Tech
DeepSeek And Nvidia Logos
(VCG/Getty Images)

The trillion-dollar mystery surrounding DeepSeek’s Nvidia GPUs

There’s a cloud of suspicion hanging over the type and number of Nvidia GPUs DeepSeek used to train its R1 models.

At the center of the story of DeepSeek’s breakthrough achievement with its R1 models lies the Nvidia hardware that powered the servers that trained those models.

In December 2024, DeepSeek researchers released a paper that outlined the development and capabilities of the new DeepSeek-V3 large language model. In the paper, the researchers said they were able to train their powerful, efficient model over 2.78 million GPU hours of computing time on a cluster of only 2,048 Nvidia H800 GPUs. That is a very small number of GPUs for a model that matched or beat OpenAI’s state-of-the-art o1 model in some benchmarks.

For comparison, Meta trained its Llama 3.1 models on two clusters, using a total of 39.3 million GPU hours with 49,152 Nvidia H100 GPUs. Last week, Mark Zuckerberg said that Meta is planning on ending 2025 with over 1.3 million GPUs.

Released in 2023, the H800 is a GPU thats similar to the H100 but is tailored for the Chinese market to comply with US export controls concerning national security parameters that the Biden administration rolled out in 2022. Reuters reported that the main thing Nvidia changed in the H800 was that it “reduced the chip-to-chip data transfer rate to about half the rate.”

But The Wall Street Journal reports that government officials found the H800 exploited technical loopholes that met the strict requirements of the ban, but still gave Chinese buyers very powerful AI chips. To close the loophole, in October 2023, the US government banned the export of H800s as well.

It appears that DeepSeek was able to acquire its H800s during that short window of availability.

DeepSeek’s claims are drawing suspicion from some observers in the AI industry, but most appear to be just speculation. Scale AI CEO Alexandr Wang told CNBC that he suspected DeepSeek has “about 50,000 H100s, which they can’t talk about obviously because it is against the export controls that the United States has put in place,” and in a tweet, Elon Musk replied, “Obviously.” Musk, meanwhile, has bragged about xAI’s “Colossus supercluster,” which is powered by 100,000 H100 GPUs, and that he plans to scale up to 1 million of the expensive Nvidia chips.

There have been reports of H100s being smuggled into China through a series of intermediaries on the black market, but no evidence that DeepSeek did so.

Adding to the confusion, DeepSeek cofounder Liang Wenfeng said that the company does own a cluster of 10,000 Nvidia A100 GPUs, a cheaper and less powerful AI chip.

The H100 has earned a status of being one of the most coveted pieces of computer hardware in the AI age. Even when other chips are used, the power is sometimes expressed as a number of “H100-equivalent” GPUs.

Nvidia is in the process of rolling out its next-gen H200 Blackwell GPUs, and last year CEO Jensen Huang hand-delivered the first DGX H200 server to OpenAI headquarters.

More Tech

See all Tech
tech

Amazon AWS and Google set up rare partnership as rivals try to prevent massive network outages

In October, Amazon AWS suffered a major outage, disrupting services for Snapchat, Reddit, Roblox, Venmo, and United Airlines, among others.

As more and more businesses move their cloud computing to AWS or Google Cloud, the risks of such large outages increase.

To address this, an unusual partnership between rivals AWS and Google seeks to let customers quickly and easily move their data and services between the platforms.

In a press release, Robert Kennedy, VP of network services at AWS, said:

“This collaboration between AWS and Google Cloud represents a fundamental shift in multicloud connectivity. By defining and publishing a standard that removes the complexity of any physical components for customers, with high availability and security fused into that standard, customers no longer need to worry about any heavy lifting to create their desired connectivity. When they need multicloud connectivity, it’s ready to activate in minutes with a simple point and click.”

To address this, an unusual partnership between rivals AWS and Google seeks to let customers quickly and easily move their data and services between the platforms.

