Who are the biggest cloud players today? Amazon, Google, and Microsoft, three companies all worth more than $2 trillion in market capitalization. All three were early to the cloud, launching AWS, Google Cloud, and Azure in 2006, 2008, and 2008, respectively.
Yet another technology company, IBM, rolled out “the cloud” four years before AWS.
In 2002, IBM announced a new service called Linux Virtual Services, which would allow customers to run their own software on IBM computers in its data centers. Clients would be charged per usage instead of having to sign long-term contracts. In a 2002 Wall Street Journal story, IBM executive James Corgel said the company saw “a huge opportunity going forward,” estimating that the on-demand computing market would be worth between $100 billion and $150 billion in five years. Corgel didn’t know just how right he was: in 2024, AWS alone had annualized revenue exceeding $100 billion. And the market is still growing.
Yet despite IBM’s early entrance to the cloud market, it lost. In 2023, IBM generated $62 billion in revenue. Meanwhile, AWS alone generated $91 billion for Amazon. While IBM may’ve been the first to launch “the cloud,” it failed to win the market because it couldn’t find product market fit. Today, IBM is worth roughly the same valuation it had in 1999 and 2000.
In the future, when we look back at all these AI companies, we may end up viewing OpenAI as the IBM of the AI wave: the first mover that failed to capture all the economic value.
There are a few reasons why.
First, AI-scaling laws appear to be providing diminishing returns. Until November, the consensus was that AI would continue to improve as computing power increased. But OpenAI, Google, and Anthropic have begun to see smaller marginal returns on further increasing compute. As a result, OpenAI’s new model, Orion, fell short of the company’s desired performance.
On top of slowing performance gains, customers don’t need frontier-edge models to accomplish their goals now that “normal,” cheaper models are quite powerful. Most companies just need an AI model to do a few specific tasks, and super-powerful, all-reaching models are unnecessarily large and expensive.
Even if OpenAI’s Orion model proves to be the most powerful model on the market, its utility to customers may be only marginally higher than a much cheaper alternative.
When you combine plateauing improvements with a plateauing customer need for improvements, models quickly grow commoditized as they converge to the same performance level. When models are commoditized, customers will choose the cheapest option, eroding OpenAI’s margins.
Fintech unicorn Ramp provides a good example of this commoditization in practice. It’s been using responses from OpenAI’s GPT-4 to help it fine-tune open-source models from Mistral and Meta, and the resulting custom models are cheaper and better than GPT-4.
Another risk facing OpenAI is a talent outflow: thanks to tender offers from outside investors, early employees have sold their now very valuable OpenAI stakes, and at least nine key executives have stepped down. In a competitive AI marketplace, where Meta and Musk are gunning for you, competition is cutthroat and losing top talent could be fatal.
That brings us to the last point: competition is fierce. Meta, xAI, and Anthropic are spending billions to keep scaling their AI models, and open-source models, which companies can fine-tune to address their own needs more cheaply, continue to improve. As baseline AI tech keeps getting better, customers will likely opt for cost efficiency over anything else, in a shift that doesn’t bode well given that OpenAI’s models are more expensive to run than open-source alternatives.
It’s possible that in 20 years we’ll live in a world where AI is ubiquitous, but OpenAI won’t be the big winner because the technology got commoditized. OpenAI is worth $157 billion today. Could it, like IBM, still be worth the same valuation in 20 years? Maybe.
