On the menu today: The sudden arrival of DeepSeek, a new Chinese-built artificial intelligence, sent U.S. tech company stock prices plummeting and is being touted as a “Sputnik moment” for American AI. But today’s newsletter will look at the evidence that this seeming story of triumphant ingenuity may well be a story of triumphant smuggling. In a world where oil, weapons, and all kinds of restricted materials cross borders easily and evade sanctions, just how sure are we that the export controls on advanced chips are really being enforced? And how much can the boasts of an up-until-recently obscure Chinese company be taken at face value?
Questioning DeepSeek’s Extraordinary Claims
On December 15, former Google CEO Eric Schmidt appeared on ABC News’s This Week. He had just published a new book entitled Genesis: Artificial Intelligence, Hope, and the Human Spirit, co-authored with the late Henry Kissinger and Microsoft’s former chief research and strategy officer, Craig Mundie. Schmidt mentioned a then-obscure AI program in the works by Chinese developers:
I used to think we were a couple of years ahead of China, but China has caught up in the last six months in a way that is remarkable. The fact of the matter is a couple of the Chinese programs, one, for example, is called Deep Seek, looks like they’ve caught up, and this is even after these extraordinary efforts by the Trump and Biden administration to withhold high-powered chips.
The Chinese are clever, and they understand the power of a new kind of intelligence for their industrial might, their military might, and their surveillance system.
Fast-forward to today, as the markets realized that the Chinese had more or less caught up to the most advanced AI systems in America, and the stock prices of Nvidia, Google, Microsoft, and AMD took a sharp tumble. In fact, the stock prices for Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla are expected to slide further, and some calculations put the losses at $1 trillion. That’s not a typo, that’s a deliberate “tr.”
Sunday evening, venture capitalist Marc Andreessen declared, “Deepseek R1 is AI’s Sputnik moment.” (You’re a smart readership, but for you younger readers, on October 4, 1957, the Soviet Union launched Sputnik, the world’s first satellite, a key moment in the space race and a wake-up call to Americans that their geopolitical rivals weren’t as far behind them as they had believed.)
Speaking Monday evening at a House Republican members conference meeting, President Trump said, “the release of DeepSeek, AI from a Chinese company, should be a wake-up call for our industries that we need to be laser-focused on competing to win.”
This past week at the World Economic Forum in Davos, Anthropic co-founder and CEO Dario Amodei warned, “I think if the United States can’t lead in this technology, we’re going to be in a very bad place geopolitically.”
Have you used OpenAI’s popular ChatGPT chatbot? (As Charles Pierce of Esquire and Ana Navarro-Cárdenas of ABC’s The View demonstrated, you should double-check anything ChatGPT tells you before using it as a source in your arguments.)
DeepSeek is basically like ChatGPT; you ask the AI questions, and it answers them. But what makes DeepSeek stand out from the rest of the AI pack is how quickly it was built and “trained,” how inexpensive it was compared to other AI models, and how it was built without the most advanced chips, a resource that the U.S. and its allies were trying to keep out of the hands of the Chinese.
The company’s claims are extraordinary . . . emphasis on the word “claims.”
According to the DeepSeek creators, “Our pre-training stage is completed in less than two months and costs 2664K GPU hours. Combined with 119K GPU hours for the context length extension and 5K GPU hours for post-training, DeepSeek-V3 costs only 2.788M GPU hours for its full training. Assuming the rental price of the H800 GPU is $2 per GPU hour, our total training costs amount to only $5.576 million.”
That sounds like a lot of jargon, but the gist is DeepSeek got up and running within two months and for less than $5.5 million. Google and OpenAI needed years and hundreds of millions of dollars to build their AI models, and OpenAI is burning through billions per year:
Early last year, OpenAI raised $10 billion. Just 18 months later, the company had burned through most of that money. So it raised $6.6 billion more and arranged to borrow an additional $4 billion.
But in another 18 months or so from now, OpenAI will need another cash infusion because the San Francisco start-up is spending more than $5.4 billion a year. And by 2029, OpenAI expects to spend $37.5 billion a year. . . .
When companies build A.I. systems, they go big first: They feed these systems enormous amounts of data. The more data companies feed into these systems, the more powerful they become. Just as a student learns more by reading more books, an A.I. system can improve its skills by ingesting larger pools of data. Chatbots like ChatGPT learn their skills by ingesting practically all the English language text on the internet.
That requires larger and larger amounts of computing power from giant data centers. Inside those data centers are computers packed with thousands of specialized computer chips called graphics processing units, or GPUs, which can cost more than $30,000 apiece.
The cost is pushed higher because the chips, data centers and electricity needed to do this digital work are in short supply.
Remember that sudden resurgence in nuclear power, and when big tech companies like Microsoft, Amazon, and Google were signing contracts to either restart decommissioned nuclear reactors or develop small modular nuclear reactors? That’s all driven by the energy needs of data centers fueling the development of artificial intelligence.