In a press release, Robert Kennedy, VP of network services at AWS, said:

“This collaboration between AWS and Google Cloud represents a fundamental shift in multicloud connectivity. By defining and publishing a standard that removes the complexity of any physical components for customers, with high availability and security fused into that standard, customers no longer need to worry about any heavy lifting to create their desired connectivity. When they need multicloud connectivity, it’s ready to activate in minutes with a simple point and click.”

tech

Tesla’s Chinese-made EV sales jumped 10% in November

Tesla sales from its Shanghai plant jumped 10% in November, only the third time those sales have risen in a month this year, Bloomberg reports. These figures don’t break down what share of sales are within China — Tesla’s second-biggest market — or exported, but typically most made there are sold locally.

The news is a rare bright spot these days for the company’s car sales, which are continuing to decline in much of the world amid fierce competition, especially from Chinese automakers, against Tesla’s aging lineup. The FactSet analyst consensus expects Tesla’s global sales to drop 7% in 2025, the second annual decline in a row.

tech

Apple retires AI head to show it’s serious about AI

Yesterday, Apple announced that its AI head, John Giannandrea, who presided over the company’s largely subpar endeavors into generative AI, would retire at the relatively young age of 60.

Giannandrea is set to leave the company this spring — right around the time Apple’s long-delayed, AI-powered Siri is expected to debut, with Microsoft’s corporate VP for AI, Amar Subramanya, taking the reins.

The move signals that Apple, which has invested far less in AI than its Big Tech peers — and has comparatively little to show for it — is finally getting serious about the technology.

“We believe that Subramanya represents the right hire at the right time with the clock ticking on Apple’s AI strategy heading into next year with outside hires a necessary move to improve the AI strategy,” writes Wedbush analyst Dan Ives. “We believe that this was a major reset while expecting more outside hires from Cook & Co. to get Apple on the right track when it comes to AI while further preparing the company for its AI Siri launch by mid 2026 which has been delayed due to development challenges and now represents a major turning point in the company’s history.”

The company put it this way: “This moment marks an exciting new chapter as Apple strengthens its commitment to shaping the future of AI for users everywhere.”

The move signals that Apple, which has invested far less in AI than its Big Tech peers — and has comparatively little to show for it — is finally getting serious about the technology.

“We believe that Subramanya represents the right hire at the right time with the clock ticking on Apple’s AI strategy heading into next year with outside hires a necessary move to improve the AI strategy,” writes Wedbush analyst Dan Ives. “We believe that this was a major reset while expecting more outside hires from Cook & Co. to get Apple on the right track when it comes to AI while further preparing the company for its AI Siri launch by mid 2026 which has been delayed due to development challenges and now represents a major turning point in the company’s history.”

The company put it this way: “This moment marks an exciting new chapter as Apple strengthens its commitment to shaping the future of AI for users everywhere.”

tech

OpenAI declares “code red” on Google three years after Google declared “code red” on OpenAI

Google’s recent AI hot streak, which includes the much-lauded release of Gemini 3 last month, has OpenAI worried — so much so that CEO Sam Altman has declared a “code red” to improve its flagship product, ChatGPT, The Wall Street Journal reports.

Scholars of chatbot history might remember that after OpenAI’s ChatGPT burst onto the scene in late 2022, Google declared a “code red” to fight the existential threat the new AI technology posed to Google’s search dominance.

And here we are today!

tech

Tesla sales continued to fall in Europe last month — but there were a few bright spots

Tesla’s self-proclaimed “weakest market” continued to look weak in November, as sales mostly kept dropping year over year across European countries, including Sweden, France, and Spain, Reuters reports.

However, there were a couple of notable bright spots for Tesla, which has suffered due to strong competition and an aging lineup. In Norway, as consumers face the end of certain EV subsidies, Tesla sales jumped 175% in November and about 35% for the first 11 months of 2025. Meanwhile, in Italy, sales increased 58% year on year after declining for six months. From January to November, Tesla sales fell 28% there.

Latest Stories

Sherwood Media, LLC produces fresh and unique perspectives on topical financial news and is a fully owned subsidiary of Robinhood Markets, Inc., and any views expressed here do not necessarily reflect the views of any other Robinhood affiliate, including Robinhood Markets, Inc., Robinhood Financial LLC, Robinhood Securities, LLC, Robinhood Crypto, LLC, or Robinhood Money, LLC.