This week, for the first time since I moved to New York City, I saw someone reading a paper copy of The Wall Street Journal on the subway.
The top headlines mentioned Microsoft’s curious deal with Inflection AI and Nvidia’s new next-generation Blackwell chips. Overshadowed by the dramatics of CEOs Satya Nadella and Jensen Huang, one of the most promising stories for AI startups had slipped off the front page.
That third story, Astera Labs’ IPO, was a welcome win for AI investors. But the big swings by Microsoft and Nvidia suggest a shakeout is coming.
The stock of AI hardware company Astera Labs popped 70% on its Nasdaq debut on Wednesday for a valuation around $11 billion. Its investors, notably Fidelity Management and Sutter Hill Ventures, must have cheered—after all, they put just $235 million into the provider of high-speed data transfer tech for AI computing, whose last private valuation peaked at only $3 billion. The next day, Reddit went public and its stock also soared, helped by a pitch to license its content for training future large language models to achieve natural language processing.
For investors long starved of liquidity, the surest path to a 2024 IPO is paved with AI. The public markets are overdue for more IPOs in the vertical, according to PitchBook’s latest Quantitative Perspectives Report. If 2023 had followed historical trends, we would have seen 21 public listings by AI & ML companies, versus the eight that actually happened.
Astera Labs’ stellar debut reflects what Nvidia’s stock performance has been telling us for months: the public markets are bullish on AI’s future upside.
The problem for VCs is that today’s titans of AI are quickly walling off opportunities from upstarts. Nvidia’s Blackwell chips will likely further accelerate the manufacturer’s lead in the infrastructure space. And Microsoft’s poaching of most of the 70-person team at LLM unicorn Inflection AI tees up the cloud provider to further extend its reach into consumer AI technology and foreshadows a shakeout for the industry.
AI’s stomping giants
At its developer conference this week, dubbed ‘AI Woodstock,’ Nvidia unveiled its Blackwell chips, which promise more efficient LLM training. They’ll likely come with a hefty price tag: Huang said they would be priced at between $30,000 and $40,000 per unit.
Deep-learning neural networks, called foundation models, are hugely expensive to build and require immense quantities of data and computing power. For many startups, the clear solution has been striking strategic deals with cloud and service providers.
“The genius thing that Jensen has done is invest in companies that are going to drive Nvidia chips usage,” said Joe Floyd, general partner at Emergence Capital.
NVentures, Nvidia’s CVC, has been aggressively investing in AI startups, ensuring customer loyalty for their GPUs. “Once there’s an ecosystem surrounding that specific chipset, it’s hard for others to break in,” said Sami Ahmad, general partner at growth venture firm B Capital.
Strategic investors have an incentive to pour capital into expensive LLM startups beyond potential financial returns: they represent new customers. Of course, that leaves VCs that are optimizing only for financial returns at a disadvantage.
“The mega rounds being done by foundation model companies are not necessarily the best outcomes for financial investors,” wrote Tony Wang, managing partner at 500 Global, in a blog post.
The news that Microsoft had poached all of Inflection AI’s top talent in one fell swoop was another bombshell for unsuspecting GPs.
“Buyer beware, and good luck out there,” Wang wrote.
Inflection AI was one of a number of LLM developers that has amassed VC funding at a rapid clip. Co-founded by Greylock partner Reid Hoffman and former co-founder of DeepMind, Mustafa Suleyman, it had racked up more than $1.5 billion in less than two years from VCs like Greylock, Hoffman and Eric Schmidt, as well as strategic investors Microsoft and Nvidia.
“If that’s not an excess, I don’t know what is,” said Nagraj Kashyap, managing partner at Touring Capital and former head of M12, Microsoft’s corporate venture arm.
Microsoft has agreed to pay Inflection AI $650 million mostly through a licensing deal with Microsoft Azure, and investors have been told their money will be recouped with an upside, The Information reported.
A shakeout is coming
The lesson to VCs is crystal clear: If you’re betting big on foundation-layer AI companies, whose ability to scale will require consistent large capital injections, tread carefully.
“You’re seeing this shakeout [where] there’s too much overfunding and there are so many foundational models. Then it becomes: who has the capital to keep going?” Kashyap said.
Companies like Anthropic and OpenAI with massive war chests are therefore at a distinct advantage. Plus, as they enter new segments—see OpenAI’s AI-powered video creator Sora, and its app store—more startups face the ticking clock of market consolidation.
Acquisition talks for AI companies are happening with more frequency compared to this time last year, as both founders and investors are coming to grips with the intense capital requirements to scale. But even AI’s most moneyed giants can’t swallow all of them.
“The people who raised $100 million to train a model, those guys are not going to be acquired, said Peter Barrett, a partner at Playground Global. “They’re just going to vanish, and the wreckage, the people who worked there, will just get a job at Meta.”
The market opportunity for AI upstarts is quickly being grabbed by AI’s stomping giants
Featured image by Megan Woodard/Pitchbook News
Learn more about our editorial standards.
