THOUGHTS OVER MY MORNING COFFEE

Huh. So Sam Altman and his AI bros were lying through their teeth all along.

The Chinese simply changed the math and laid out how the AI crowd are just a bunch of opportunists.

The Emperors have no clothes.

28 January 2025 -- So we received an early Chinese New Year surprise from DeepSeek. The Chinese AI firm launched its reasoning model last week - and analysts belatedly woke up to it.

The firm’s consumer app jumped to #1 in the Apple AppStore and U.S. stock markets, overly indexed on big tech, took a pounding. About $1.2 trillion was been wiped off the U.S. markets yesterday, led of Nvidia getting a hammering. It had a $600 billion one-day loss in its capital - the largest ever drop for one stock in the history of the U.S. stock markets.

What happened? The Chinese artificial-intelligence upstart has trained high-performing AI models cheaply - very, very, very, very cheaply - without the most advanced gear provided by Nvidia and others. 

And it is better than or equal to OpenAI on almost every task. And remember: DeepSeek is open source.

That has pulled the rug from under global companies riding the AI wave, including chip makers, infrastructure suppliers and power stocks, as investors question the outlook for AI spending.

My Chief Technology Officer, Eric De Grasse, went into the fine detail in a post yesterday.

The bottom line? DeepSeek showed you could do the training for pennies, not the astronomical sums Sam Altman and his Robber Barons have been demanding.

The Chinese changed the math. Google, OpenAI, Meta, and Nvidia have all bet on capital spending being the path forward and huge amounts of it. Cash would buy chips. Lots of chips. This was going to provide the moat, the source of advantage. U.S. model makers have been locked into a single paradigm of building ever-larger, more compute-hungry models.

And, after all, the capital markets were willing to fund this outsize spending on GPUs, so why not go for it? 

With China’s venture capital market becoming moribund, local players could not access enough capital. Even those that could, such as the Qwen team from Alibaba or the Doubau from ByteDance, export restrictions hampered access to compute power. So they went internal.

Steven Sinofsky (who has been up to his knees and elbows in the whole world of tech, but especially AI, for ages) put it aptly when he observed that the history of computing is one of innovation followed by a scale-up, eventually disrupted by a “scale-out” approach - when bigger and faster methods are replaced by smaller, more numerous alternatives. As he noted last night on his blog:

China faced an AI situation not unlike Cisco did in its early years. Many point to the Nvidia embargo as the cause, but the details don’t really matter.

The point is they had different constraints: more engineers than data centers to train in. Inevitably, they would develop a different kind of solution. One thing for certain is that all firms will look at model development practices with an emphasis on driving efficiencies.

As I wrote last month about OpenAI’s o3, early versions are often expensive, but we can assume that the performance we get at $3,500 will cost us substantially less, perhaps a dollar or two, within no more than a couple of years. The cost of GPT4 quality results has declined by more than 99% in the last two years. GPT-4 launched in March 2023 at $36 per million tokens.

Except today, China’s DeepSeek offers similar performance for $0.14, or 250 times cheaper. 

Former Clubhouse influencer Marc Andreessen probably had it right when he posted on Twitter "AI’s Sputnik moment has arrived".

So panic all around. AI and chip manufacturer stocks went into free fall this morning as the market reacts to DeepSeek.

Oh, and yesterday DeepSeek was subject to a massive cyber attack. The tech mavens and financiers with so much $$$$$$$$$$$$$$ invested in OpenAI and its kin will do anything for revenge 😈

But if you’re looking for a real break down of what DeepSeek can’t do that ChatGPT can, it’s a lot of quality of life stuff. It can’t generate images, can’t talk to you, doesn’t support third party plugins, and doesn’t have “vision” like ChatGPT does.

All that said, on Monday, DeepSeek released an open-source image generator called Janus-Pro-7B that is, once again, as good, if not better, than OpenAI’s DALL-E 3. 

But the key thing is this. Limitations aside, the fact DeepSeek is essentially free, costing cents to use its API, open source, and was reportedly created by a team for only around $5 million, has raised several existential questions for America’s tech giants.

Or as noted AI evangelist and OpenAI (former?) superfan Ed Zitron wrote on Bluesky yesterday:

The AI bubble was inflated based on the idea that we need bigger models that both are trained and run on bigger and even larger GPUs. A company came along that has undermined the narrative - ways both substantive and questionable.

But to put all of this in larger context, Andreessen’s Sputnik comparison isn’t totally inaccurate. Especially if you, like him, believe that artificial general intelligence is both possible and a genuine nuclear-level threat to our existence or space-race-esque quest to change the future of humanity. But I’d actually compare DeepSeek to something much more recent: TikTok.

