Nvidia (NASDAQ: NVDA) CEO Jensen Huang’s meeting with President Donald Trump on Friday capped a whirlwind week for the increasingly life sciences-focused Silicon Valley microprocessing giant, in which its stock slid 16% as investors assessed how much of a competitive threat the company faces from the Chinese-owned artificial intelligence (AI) startup DeepSeek.
“We appreciated the opportunity to meet with President Trump and discuss semiconductors and AI policy,” an Nvidia spokesperson said in a statement to Japan’s NHK and other news outlets. “Jensen and the President discussed the importance of strengthening U.S. technology and AI leadership.”
Trump offered few specifics about his meeting with Huang, but said Friday: “He’s the biggest in the world in terms of chips. I can’t say what’s going to happen [except] we had a meeting. It was a good meeting.”
“Eventually, we’re going to put tariffs on chips” as well as oil and gas, he continued: “That’ll happen fairly soon, I think around the 18th of February.”
DeepSeek rattled Wall Street late last month by announcing it had gone live with its open source R1 model, one that it said rivaled the Model o1 of privately held OpenAI, the developer of ChatGPT. What especially shook investors was the prospect that China may have closed the tech gap with the United States in AI and especially DeepSeek’s claim that it developed an earlier model it launched in December, V3, for just $5.58 million—a far cry from the tens of billions and hundreds of billions of dollars that U.S. AI tech giants have committed toward expanding the technology.
However, Palmer Luckey, co-founder of defense tech company Anduril Industries and co-founder of virtual reality company Oculus (acquired by Meta predecessor Facebook in 2017), dismissed that cost estimate as CCP [Chinese Communist Party] propaganda, telling Fox Business that much of DeepSeek’s infrastructure costs remain unknown because the company’s widely-quoted cost figure did not include full costs of V3 and R1, whose costs have not been disclosed.
Analyst: DeepSeek lowballing expenses
At least one analyst agreed that DeepSeek was publicly lowballing its expenses: “Did DeepSeek really ‘build OpenAI for $5M?’ Of course not,” Bernstein analyst Stacy Rasgon wrote in a research note as reported by Dow Jones, contending that the $5.58 million figure “does not include all the other costs associated with prior research and experiments on architectures, algorithms, or data,” though he added that the model was designed “to significantly reduce the cost to train and run, given that only a portion of the parameter set is active at any one time.”
V3 is a 671-billion parameter, reinforcement learning-based general-purpose large language model with an emphasis on scale and efficiency, with key strengths that include excelling in problem-solving, mathematics, and logical reasoning, according to the Indian tech education platform GeeksforGeeks. R1, by contrast, is an advanced AI model designed for high-speed processing, logical thinking, self-verification, and accurate content generation across various applications, with applications that include 3ducational tools, research applications, and AI-driven reasoning tasks.
“The main difference lies in their processing capabilities,” GeeksforGeeks contributor Shubham Kumar explained. “DeepSeek R1 is designed for speed and efficiency, making it ideal for tasks that require quick responses like content creation and coding, while DeepSeek V3 excels in complex reasoning, handling multi-domain tasks, and offering a deeper understanding of language and context.”
Trump and several U.S. tech giants late last month sought to reinforce U.S. leadership in AI when they announced the up-to-$500 billion Stargate AI infrastructure joint venture—an effort promoted in part on its ability to help achieve the long-sought goal of developing a cancer vaccine.
Investors quickly interpreted DeepSeek’s success as posing a competitive threat to Nvidia since it is the leading maker of graphics processing units (GPUs) and other hardware used in AI. That fear sent Nvidia’s shares tumbling 17% on January 27, from $142.62 to $118.42. The drop wiped out about $600 billion in Nvidia’s market capitalization (the product of the share price and the number of outstanding shares), the largest ever one-day market-cap decline for a public company, briefly bringing the company back under $3 trillion in market cap.
Shares recovered Tuesday, bouncing back 9% to $128.99, and restoring the $3 trillion market cap. But on Wednesday, Nvidia shares resumed their slide, falling 4% to $123.70 after Bloomberg News reported that the Trump administration was considering further curbs on exports of the company’s GPUs to Chinese companies, citing unnamed sources.
Export restrictions
Starting in 2022, the administration of Trump’s predecessor Joe Biden began restricting the export to China of advanced AI GPUs, among chips deemed critical because of their potential military use.
