In early 2025, NVIDIA lost nearly $600 billion in market value after China’s emerging generative AI large-scale language model tried to prove that it didn’t need to spend billions to develop such a platform.
Now, after NVIDIA became the first company to cross the $5 trillion market capitalization threshold on Friday, Chinese platform DeepSeek marked the occasion by introducing another model that is decidedly powerful and pocket-friendly. In theory, this could increase the cap tank again.
DeepSeek is bursting onto the scene with a highly affordable generative AI service that aims to challenge the dominance of American AI technology. On the same day that Nvidia closed on a $5.06 trillion cap, the company launched V4-Pro, outselling second-place Alphabet by nearly $1 trillion.
V4-Pro uses a 1.6 trillion parameter AI model that is approximately 50 times cheaper to run than Anthropic’s Claude Opus using an application programming interface.
It’s also worth noting that it was trained on Chinese chips, specifically those from Huawei Technologies, rather than Nvidia chips. DeepSeek’s initial V3 and R1 LLMs were trained on chips from the world’s most valuable publicly traded companies.
According to the feature, V4-Pro only activates 49 billion parameters per token (despite a total of 1.6 trillion). This means “frontier-level output at a computational cost of 37 billion models,” said Christian Schmidt, chief commercial and revenue officer at Samsung Mena.
V4-Pro costs just $3.48 for an output of 1 million tokens, while Anthropic costs $30 and OpenAI costs $25 for the same workload. This is already rock bottom, but lower-tier V4-Flash is currently ridiculously cheap at $0.28.
“The numbers are real…the cost of frontier AI has just come down again,” Schmidt said. “The chip monopoly that the US relied on to slow China is no longer guaranteed. Open source now competes at the highest level with closed source.”
familiar playbook
California-based Nvidia suffered nearly $590 billion in losses in the aftermath of the DeepSeek R1 launch. DeepSeek also managed to dethrone ChatGPT from the top of Apple’s App Store at the time.
It’s no surprise that DeepSeek emerged from China to offer such value. The world’s second-largest economy is considered America’s biggest technology rival and has competed with other countries by offering more affordable products such as smartphones and cars.
DeepSeek claims that it spent just $6 million and just two months training its R1 model. By comparison, Meta Platforms spent $60 million on its Llama model, while OpenAI reportedly spent more than $6 billion, and both companies also took longer to develop their models. It is unclear how much was invested in the development of the V4 LLM.
The humble fact here is that Huawei is committed to supporting DeepSeek by providing Ascend 950 chips. Huawei started making more noise about generative AI almost a year ago when it was reported that the company was developing the Ascend 910D.
“That Huawei product is lost in most of the press coverage. So here we have a frontier-adjacent model trained and delivered on non-NVIDIA silicon, and offered at a lower price than what closed-source can sustain,” said Shwetank Kumar, principal scientist at California-based EnCharge AI.
Nvidia CEO Jensen Huang seemed unhappy, warning that a partnership between DeepSeek and Huawei would be “scary for the United States.” Perhaps he was also upset that Huawei reportedly denied Nvidia access to its new Ascend chips, as did OpenAI.
“The old assumption that you can’t train a frontier model without Nvidia hardware is now empirically incorrect,” said Rishav Ganguli, founder of New Dawn AI in India.
He cited DeepSeek’s admission that the company was three to six months behind GPT-5.4 and Gemini 3.1 Pro, but the 2026 Stanford AI Index said the Chinese AI institute had “effectively closed” the performance gap.
“This is no longer an outdated phenomenon…For the past two years, a lot of strategies have been built on the assumption that a small number of US labs will be at the top of a steep capacity curve and all the remaining labs will pay the rent from there,” Ganguly said. “That assumption is being repriced in real time.”
Kumar said companies are spending tens of billions of dollars on LLM training, making this price point a “strategic tipping point.”
He added that DeepSeek’s pricing is an “intentionally set ceiling lower than what OpenAI and Anthropic would need to charge to cover their own frontier training.” “DeepSeek doesn’t need margin on the token. They need OpenAI and Anthropic, and they can’t raise the price unless they see an extraction effect.”
Is Nvidia still worth it?
As of now, NVIDIA’s market position is still unaffected by DeepSeek’s latest salvo, but it’s worth mentioning that last year’s multi-billion dollar market cap drop took about a week to occur.
Perry Wu, founder and chief executive officer of Darius FinTech, which operates the Darius.AI platform designed to help investors understand changing market dynamics, said the rise in NVIDIA’s stock price is a “trend-level advance.”
Darrius.AI’s analysis showed an “early buy signal” on April 9, when the stock price was $183, and changed it to a “firm buy” when the stock reached $187. The stock has since trended upward, closing at a record high of $208.26 on Friday, “confirming its upward trend.”
On that fateful day, January 27, 2025, NVIDIA’s stock price plummeted, but the stock has since risen nearly 76%. “This is no longer a question of ‘will we get in?’ but a question of the fact that smart money has been in place for some time already,” Wu said.
However, his optimistic assessment comes with a caveat to Nvidia itself, not investors. “NVIDIA must continue to outperform. There is little room for disappointment.”
