Nvidia took San Jose by storm this year, with a record-breaking 25,000 attendees flocking to the San Jose Convention Center and surrounding downtown buildings. Many workshops, talks, and panels were so packed that people had to lean against walls or sit on the floor \u2014 and suffer the wrath of organizers shouting commands to get them to line up properly.<\/p>\n
Nvidia currently sits at the top of the AI world, with record-breaking financials<\/a>, sky-high profit margins, and no serious competitors yet. But the coming months also hold unprecedented risk for the company as it faces U.S. tariffs, DeepSeek<\/a>, and shifting priorities from top AI customers.\u00a0<\/p>\n At GTC 2025<\/a>, Nvidia CEO Jensen Huang attempted to project confidence, unveiling powerful new chips<\/a>, personal \u201csupercomputers,\u201d<\/a> and, of course, really cute robots.<\/a> It was an exhaustive sales pitch \u2013 one aimed at investors reeling from Nvidia\u2019s nosediving stock<\/a>.<\/p>\n \u201cThe more you buy, the more you save,\u201d Huang said at one point during a keynote on Tuesday. \u201cIt\u2019s even better than that. Now, the more you buy, the more you make.\u201d<\/p>\n More than anything, Nvidia at this year\u2019s GTC sought to assure attendees \u2013 and the rest of the world watching \u2013 that demand for its chips won\u2019t slow down anytime soon.\u00a0<\/p>\n During his keynote, Huang claimed that nearly the \u201centire world got it wrong\u201d<\/a> on traditional AI scaling falling out of vogue<\/a>. Chinese AI lab DeepSeek, which earlier this year released a highly efficient \u201creasoning\u201d model<\/a> called R1<\/a>, prompted fears among investors that Nvidia\u2019s monster chips may no longer be necessary for training competitive AI.\u00a0<\/p>\n But Huang has repeatedly insisted that power-hungry reasoning models will, in fact, drive more demand for the company\u2019s chips, not less. That\u2019s why at GTC, Huang showed off<\/a> Nvidia\u2019s next line of Vera Rubin GPUs, claiming they\u2019ll perform inference (that is, run AI models) at roughly double the rate of Nvidia\u2019s current best Blackwell chip.<\/p>\n The threat to Nvidia\u2019s business Huang spent less time addressing was upstarts like Cerebras, Groq, and other low-cost inference hardware and cloud providers. Nearly every hyperscaler is developing a custom chip for inference, if not training, as well. AWS has Graviton and Inferentia (which it\u2019s reportedly aggressively discounting<\/a>), Google has TPUs, and Microsoft has Cobalt 100<\/a>.<\/p>\n Along the same vein, tech giants currently extremely reliant on Nvidia chips, including OpenAI and Meta, are looking to reduce those ties via in-house<\/a> hardware<\/a> efforts. If they \u2013 and the aforementioned other rivals \u2013 are successful, it\u2019ll almost assuredly weaken Nvidia\u2019s stranglehold on the AI chips market.<\/p>\n That\u2019s perhaps why Nvidia\u2019s share price dipped<\/a> around 4% following Huang\u2019s keynote. Investors might\u2019ve been holding out hope for \u201cone last thing\u201d \u2014 or perhaps an accelerated launch window. In the end, they got neither.<\/p>\n Nvidia also sought to allay worries about tariffs at GTC 2025. <\/p>\n The U.S. hasn\u2019t imposed any tariffs on Taiwan (where Nvidia gets most of its chips), and Huang claimed<\/a> tariffs wouldn\u2019t do \u201csignificant damage\u201d in the short run. He stopped short of promising that Nvidia would be shielded from the long-term economic impacts, however \u2014 whatever form they ultimately take.<\/p>\n Nvidia has clearly received the Trump Administration\u2019s \u201cAmerica First\u201d message, with Huang pledging at GTC<\/a> to spend hundreds of billions of dollars on manufacturing in the U.S. While that would help the company diversify its supply chains, it\u2019s also a massive cost for Nvidia, whose multitrillion-dollar valuation depends on healthy profit margins.<\/p>\n As it looks to seed and grow businesses other than its core chips line, Nvidia at GTC drew attention to its new investments in quantum, an industry that the company has historically neglected. At GTC\u2019s first Quantum Day, Huang apologized<\/a> to the CEOs of major quantum companies for causing a minor stock crash in January 2025 after he suggested that the tech wouldn\u2019t be very useful for the next 15 to 30 years.<\/p>\n On Tuesday, Nvidia announced<\/a> that it would open a new center in Boston, NVAQC, to advance quantum computing in collaboration with \u201cleading\u201d hardware and software markers. The center will, of course, be equipped with Nvidia chips, which the company says will enable researchers to simulate quantum systems and the models necessary for quantum error correction.<\/p>\n In the more immediate future, Nvidia sees what it\u2019s calling \u201cpersonal AI supercomputers\u201d<\/a> as a potential new revenue-maker.\u00a0<\/p>\n At GTC, the company launched<\/a> DGX Spark (previously called Project Digits<\/a>) and DGX Station, both of which are designed to allow users to prototype, fine-tune, and run AI models in a range of sizes at the edge. Neither is exactly inexpensive \u2013 they retail for thousands of dollars \u2013 but Huang boldly proclaimed that they represent the future of the personal PC.\u00a0\u00a0<\/p>\n \u201cThis is the computer of the age of AI,\u201d Huang said during his keynote. \u201cThis is what computers should look like, and this is what computers will run in the future.\u201d<\/p>\n We\u2019ll soon see if customers agree.<\/p>\n<\/div>\nInference boom<\/h2>\n
![\"\"](\"https:\/\/techcrunch.com\/wp-content\/uploads\/2025\/03\/GettyImages-2205761750.jpg?w=680\")
Tariff tensions<\/h2>\n
New business<\/h2>\n
![\"\"](\"https:\/\/techcrunch.com\/wp-content\/uploads\/2025\/03\/GettyImages-2205217071.jpg?w=680\")