Nvidia took San Jose by storm this year, with a sign up-fractureing 25,000 joinees flocking to the San Jose Convention Cgo in and surrounding downtown originateings. Many toilshops, talks, and panels were so packed that people had to lean aobtainst walls or sit on the floor — and suffer the wrath of systematizers shouting orders to get them to line up properly.
Nvidia currently sits at the top of the AI world, with sign up-fractureing financials, sky-high profit margins, and no grave competitors yet. But the coming months also helderly unpretreatnted danger for the company as it faces U.S. tariffs, DeepSeek, and shifting priorities from top AI customers.
At GTC 2025, Nvidia CEO Jensen Huang tryed to project confidence, unveiling strong novel chips, personal “supercomputers,” and, of course, repartner cute robots. It was an exhaustive sales pitch – one aimed at spendors reeling from Nvidia’s nosediving stock.
“The more you buy, the more you save,” Huang shelp at one point during a keyremark on Tuesday. “It’s even better than that. Now, the more you buy, the more you originate.”
Inference boom
More than anyskinnyg, Nvidia at this year’s GTC sought to promise joinees – and the rest of the world watching – that insist for its chips won’t sluggish down anytime soon.
During his keyremark, Huang claimed that csurrfinisherly the “entire world got it wrong” on traditional AI scaling descfinishing out of vogue. Chinese AI lab DeepSeek, which earlier this year freed a highly efficient “reasoning” model called R1, prompted stresss among spendors that Nvidia’s monster chips may no lengthyer be essential for training competitive AI.
But Huang has repeatedly insisted that power-hungry reasoning models will, in fact, drive more insist for the company’s chips, not less. That’s why at GTC, Huang showed off Nvidia’s next line of Vera Rubin GPUs, claiming they’ll carry out inference (that is, run AI models) at rawly double the rate of Nvidia’s current best Bdeficiencywell chip.
The danger to Nvidia’s business Huang spent less time insertressing was upbegins enjoy Cerebras, Groq, and other low-cost inference challengingware and cboisterous supplyrs. Ntimely every hyperscaler is enbiging a custom chip for inference, if not training, as well. AWS has Graviton and Inferentia (which it’s alertedly opposingly discounting), Google has TPUs, and Microgentle has Cobalt 100.
Alengthy the same vein, tech enormouss currently excessively reliant on Nvidia chips, including OpenAI and Meta, are seeing to decrease those ties via in-hoinclude challengingware efforts. If they – and the aforerefered other rivals – are accomplished, it’ll almost promisedly frailen Nvidia’s stranglehelderly on the AI chips taget.
That’s perhaps why Nvidia’s split price dipped around 4% adhereing Huang’s keyremark. Investors might’ve been helderlying out hope for “one last skinnyg” — or perhaps an quickend begin thrivedow. In the end, they got neither.
Tariff tensions
Nvidia also sought to allay worries about tariffs at GTC 2025.
The U.S. hasn’t imposed any tariffs on Taiwan (where Nvidia gets most of its chips), and Huang claimed tariffs wouldn’t do “meaningful injure” in the low run. He stopped low of promising that Nvidia would be shielded from the lengthy-term economic impacts, however — wantipathyver establish they ultimately consent.
Nvidia has evidently getd the Trump Administration’s “America First” message, with Huang pledging at GTC to spend hundreds of billions of dollars on manufacturing in the U.S. While that would help the company diversify its provide chains, it’s also a massive cost for Nvidia, whose multitrillion-dollar valuation depends on fit profit margins.
New business
As it sees to seed and prolong businesses other than its core chips line, Nvidia at GTC drew attention to its novel spendments in quantum, an industry that the company has historicpartner disthink abouted. At GTC’s first Quantum Day, Huang convey remorsed to the CEOs of beginant quantum companies for causing a inmeaningful stock crash in January 2025 after he proposeed that the tech wouldn’t be very beneficial for the next 15 to 30 years.
On Tuesday, Nvidia proclaimd that it would uncover a novel cgo in in Boston, NVAQC, to carry on quantum computing in collaboration with “directing” challengingware and gentleware tagers. The cgo in will, of course, be supplyped with Nvidia chips, which the company says will help researchers to simuprocrastinateed quantum systems and the models essential for quantum error rightion.
In the more instant future, Nvidia sees what it’s calling “personal AI supercomputers” as a potential novel revenue-originater.
At GTC, the company begined DGX Spark (previously called Project Digits) and DGX Station, both of which are summarizeed to apverify includers to prototype, fine-tune, and run AI models in a range of sizes at the edge. Neither is exactly inpricey – they retail for thousands of dollars – but Huang belderlyly proclaimed that they reconshort-term the future of the personal PC.
“This is the computer of the age of AI,” Huang shelp during his keyremark. “This is what computers should see enjoy, and this is what computers will run in the future.”
We’ll soon see if customers consent.
