This is not investment advice. The author has no position in any of the stocks mentioned. Wccftech.com has a disclosure and ethics policy.
With DeepSeek shocking the global tech sector with its breakthrough R1 model, which is able to more than compete with OpenAI's o1 model but with just around 1/50th of the typical costs associated with training a complex LLM, the internet chatter is rife with speculation as to how the Chinese startup was able to achieve such a feat. Now, NVIDIA has jumped into this melee with an endorsement of DeepSeek's efforts.
Recently, the founder and CEO of Scale AI, Alexandr Wang, set the proverbial tongues wagging when he declared in an interview that DeepSeek "has about fifty thousand H100s." NVIDIA's H100 GPUs cannot be sold to China under US law, and Wang's statement drew mixed reactions with some outright rejecting his claims.
Obviously
— Elon Musk (@elonmusk) January 27, 2025
Then, Elon Musk pitched in by endorsing Wang's statement in relation to DeepSeek's alleged H100 stash.
This then quickly morphed into a speculative frenzy, with netizens pointing to Singapore as one possible vector that could have allowed DeepSeek to procure NVIDIA's H100 chips.
Over 1/3rd of Nvidia sales go to China, probably ~$40B last year.
The Singapore backdoor is real: Nvidia even says “shipments to Singapore are insignificant” while 22% of billings last qtr were to Singapore. pic.twitter.com/AGeArCgTI1
— Sheel Mohnot (@pitdesi) January 25, 2025
Some pointed to the fact that Singapore accounted for 22 percent of NVIDIA's billings last quarter, but actual shipments were "insignificant," as per NVIDIA's filings with the SEC.
To clarify this matter, we approached NVIDIA on Sunday for a statement. The GPU giant has now offered its take on this subject:
“DeepSeek is an excellent AI advancement and a perfect example of Test Time Scaling. 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. We now have three scaling laws: pre-training and post-training, which continue, and new test-time scaling.”
NVIDIA believes that DeepSeek was able to achieve its breakthrough in a manner that was fully compliant with US export controls, seemingly rubbishing the rampant speculation about the startup's sizable GPU stash.
Meanwhile, NVIDIA has taken quite a beating today on fears that DeepSeek's breakthrough would reduce the need for hyperscalers, affecting the GPU giant's order book in the process. Its stock has lost 17 percent during the day's trading at the time of publishing, pushing it down to the third place in the list of the world's most valuable companies.
