Nvidia (NVDA) CEO Jensen Huang took to the stage in San Jose, Californi🐬a on Tuesday to kick off the AI giant's weeklong GTC 2025 conferenc♍e.
Huang's two-hour presentation emphasized the breadth of AI applications supported by Nvidia systems. Huang elaborated on Nvidia's contributions to the development of autonomous vehicles, better wireless networks, and advanced robotics. He also laid out the company's product roadmap for the next two years. Readers can watch the full livestream below.
The event came with Nvidia shares down more than 10% so far this year. The stock has come under pressure from concerns about 澳洲幸运5官方开奖结果体彩网:overspending on AI infrastructure, 澳洲幸运5官方开奖结果体彩网:economic uncertainty related to President Trump’s unpredictable tariff policies, and worries that restrictions on chip exports could be tightened to hamstring 澳洲幸运5官方开奖结果体彩网:AI’s development in China.
Nvidia shares tr𓃲aded in the red tﷺhroughout the event and finished the day 3.4% lower.
Replay: Watch Huang's Keynote Here
Watch Huang's keynote in full with the video player below.
Huang Wraps Up Keynote With Help of Star Wars-Style🌟 Robot
Huang ended his GTC 2025 keynote address talking about Nvidia's robotics infrastructure.
The presentation was capped off by the announcement of a joint venture with Google's DeepMind and Disney Research.
A short video animated in the style of Disney's Pixar introduced a bipedal robot, similar to those that accompany humans in the Disney-owned Star Wars franchise, wandering through sand dunes. A physical prototype of the robot, which Huang called Blue, was then brought onto the stage, where it followed his directions and reacted to his questions with gestures and noises.
Nvidia Annꦏounces Next-G🃏eneration Systems Blackwell Ultra and Vera Rubin
As expected, Nvidia announced its Blackwell Ultra chip, which the chipmaker s🃏aid 🀅can offer 1.5 times the AI performance of earlier Blackwell chips.
Blackwell Ultra is set to come in the second half of the year, Huang said, consistent with Nvidia's plans to roll out a new product line every year. Its Vera Rubin, the next generation, is expected to ship in 2026.
Ama𝄹zon’s AWS, Google Cloud, Microsoft’s Azure and Oracle will be among the first cloud service providers to offer services powered by Blackwell Ultra, the company said.
Server makers Cisco, Dell Technologies, Hewlett Packard En💎terprise, Lenovo and Supermicro will also be among those expected to deliver a wide range of servers based on Blackwell Ultra.
Nvidi💃a ⭕Says Its Dynamo Software Can Boost Efficiency and Lower Costs
During the ev𝕴ent, Huang announced Nvidia’s Dynamo, an open-soꦆurce software that the company said can raise efficiency and lower costs for AI inference.
Th𝔉e company said that Dynamo can even double performance serving Meta’s Llama models on Nvidia’s Hopper platform, Blackwell’s predecessor.
“It will enable users to accelerate the adoption of AI inference, including at AWS, Cohere, CoreWeave, Dell, Fireworks, Google Cloud, Lambda, Meta, Microsoft Azure, Nebius, NetApp, OCI, Perplexity, Together AI and VAST,” the companꦆy said in a release.
The AI industry has become increasingly i💟nterested in efficiency since Chinese start-up DeepSeek released its super-efficient R1 reasoning model earlier this year.
Huang Lays Out AI System Roadmap
We’re one year out from Nvidia's follow-up to the immensely powerful Blackwell system, said Huang.
Vera Rubin is Nvidia’s next-generation system, and it will contain 144 individual Nvidia GPUs. Rubin was an astrophysicist who discovered dark matter. Th🙈e system is expected to roll out in the second half of 2026.
Huang said the company would release the Rubin Ultra, containing 576 GPUs, in 🔯the second half of 2027. The system, he said, will dramatically increase AI computing power and efficiency.
Bl✨ackwell 40x More Powerful Than Hopper, Huang Says
“I’m the Chief R♑evenue Destroyer,” Jensen Huang qu𝔉ipped after noting that Nvidia’s new Blackwell system was 40 times more powerful than its predecessor Hopper system.
Huang joked there were few instances in which he'd recommend buying a Hopper rather than Blackwell system.
Nvidia Shares Slide During Huang's Address
Nvidia shares were down about 3% about an hour into Jensen Huang's keynote address.
