OpenAI is a $157 billion behemoth, but now comes the hard part

• OpenAI is now worth $157 billion, but it still can’t be easy.
• High costs, competition and lack of profitability remain challenges as they try to build more powerful AI models.
• Investors and founders must now consider how long this competition can last.

OpenAI just closed the most lucrative funding round in Silicon Valley history. Now comes the hard part: emerging victorious in a highly competitive AI industry.

Although Sam Altman’s company cemented its status as a leader in the generative artificial intelligence boom this week, securing a new $157 billion valuation after raising $6.6 billion in fresh capital from leading investors , its leading position is hardly guaranteed.

Since OpenAI launched ChatGPT in late 2022, it has become clear that its mission to build large language models that rival human intelligence will involve exorbitant costs that require vast resources.

Sure, Altman’s company now casts a massive shadow over the industry with a new valuation of $157 billion. That said, a host of rivals are chasing it in an effort to compete for capital and resources, complicating the startup’s path to profitability.

So while OpenAI is having a moment to celebrate, we’re about to find out just how deep its moat is and whether a brutal wave of consolidation is ahead for Silicon Valley’s hottest industry.

An expensive race begins

While OpenAI’s new valuation and its new capital injection represent huge sums that would be the envy of any Silicon Valley founder, there are signs that Altman remains unfazed.

According to a Financial Times report on the fundraising, Altman’s nearly nine-year-old venture has asked its new backers — a lineup led by Thrive Capital and including Nvidia, SoftBank and Microsoft — to avoid investing in firms rivals, of which there are plenty.

Anthropic and Mistral, both worth billions of dollars, are looking to embrace OpenAI. So are xAI and Elon Musk’s Safe Superintelligence (SSI), the startup launched in June by Ilya Sutskever, the former chief scientist at OpenAI who led a failed coup against his former boss.

“For the biggest model builders, these mega rounds are the new normal as the cost of training runs for the biggest models climbs into the hundreds of millions of dollars,” said Nathan Benaich, founder and partner at Air Street Capital, a firm of investments.

There are some good reasons why OpenAI cannot rest easy.

First, the costs of delivering game-changing generative AI advances appear to be spiraling. Dario Amodei, CEO of Anthropic, said earlier this year that he expected AI model training revenue to exceed $10 billion by 2026 and reach $100 billion after that.

OpenAI’s own training costs could reach up to $3 billion a year, The Information previously estimated. GPT-4o costs about $100 million to train, but the cost will likely increase depending on the complexity of future AI models.

In part, the costs are driven by purchases of powerful chips — known as GPUs — bought mainly from Jensen Huang’s Nvidia to build clusters in data centers. These are essential for providing the computing power required to run LLMs.

The battle to retain talent has also been vicious in the current frenzy, with AI labs jostling for an edge over competitors being pushed to offer increasingly attractive compensation packages.

“These costs will only increase as companies invest more and more to fight for often marginal performance advantages over their competitors. This race has no obvious historical parallels due to the attractive capex requirements and the lack of a clear path to profitability.” Benaich told BI.

While OpenAI’s new arsenal of capital will help it fund some of the more expensive elements of its business, it’s still not exactly operating from a position of strength. Last week, a report from The New York Times revealed that the hottest artificial intelligence lab on the planet was on track to end the year with a $5 billion loss.

And OpenAI’s rumored attempt to push for exclusivity among investors could have downsides. Benaich described the move as “unusual” – but also a “perception of how it perceives its own market power”.

“It’s also a bold move that risks attracting unwanted attention from competition authorities,” he added.

For industry experts, it raises questions about how sustainable this is for the industry in the long run.

Investors see consolidation on the horizon

As OpenAI cements its position as an industry leader, investors anticipate some degree of consolidation among startups in the base model layer in the coming year.

LLM startups need constant access to a lot of capital, but not everyone can take advantage of the same influx of cash as OpenAI. With Microsoft acquiring Inflection.ai and Google similarly absorbing the founding team of Character.ai, investors expect more of these types of acquisitions in the near future.

“This is also a race for capital and only investors like sovereign wealth funds will ultimately be able to provide the kinds of capital these LLM startups need,” a VC at the stage told BI of European growth.

Once funding dries up, incumbents, including Big Tech giants, can scoop up smaller upstarts with a vertical focus. These companies have access to a wealth of proprietary data on which to train their models.

VCs are also seeing a more serious approach to backing LLM giants at high valuations. “A lot of other companies are raising capital for vision and hope, and I think you’re going to start to see a rationalization of that,” said another growth-stage VC, adding that there would be “a level of cooling. deflation of AI foam next year”.

“You don’t need 50 core model companies — you end up with two or four, maybe,” he said.

He added that those companies that survive will be those that tangibly serve consumers. “You might have Amazon, Anthropic, OpenAI, Meta and Google, but I can’t imagine there will be many more.”