The battle among frontier AI companies in startup land is looking more and more like a survival of the richest. Just take a look at the latest from OpenAI.
On Wednesday, Bloomberg reported that the poster company of the generative AI boom is in talks to raise $6.5 billion from investors at a $150 billion valuation. It is also in separate talks with banks over a debt raise through a $5 billion revolving credit facility.
At a $150 billion valuation, OpenAI would almost double the $86 billion value it held during a tender offer earlier this year. It would also vault the company closer to the heights of Elon Musk’s SpaceX, reported to be valued at $210 billion, and TikTok’s parent company ByteDance, said to be worth almost $230 billion.
It’s a sign of just how far OpenAI has come since November 2022, when its CEO Sam Altman described ChatGPT as “an early demo of what’s possible.” It’s also a sign of just how expensive the business of training and running AI models is getting.
Generative AI is a cash-guzzler
It was only last year that OpenAI revealed that it had secured a $10 billion investment from Microsoft in a deal that extended a partnership first struck in 2019.
Back then, in January 2023, OpenAI said it had already worked with Microsoft to “build multiple supercomputing systems” — powered by the tech giant’s cloud platform Azure — that could be used to train its models. OpenAI said it would deploy that $10 billion investment to further boost these systems.
However, as the generative AI boom has progressed since OpenAI launched ChatGPT in late 2022, it has become increasingly evident that making the large language models behind generative AI applications smarter — a fundamental goal — is an expensive endeavor.
Just look at the spending of Big Tech firms working to improve their AI models. In the last quarter alone, Alphabet, Amazon, Meta, and Microsoft saw capital expenditure hit $52.9 billion — almost double what it spent in the same quarter in 2023, per The Wall Street Journal.
Much of that has been directed toward assets critical for strengthening their AI plays, such as buying expensive chips from the likes of Nvidia and building new data centers to host these chips for training and inference purposes.
While privately held startups working on large language models, or LLMs, aren’t in the business of building data centers like Big Tech’s cloud players, they have been busy trying to secure as many powerful chips as they can. This is to ensure they have enough computing power to build their next-generation models.
Industry projections note just how expensive this is going to get. Earlier this year, Dario Amodei, CEO of OpenAI rival Anthropic, noted that it could cost $10 billion to train AI models within the next two years. Pretty soon, that could hit $100 billion.
This is a tricky situation for startups, of course, given that they don’t have quite the same deep pockets as their publicly traded counterparts. Their traditional funding sources in venture capital may not suffice.
According to Bloomberg, OpenAI has lined up blue-chip names, including Apple, Microsoft, and Nvidia as potential backers for its upcoming investment round, alongside lead investor Thrive Capital. That raises the bar for the handful of startup competitors it faces.
Amodei’s Anthropic seemed to have got the memo, as it raised $2.75 billion from Amazon in March, taking its total raised from the Big Tech firm to $4 billion. It was last valued at $18.4 billion.
OpenAI’s top Europe rival, Mistral, raised fresh capital at a $6.2 billion valuation in June, with its funding round peppered with big corporate names such as Nvidia and Samsung, alongside VC heavyweights such as A16z and Lightspeed.
It remains to be seen which companies will stay the course in the long term. Consolidation has struck the AI industry in recent months, as some top startups trying to build AI models have been gutted.
In August, for instance, news emerged that Google had hired Character AI CEO Noam Shazeer — a former Googler — along with top talent, and secured a licensing agreement for its LLM technology.
As my colleagues Ben Bergman and Sri Muppidi noted last month, the startup’s growth “has not been blockbuster and revenue is minimal.” As a result, it “did not have enough momentum to raise another big round in today’s ultracompetitive funding environment.”
This looks to be the key challenge going forward for startups seeking to get an edge in the LLM space and cut into OpenAI’s lead, especially as profit remains a distance away. OpenAI’s annualized revenue is on track to be $3.4 billion this year, a report said in June, though it’s not yet profitable.
In a recent episode of the No Priors podcast, tech investor Elad Gil suggested that “enormous capital moats are emerging” among startups looking to scale their LLMs to the biggest scale. The goal for many, of course, is AI that is as intelligent as humans.
“Everybody’s partnering up, and so it’s a really interesting question to ask,” Gil said. “For all the other players in the market, where are they going to get these ever-rising amounts of capital, and who do they partner with?”
AI startups serious about progressing on LLMs will need to figure out the answer to this pretty quickly. Failure may lie ahead if they don’t.