Days before leaving office, Biden drew criticism from Nvidia and the Semiconductor Industry Association for further restrictions set to take effect May 15 unless Trump’s administration changes or withdraws them. Those rules would go beyond a 2023 expansion of the limits from Nvidia’s A100 and H100 GPUs to its A800 and H800 GPUs—reduced capability versions of the original 100 series chips initially created for export to China.
The restrictions prevented DeepSeek from using Nvidia’s most advanced GPUs, which are now the Blackwell GPUs launched last year, instead relegating the Chinese company to using Nvidia GPUs not covered by the original ban.
While DeepSeek said it trained R1 using 2,048 Nvidia H800 GPUs, several market watchers said that claim could not be verified, reasoning that the company may have more GPUs and even more advanced chips than it has disclosed.
“Wake-up call”
Bloomberg reported Friday, citing unnamed sources, that the White House and FBI were investigating whether DeepSeek acquired the H800s from third parties in Singapore to get around the Chinese export ban.
“The release of DeepSeek AI from a Chinese company should be a wake-up call for our industries that we should be laser-focused on competing to win because we have the best scientists in the world. The Chinese leadership told me that,” Trump said on January 27, addressing the House Republican Issues Conference in Miami.
“It could be a positive development. Instead of spending billions, you will spend less, and hopefully come up with the same solution,” Trump added. “We are going to unleash our tech companies and dominate the future like never before.”
DeepSeek’s impact on AI technology “all depends on the veracity of DeepSeek’s claims,” William Stein, managing director/senior analyst-technology with Truist Securities, wrote in a research note on January 27, as reported by Investors Business Daily. “We cannot determine the veracity of DeepSeek’s claims. If they’re true, it magnifies the already-known risk of an AI spending slowdown.”
“Excellent AI advancement”
Whatever the number and type, DeepSeek’s use of Nvidia GPUs may explain why Nvidia has rejected arguments that its dominance of AI hardware is threatened by the competitive challenge posed by DeepSeek-R1.
“DeepSeek is an excellent AI advancement and a perfect example of Test Time Scaling,” Nvidia stated, using the industry term for the practice of AI models performing multiple inference passes over a query, conducting chain-of-thought, consensus, and search methods to generate the best answer, rather than offer a single direct, pre-trained response. “DeepSeek’s work illustrates how new models can be created using that technique, leveraging widely available models and compute that is fully export control compliant.”
“Inference requires significant numbers of Nvidia GPUs and high-performance networking,” the company noted.
On Thursday, Nvidia said it was making DeepSeek R1 available as an NVIDIA NIM or “NVIDIA Inference Microservice.” NIMs are optimized cloud-native “microservices” designed to let developers accelerate deployment of generative AI models anywhere—whether through local workstations, on-premises data centers, cloud services, or GPU-accelerated workstations.
Lessons from DeepSeek
Pat Gelsinger, who in December announced his retirement as Intel’s CEO—reportedly under pressure from board members dissatisfied with the company’s inability to compete with Nvidia—commented on LinkedIn that DeepSeek succeeded by following three lessons of computing industry success:
- Making compute resources broadly available at radically lower price points will drive an explosive expansion, not contraction, of the market.
- Constraints compelled DeepSeek to find creative ways to deliver a “world-class solution in every respect at 10–50X lower costs.”
- Open source wins every time it is given a proper shot: “AI is much too important for our future to allow a closed ecosystem to ever emerge as the one and only in this space.”
“The market reaction is wrong, lowering the cost of AI will expand the market. Today I’m an Nvidia and AI stock buyer and happy to benefit from lower prices,” wrote Gelsinger, who told TechCrunch his startup Gloo, a messaging platform for churches, will use DeepSeek rather than OpenAI technology. “DeepSeek is an incredible piece of engineering that will usher in greater adoption of AI. It will help reset the industry in its view of open innovation. It took a highly constrained team from China to remind us all of these fundamental lessons of computing history.”
Morningstar senior equity analyst Dan Romanoff observed January 27 in a commentary that DeepSeek’s “materially lower” training costs “raised investor concerns about the necessity of the billions of dollars in capital expenditures that large U.S. tech companies have made (and the billions more they plan to spend) on generative AI.”
