Who shoulders the burden of proof?
That’s a key question for investors pondering the prospect of a Chinese disruptor in the heretofore US-centric AI narrative.
I’ve read as much as one man can possibly read about DeepSeek over the past 36 hours, and I gotta tell you: I’m not sure I’m much the wiser for it. Details are sparse, and you won’t be blamed for wondering, as one bank did early Tuesday, if DeepSeek might be “a CCP deep fake.”
That’s not to say it isn’t a real company (it plainly is), nor that it’s not onto something (again, it plainly is), but it’s certainly to suggest that when it comes to the implicit contention that at least some of OpenAI’s world-changing advancements can be replicated by a Chinese hedge fund manager with a few million dollars and a handful of super-smart graduate students, the burden of proof’s on the guy making the astonishing claim.
On the other hand, if you’re plowing tens (hundreds) of billions of dollars into AI development and you plan to keep doing so for the foreseeable future, you also have something to prove — namely that it’s worth the money, and that eventually, shareholders can expect a return on those investments.
As the figure shows, costs are piling up, and the Mag7 C-suite generally insists that pursuing this dream requires throwing ever more money down the rabbit hole from now until the advent of AGI.
Up until this week, big US tech got the benefit of the doubt on that score, but DeepSeek’s audacious claims (and it might be more accurate to say claims people are making on behalf of DeepSeek) threaten to pull forward the “show me” moment for America’s giants, without whom the S&P would be ~12% lower.
The simple chart below, from SocGen, gives you a sense where we’d be right now were it not for Nvidia and the companies who buy from it.
“Nvidia and its top four customers — Microsoft, Google, Amazon and Meta — have contributed approximately 700 points to the S&P 500 over the last two years,” SocGen noted, adding that Nvidia alone accounts for 4% of index performance.
The bank went on to suggest that the answer to the question posed here at the outset is in fact mega-cap US tech, not DeepSeek.
“There are many questions about the ‘true cost’ of DeepSeek AI rollouts, their capabilities and the wide acceptability of their products [but] arguably, the onus may be on the hyperscalers to justify their capex projections,” Manish Kabra said, before noting that ultimately, cost-efficient AI should be a “net positive” for the world.
The burden of proof has always been on the ones spending $100BN of shareholder money on mammoth liquid-cooled nuclear-powered datacenters full of rapidly obsolescing but oddly not so rapidly depreciating GPUs.
The one Mag that didn’t jump on the Magnificent capex bandwagon is the one whose stock is doing the best through this. Better to be a user of AI models than a provider of AI models?
I have lived on the fringe of this DeepSeek, AI, Crypto, FX, global capital flows ever since my Dad , MIT and in 1963 head of Central Data Processing at FW Dodge paid me to learn PL-1. Grad school was writing models twards a PhD before Nixon blew up Bretton Woods. I am like Rodney Daingerfield – no respect. In 1978 FX made more money for Citibank, we invented global swaps and I couldn’t afford an apartment in Manhattan. This whole area has been little mice scurrying around the ignorant giants. Even hedge funds ignored us as we spent 20 years arguing with the capital holders as to whether we were even an asset class. I spent a summer in China in 2007 and they stole – I stole – ideas. We were all mice, not trusting but listening. They are smart and hungry despite how badly their CCP works. AI, etc. js a business for brains and machines but still brains first. Big companies are blowhards and assholes plus a danger to the world. the big US guys are burning lots of capital led by Trump, we need to think our way around those guys.