And some of the commentary was hilarious:

But think of this along the lines of TikTok.

If the 2010s was the story of American tech platforms eating the world then the 2020s has been the story of Chinese platforms biting back in a big way. Not just because users have, in the case of TikTok, liked them better or found them more addictive, but because they are, also, much cheaper.

And in September 2021, when TikTok broke a billion monthly active users, the mistake — well, one of the many mistakes — Silicon Valley made in response was assuming that they could simply change the game they were suddenly losing.

Meta, in particular, had lobbyists pushing hard for the TikTok ban as they, at the same time, renamed themselves Meta and went all in on, first, the metaverse, then AI, and now, some sort of vague middle ground between the two.

But it wasn’t just Meta. Andreessen and every other grand wizard of that San Fransisco fascist and Dark Enlightenment crowd decided that AI, not social, was the future and that future was assuredly still American. Instead of accepting the new bi-polar tech landscape increasingly shared by both China and the US, they shut out Chinese competition at home and immediately began throwing other peoples’ money into a furnace, promising it would eventually bring about the new revolution that they, coincidentally enough, just so happened to be early to.

And so this weekend, thanks to DeepSeek, we not only learned they never needed all that money to build the future - but that they weren’t even that good at it at building it. 

Now, we don’t yet know how the American AI industry will react to DeepSeek, but OpenAI’s Sam Altman immediately announced that free ChatGPT users are getting access to a more advanced model. Likely as a way to quickly respond to the DeepSeek hype.

Meta are also frantically beefing up their own AI tools (in a way that damns your privacy, by the way).

But it’s hard to imagine how American AI companies can compete after they spent the last 4 years insisting that they need infinite money to buy infinite computing power to accomplish what is now open source.

Oh, and DeepSeek r1 can even run without an internet connection. So it’s possible that OpenAI, the biggest money sink of all, may, as cognitive scientist and AI critic Gary Marcus wrote today, “some day been seen as the WeWork of AI".

And that some day might be sooner than you think. The mood is changing fast. AI analysts came out in droves and said:

Hey, you AI bros? So what are you telling us. That more than 95% of your the cost of developing new AI models is ... purely overhead?

But, like TikTok, it’s doubtful that American tech oligarchs are actually capable of accepting how screwed they are because AI is not just a massive pyramid scheme to them. It has ballooned out into a psuedo-religion. And Altman and Andreessen and their kin have spent their time frantically posting through it, doing their best impression of a doomsday evangelist trying to convince his flock that, "yes, he knew that the roadmap was changing" and that, "yes, the promised revelation is still coming". 

Your three key take-ways from all this stuff:

• Cost breakthrough: DeepSeek’s R-1 rivals OpenAI’s o1 in performance at 10% the cost, enabling affordable, high-quality AI.
• Open-source edge: Freely available, runs on modest hardware, sparking rapid developer adoption and open-source innovation.
• Industry disruption: Big Tech (OpenAI, Google, Meta) and Nvidia’s GPU business face pressure as low-cost models challenge closed, expensive systems.

For the gritty details, comb through Eric post linked at the top of this post.

And most intriguing for me, and just too much to adequately address in this post, is the business model. DeepSeek was a model developed as a side project by High-Flyer, a hedge fund rather than a conventional AI lab - underscoring the growing synergy between AI and quantitative finance which should be a U.S. advantage but the Chinese have shown they are masters.

In the United States, it is commonplace for some of the brightest minds from top universities to gravitate toward finance roles, drawn by high-paying opportunities in hedge funds and investment banks. This often leaves fewer brilliant graduates working on innovations far away from tech. 

China looked at the West—the U.S. in particular—and saw the overbearing importance of the finance industry at the expense of the real economy. Many of the country’s most brilliant graduates from Ivy League schools ended up in the increasingly parasitic finance industry instead of working on projects that move society forward.

Simply put: you want your best minds building real value, not just extracting it. 

The success of DeepSeek highlights the potential reallocation of talent that may lie ahead. Rather than remaining tied to traditional finance, highly skilled technologists are beginning to shift toward AI research and development. This phenomenon can be described as a “brain drain” in traditional sectors, with specialists in mathematics, statistics and computer science moving into AI instead of staying in finance or related fields.

By merging financial ingenuity with cutting-edge machine learning, projects like DeepSeek push the boundaries of what is possible, even if they emerge from unexpected corners of the tech-finance landscape.

As Eric noted in his analysis, one reporter noted at the end of his piece on DeepSeek:

"This is the end of the beginning of AI".

That could be true on many levels. Get out your 🍿

* * * * * * * * * * * * * * 



For the URL link to this post, please click here.

To read my other posts,

please visit my full archive by clicking here

* * * * * * * * * * * * * *