The stock remained off its session lows from early in the morning but has fallen relatively steadily since Huang's presentations began.
Nvidia to Partner With GM on Self-Driving Cars
Jensen Huang on Tuesday announced Nvidia would partner with GM (GM) to develop and build the automaker's future fleet of s𝓀elf-dri𝄹ving cars.
The announcement focused heavily on autonomous vehicle safety, with Huang announcing Nvidia HALO, "a comprehensive safety system."
The announcement was followed by a video demonstration of autonomous driving tests explaining how Nvidia's chips and software are enabling self-driving cars.
GM shares rebounded from their session lows foll🌞owing the announcement♔ but remained slightly lower for the day.
Huang Predicts Data Center Spending To Soar
Huang said data center buildouts are at an “i🐼nflection point,” expecting data center capital expenditures could exceed $1 trillion by 2028.
Huang pointed to sales of 3.6 million Blackwell GPUs so far this year to the big four cloud platforms. The chipmaker counts Microsoft (MSFT), Meta (META), Amazon (AMZN), and Google-parent Alphabet (GOOGL) among its m🍃ajor clients, with all of🉐 them announcing plans to boost spending on AI in recent earnings calls.
Last month, Wedbush analysts said about 70% of Nvidia customers they spoke with said they are ꧂raising their budgets for AI spending.
Huang Points to Surging GPU Demand from CSPs
Huang said AI computing is at an inflection point, pointing to a surging demand for Nvidia graphics pr🍰ocessing units.
Nvidia shipped 1.3 million Hopper GPUs to America’s four largest cloud service providers—Microsoft (MSFT), Alphabet (GOOG), Amazon (AMZN), and Meta (META)—in 2ꦑ024. In 2025, it has already shipped 3.6 million Blackwell GPUs.
Correction: A prior version of this entry understated the number of Blackwell GPUs shipped so far this year.
Huang Maps Out the Phases of AI
Huang open🎐ed his keynote with a timeline of AI’🐠s development.
“AI really came into the world’s consciousness 10 years ago,” Huang said. The first iteration Huang called Perception 𝔉AI, and it was characterized by speech recognition and other simple processes.
澳洲幸运5官方开奖结果体彩网:Generative AI followed, allowing AI 𝐆to produce text and images through predictive patterns.
Now, he says, we’re in the middle of the rollout of 澳洲幸运5官方开奖结果体彩网:Agentic AI, artificial intelligence that can interact with the digital world, performing tasks onܫ a user’s behalf.
Next is Physical AI, the intelligence that’s expected to power humanoid rꦏobots.
Analysts Expect GTC To Be Positive Catalyst fꩲor the AI Trade
Nvidia shares were 2% lower in intraday trading Tuesday ahead 🐠of the eve෴nt some Wall Street analysts are calling the “AI Woodstock.”
Wedbush analysts suggested the event could be a ”wake-up moment for the tech bulls” after a 澳洲幸运5官方开奖结果体彩网:tough start to the year, with Nvidia shares down about 13% for 2025 amid a broader tech rout as concerns about AI spending and the potential impact of pol✅icies on tariffs and export restrictions weighed on sentiment.
Wedbush, along with Bank of America, UBS, and others have suggested the stock’s recent losses could be an opportunity to 澳洲幸运5官方开奖结果体彩网:buy the dip, pointing to strong demand for AI hardware. The consensus price target of an꧑alysts tracked by Visible Alpha at $177 would suggest roughly 50% upside from Nvidia’s level ahead of the eജvent.
What To Expect From Jensen Huang's GTC 2025 Keynote
Investors and analysts will be watching for updates on the company’s latest 澳洲幸运5官方开奖结果体彩网:artificial intelligence chips, upcoming releases, and developments in gaming and rob☂otics.
The AI chipmaker is expected to showcase its 澳洲幸运5官方开奖结果体彩网:Blackwell Ultra GB300 family of c♍hips, which Deutsche Bank analysts said is expected to deliver over 50% more memory capacity and significantly higher performance than its earlier Blackwell offerings. The timing of GB300’s rollout will be a focus, 💝the analysts said, particularly as Nvidia has faced delays in fully ramping up Blackwell production.
Nvidia could also offer more details on its Rubin GPU, the successor to Blackwell expected in 2026, along with its associated Vera CPU, and the Rubin Vera platfo💃rm. It’s possible Huang’s keynote could offer breadcrumbs at what lies a generation beyond Rubin, analysts said.