Romanoff’s colleague Brian Colello, an equity strategist with Morningstar, opined in the same commentary he didn’t foresee a halt to the buildout of AI systems now in progress: “Despite DeepSeek’s promise, we doubt the leading cloud vendors and AI builders will pause their plans, although it’s a risk that certainly bears watching.”
“We believe AI GPU demand still exceeds supply, so while slimmer models may enable greater development for the same number of chips,” Colello added, “we still think tech firms will continue to buy all the GPUs they can as part of this AI ‘gold rush’.”
Leaders and laggards
- Akero Therapeutics (NASDAQ: AKRO) shares nearly doubled, rocketing 97.5% from $26.18 to $51.71 Monday after the company announced positive preliminary topline week 96 results from its Phase IIb SYMMETRY trial assessing lead product candidate efruxifermin (EFX) in biopsy-confirmed compensated cirrhosis (F4), Child-Pugh Class A, due to metabolic dysfunction-associated steatohepatitis (MASH). Among the 134 patients with baseline and week 96 biopsies, 39% of the 46 patients treated with 50 mg EFX showed reversal of cirrhosis with no worsening of MASH, vs. 15% of the 47 placebo patients. In the 181-patient Intent to Treat (ITT) population, with all missing week 96 biopsies treated as failures, 29% of the 63 patients in the 50 mg EFX group showed a reversal of cirrhosis with no worsening of MASH, vs. approximately 12% of the 61 placebo patients.
- Metsera (NASDAQ: MTSR) shares soared 47% from its initial public offering (IPO) price of $18 to $26.50 Friday, the company’s first day of trading public shares following an IPO of 15,277,778 shares that generated $275 million in gross proceeds. Metsera is developing MET-097i, a once-monthly injectable GLP-1 receptor agonist. Net proceeds would be approximately $250.8 million, rising to approximately $289.1 million if underwriters exercise their option to purchase all of an additional 2,291,666 shares at the IPO price less underwriting discounts and commissions. Last month, Metsera announced positive topline Phase IIa data for MET-0971i in obese and overweight patients showing up to 11.3% mean placebo-adjusted weight loss at 12 weeks.
- Tectonic Therapeutic (NASDAQ: TECX) shares roller-coastered this past week, plummeting 41% from $41.18 to $24.44 Tuesday after Eli Lilly (NYSE: LLY) disclosed on ClinicalTrials.gov that it terminated a Phase II trial (NCT06598631) of its chronic kidney disease candidate volenrelaxin, a relaxin analog—a modality with similarities to Tectonic’s lead candidate, the Fc-relaxin fusion protein TX45. “Study was terminated due to a lack of foreseeable clinical benefit in the proposed chronic kidney disease population, following the termination of a related heart failure study that demonstrated no benefit in an overlapping patient group,” Lilly wrote. But on Thursday, Tectonic shares more than doubled, rocketing 113% from $25.72 to $54.84 after the company announced positive interim data from a Phase Ib open-label acute hemodynamic trial showing that a single intravenous dose of TX45 resulted in meaningful improvements in both left ventricular function and pulmonary hemodynamics in patients with Group 2 Pulmonary Hypertension in Heart Failure with Preserved Ejection Fraction (PH-HFpEF). Patients treated with TX45 achieved 17.9% reduction in Pulmonary Capillary Wedge Pressure (PCWP) in the total study population of PH-HFpEF and >30% reduction in Pulmonary Vascular Resistance (PVR) in combined pre- and post-capillary PH (CpcPH), a subpopulation with more severe disease.
- 89bio (NASDAQ: ETNB) shares climbed 26% from $7.13 to an even $9 Monday as investors embraced companies whose pipelines feature FGF21 analogs—a category that includes 89bio’s metabolic dysfunction-associated steatohepatitis (MASH) candidate pegozafermin. The buying surge for 89bio was triggered by Akero Therapeutics reporting positive preliminary topline week 96 results for its FGF21 analog for a form of MASH, lead candidate efruxifermin (see above). On January 13, Akero reported that its Phase III ENLIGHTEN program consisting of separate trials in patients with non-cirrhotic F2-F3 MASH (ENLIGHTEN-Fibrosis, NCT06318169) and compensated cirrhotic F4 MASH (ENLIGHTEN-Cirrhosis, NCT06419374) continued to enroll patients across both trials.