Deirdre Bosa of CNBC reports that DeepSeek was supposedly built without the advanced chips that every other AI creator thought was required:
They accomplished all that despite the strict semiconductor restrictions that the U.S. government has imposed on China, which has essentially shackled the amount of computing power. Washington has drawn a hard line against China in the AI race, cutting the country off from receiving America’s most powerful chips like Nvidia’s H-100 GPUs. Those were once thought to be essential to building a competitive AI model, with startups and big tech firms alike scrambling to get their hands on any available. But DeepSeek turned that on its head, side-stepping the rules by using Nvidia’s less performant H-800s to build the latest model and showing that the chip export controls were not the chokehold D.C. intended.
Now, when something seems too good to be true, it probably is, and that does double for any good news coming out of China. (“No human-to-human transmission of Covid-19!”)
Scale AI CEO Alexandr Wang — the youngest self-made billionaire in the world — told CNBC’s Squawk Box on Thursday that DeepSeek probably used tens of thousands of under the table Nvidia chips to create it.
“My understanding is that DeepSeek has about 50,000 H-100s, which they can’t talk about, obviously, because it is against the export controls that the United States has put in place,” Wang said. “And I think it is true that, you know, I think they have more chips than other people expect, but also [on a] going forward basis, they are going to be limited by the chip controls and the export controls that we have in place.”
For what it’s worth, Elon Musk concurs, “obviously.”
Back in July, the Wall Street Journal did an investigative report about the booming AI chip-smuggling business, going through Singapore to China:
A 26-year-old Chinese student in Singapore was packing suitcases last fall to return home for vacation. Besides his clothes and shoes, his luggage included six of Nvidia advanced artificial-intelligence chips.
A connection from college asked him to bring the chips because the U.S. restricted their export to China. Each chip was roughly the size of a Nintendo Switch game console, and the student didn’t flag any suspicions at the airport.
Upon arrival, the student said he was paid $100 for each chip he carried, a fraction of the underground market worth.
The student is part of a barely concealed network of buyers, sellers and couriers bypassing the Biden administration’s restrictions aimed at denying China access to Nvidia’s advanced AI chips, The Wall Street Journal has found. Nvidia’s chips are highly coveted for their ability to handle the massive computations needed to train AI systems that are critical to China-U.S. tech rivalry.
More than 70 distributors are openly advertising online what they purport to be Nvidia’s restricted chips, and the Journal got in direct contact with 25 of them. Many of the verified sellers said they have supplies amounting to dozens of the high-end Nvidia chips each month.
The flow of Nvidia chips is so steady that most of those sellers take preorders and promise delivery in weeks, the Journal found. Some also sold entire servers — costing upward of roughly $300,000 — with each containing eight high-end Nvidia chips.
For what it’s worth, Nvidia insists that the GPUs that DeepSeek used were “fully export compliant.”
That CNBC report from Bosa asked the right questions, but didn’t get many revealing answers:
Who’s behind DeepSeek anyway? Despite its breakthrough, very, very little is known about its lab and its founder, Liang Wenfeng. According to Chinese media reports, Deepseek was born out of a Chinese hedge fund called High Flyer Quant. That manages about $8 billion in assets. The mission, on its developer site, it reads simply: “unravel the mystery of AGI with curiosity. Answer the essential question with long-termism.”
The leading American AI startups, meanwhile — OpenAI and Anthropic — they have detailed charters and constitutions that lay out their principles and their founding missions, like these sections on AI safety and responsibility. Despite several attempts to reach someone at DeepSeek, we never got a response.
Gee, that’s reassuring!
Perhaps it’s my hobby coming to the surface, but I find myself wondering about some of the boasts behind DeepSeek’s announcement.
- How certain are we that it really only cost $5.6 million to develop? Who’s in a position to verify that number? How do we know this really only took two months to get “trained”?
- In a world where nefarious actors can smuggle whole tankers full of oil in defiance of trade sanctions, it’s not that hard to smuggle small flat computer equipment. How do we know that DeepSeek is running on a bunch of lower-rated Nvidia H-800 chips and not some of these advanced H-100s?
- How certain are we that DeepSeek didn’t get some help from the Chinese government along the way? As with most Chinese companies, where does the company stop and the regime start?
Now, I don’t know about you, but when something allegedly amazing comes along for free from a mysterious Chinese company, the last thing I would want to do is download it onto my phone. Alas, not everyone else sees it that way; this week, DeepSeek is the most downloaded phone application on the Apple app store.
And oh, by the way, if you thought TikTok was a security risk on your phone, DeepSeek “collects your IP [address], keystroke patterns, device information and stores it in China, where all that data is vulnerable to arbitrary requisition from the Chinese government.”
Many of our fellow citizens have learned nothing from the experiences of recent years.
*Some guest booker at ABC News had fun, putting Missouri Republican senator Eric Schmitt and former Google CEO Eric Schmidt on the same program.
ADDENDUM: In case you missed it yesterday, the online petition to demand Los Angeles mayor Karen Bass immediately resign isn’t going to do anything. But a recall election could, and the process for that is complicated, but achievable. I would urge critics of Bass to strike while the iron (and embers) are hot, and get the process started.