Until two weeks ago, AI was the most expensive tech pitch ever told.

It went like this: AI is the most revolutionary technology in decades and the biggest pile of money will win the race. As the world’s most expensive poker game, only the deepest pockets need apply.

Then, DeepSeek. An obscure Chinese company announced a competitive AI model that had been trained for what seemed like pennies.

Did DeepSeek fatally puncture the pitch? What can we learn from the reactions of the tech giants as the narrative ground shifted beneath their feet?

To be clear, the ground is not done shifting. I recently met the CEO of CTGT, Cyril Gorlla, an impressive 23-year-old who recently raised an oversubscribed $7M seed round from a who’s-who of AI royalty. CTGT has technology that claims to improve deep learning for AI models by 500x. DeepSeek is more starting gun than closing bell.

It’s hard to overstate just how extreme the assumption of centralized gains for the biggest AI players had become. It fueled hundreds of billions in investment, and pretty much anyone with a 401k enjoyed the surging stock values of the “Magnificent 7.”

The $500B Stargate AI development project that Trump pitched on his second day in office was perhaps peak AI. There weren’t even concepts of a plan for $400B of it. Six days later, DeepSeek erased more wealth in 24 hours from one stock (Nvidia) than Stargate proposed raising over four years.

And just like that, the narrative had shifted. Even if DeepSeek cost more to train than reported, it clearly wasn’t Google/Anthropic/OpenAI-level money. Capital might still be necessary, but perhaps was no longer sufficient. Also, if low-cost open-source AI models are great, will AI become commoditized—perhaps showering humanity with untold economic benefit, but whose value is not necessarily captured by the tech oligarchy?

Here's how I see it: For years, the AI oligarchy told the story that the AI industry was a steep pyramid, and that money insulated them from competition further back. DeepSeek squashed that pyramid:

So how did the oligarchy respond? We saw pretty much every page of the crisis communications playbook. The scramble and panic that my contacts told me was pervasive inside of these companies was nowhere to be seen in their public comments:

• Steady as she goes: OpenAI’s Altman said that DeepSeek was “impressive,” but that they’ll continue to “lead the world with model capability.” In other words, meh. No changes announced on the $500B Stargate project.

• What, me worry? Channeling the spirit of Alfred E. Neuman, Mark Zuckerberg shrugged off the news. On Meta’s 4th quarter earnings call, he allowed that what DeepSeek is doing some “novel” things but that “I continue to think that investing very heavily in CapEx and infra is going to be a strategic advantage over time.”

• We already knew that. Microsoft has a dual strategy: work with OpenAI but ensure that Azure can deploy all models. CEO Satya Nadella kept the spotlight away from him by saying bland things like that DeepSeek has “real innovations.”

• We’re already doing that. Sundar Pichai of Google said, “for us, it’s always been obvious that frontier models could be made more efficient over time,” and that their models can “go toe to toe with DeepSeek” on efficiency. In other words, DeepSeek has merely highlighted the efficiency at which Google already excels.

• This is actually a blessing. Perhaps the happiest spin was from Nvidia, where CEO Huang pointed out that DeepSeek still used Nvidia chips (albeit older ones that we don’t have export restrictions on). The CEO of ARM, another chipmaker, went further and said that DeepSeek is “great for the industry,” and “expands the demand for overall compute.”

The broad point of the cash-rich tech oligarchy is likely correct: being able to outspend everyone still confers advantage. And, illustrating Jevons paradox, cheaper and more efficient AI will likely do nothing but increase AI adoption, sales of Nvidia chips, and total energy consumption.

But the shift in story creates a real threat for AI-only companies like Anthropic and OpenAI. Like Warren Buffett said, when the tide goes out you see who’s not wearing a bathing suit. Companies with built-in cash gushers like Meta, Google, and Amazon can weather pretty much anything. The only cash gusher OpenAI has is investors convinced that the AI door will close behind the few richest companies.

There’s an important distinction between your pitch and your plan. It’s not that you don’t believe what you’re selling, but that you realize everything might not go your way, at least initially. In other words, don’t breathe your own exhaust. I hope/presume OpenAI and the others are having that conversation.

But what does Altman of OpenAI do, at least publicly? Instead of retreat, he is doubling down. Just this past October, OpenAI closed a $6.6B investment, the largest VC round in the history of tech. Now, according to post-DeepSeek media reports, Altman is pitching a $40B round at $300B post-money, which is double the valuation from 4 months ago.

Who would buy that story? Meet Masayoshi Son, the patron saint of metastasized tech ambition. This is the man who invested $16B into WeWork, some of it at the company’s peak valuation of $47B and shortly before it crumbled to $44M in bankruptcy.

There might be some life left in the most-money-wins-AI story after all.