in a private tent. Media will receive embargoed specs under 24-hour lock; product pages go live at 06:00 Tuesday.

Wall Street’s first reactions typically hit sell notes before lunch, while Asian markets open. By sundown California time, derivative desks in Hong Kong will have listed weekly call options struck 10 percent above Nvidia’s Monday close—bets that history repeats when the world clicks to watch Jensen Huang keynote the next leg of the AI build-out.

Next up: Huang takes moderator questions at 13:30 Pacific during a press round-table; breakout sessions on cuAgent follow throughout the week. Nvidia reports fiscal-Q1 earnings on 22 May, giving investors a formal channel check on whatever sizzles on stage.