‘A million dollars isn’t cool; you know what’s cool? A billion dollars!’

Hollywood movie buffs will remember this iconic line by the fictional Sean Parker to the Mark Zuckerberg character in the film The Social Network. It’s a scene that succinctly captures a crucial moment in the early days of Facebook, where Parker explains how valuable Facebook could be.

Back then, a billion dollars was considered big and super cool. Since then, the world has moved on to trillion dollars in market cap for a few companies including Facebook (Meta Platforms). The more the number of zeroes following a non-zero number, the more cool it was. But that was only until the release of DeepSeek R1  last week—the open source AI model that many experts consider to be revolutionary. Marc Andreesen, co-founder and general partner at one of the most successful and largest technology VC firms, termed DeepSeek R1 as ‘one of the most amazing and impressive breakthroughs I’ve ever seen—and as open source, a profound gift to the world’.

Pioneers and arrows in the back

AI pioneer OpenAI was valued at a whopping $157 billion at the time of its latest fund-raising event in October 2024 (Andreesen Horowitz is an investor in Open AI). For a company which, based on estimates, is likely to have made revenue of around $3.7 billion and losses of $5 billion in 2024, the $157-billion valuation reflects an unprecedented level of future growth expectations loaded in!

But now, questions will rise on the bloated valuations that OpenAI and some of the listed Magnificent Seven companies—Apple, Alphabet, Amazon, Microsoft, Meta Platforms Nvidia and Tesla—command. All these companies have an AI element that has pushed up their valuations. These will come into question now.

DeepSeek R1 has proven that a competitive LLM can be apparently built at roughly 3 per cent of the cost at which the US companies have built their best models. While there is a view from Alexandr Wang of Scale AI that Chinese AI labs (including DeepSeek) could be under reporting the Nvidia GPUs they have as they could have got it in violation of US chip export control restrictions. Even considering this speculative point, DeepSeek R1 costs are substantially lower.

DeepSeek, in essence, has done to Silicon Valley what many Silicon Valley companies had done to each other—history is loaded with examples of many companies that were pioneers in their field, but were outdone by a few who copied them and then did a better job—Yahoo-Google, Nokia/Motorola/Blackberry-Apple, Myspace-Facebook etc. Many pioneers have ended with arrows in the back.

Along with this, DeepSeek is also sending a message — ‘A billion dollars isn’t cool, you know what’s cool? A million dollars!’ It’s far cooler to build AI capabilities with a few million dollars, rather than pouring in billions. Maybe as Warren Buffet said, ‘a fat wallet is the enemy of superior investments.’

With over $200 billion of capex by just 5 of the Magnificent Seven companies in 2024, and with their capex expected to surge even more in 2025 to solidify their AI fortresses, these US Big Tech companies might now be pushed back to their drawing boards and reassess their approach to investing in AI. Investor scrutiny in assessing the ROI on these investments may also increase.

This will have implications for the AI rally from here. Last week could have well been an inflexion point in the AI boom.

Related Topics