The end of artificial intelligence’s venture era

May was an unusually big month for the AI industry.

Here’s some of what happened:

1. Anthropic said it’s on trackCNBCAnthropic set to hit $10.9 billion in revenue during second quarter, source says for its first-ever profitable quarter, with $10.9 billion revenue for Q2. A big piece of this was Claude Code enterprise deployments alongside contracts to create agents for PwC and JPMorgan. But part of that is presumably because of tokenmaxxing, which may notYahoo Finance'Harder to justify': Uber COO says no clear link between AI spending and useful product improvements persistThe VergeMicrosoft starts canceling Claude Code licenses. And there’s also the nagging part of SpaceX conveniently offering discountsSECSpace Exploration Technologies Corp. in May and June for compute.

2. OpenAI reportedly filedCNBCOpenAI to confidentially file for IPO as soon as Friday: Source a confidential IPO prospectus with Goldman Sachs and Morgan Stanley. It’s aiming to go public as early as September with a valuation above $1 trillion.

3. Anthropic boughtTech CrunchAnthropic has acquired the dev tools startup used by OpenAI, Google, and Cloudflare Stainless, which makes SDK/MCP tooling. The deal itself was not huge, but it’s a manifestation of some changes in how AI majors move.

4. Google Deepmind absorbedReutersGoogle DeepMind hires staff from Contextual AI in licensing deal, source says the Contextual AI team in what it frames as an $80–90 million licensing deal, a relatively small exchange in the grand scheme of things.

Zoom out a bit, and there are a couple of ways to read this. The first is that we’ve reached the beginning of the end of the AI industry’s venture-capital era and are entering its public-markets era. That’s exciting if you’re an early investor, but for now, not so meaningful for everyone else.

Another read is that there is consolidation taking place, but it is buried beneath deal structures that are designed to look very much not like consolidation. The two paths will converge if you keep reading; let’s go down the second one.

To use an analogy that I’m sure you’ve heard before, Stainless makes the plumbing that connects models with everything else. That means Anthropic folded in the infrastructure layer that developers use to build on top of what it offers; it didn’t buy a specific product suite.

The licensing deal between Google Deepmind and Contextual AI sidesteps a review by antitrust regulators since it isn’t an outright acquisition. This isn’t the first time Google has used this structure. In 2025, it hired key staff from AI code generation provider Windsurf and used the startup’s technology, paying $2.4 billion in license fees. A year prior, it set up a similar arrangement with Character.AI. Like these cases, the move with Contextual AI is more about acquiring expertise than assets.

In all of the deals, the major labs are after capabilities that form or strengthen moats for them, rather than revenue or users. This series of unrelated deals knits together a compelling answer to the question of what constitutes a moat when AI is the hottest business to be in, and when its proliferation is only going to become wider.

Put it this way: if you’re in a sector where the competitive window is measured in months or even weeks, a purchase to the tune of tens of millions of dollars for what you would otherwise take two years to build is the rational move, particularly when your valuation is $965 billionReutersAnthropic's valuation surges to $965 billion, surpassing OpenAI. That expense is a mere rounding error. And since the consolidation takes place at the infrastructure and talent layers, not in products, it doesn’t really look like consolidation from the outside.

This matters the most to people who, not long ago, formed the belief that it would be different to build in the AI industry. The venture-capital model assumes that the frontier is contestable, and that an agile team with a good idea, solid execution, and sufficient compute can challenge incumbents that are, well, not as agile or even resistant to change. But these deals signal that the AI industry’s own incumbents are closing off the paths that challengers can use. You can still build on top of these platforms, but they’re making it harder to build around them.

For the past three to four years, the venture market bankrolled the experiments that the AI industry’s incumbents were either too slow or too cautious to run themselves. (That was a strange sentence to write, given how we’ve repeatedly described the sector as moving at breakneck speeds.) With this arrangement, the labs benefitted from a constellation of builders stress-testing their platforms, while builders got access to frontier technology and a shot at building something meaningful. The incumbents that are still private, although not for much longer, don’t need that setup anymore, and are now ready to stop outsourcing the valuable parts of the industry and start owning them.

The result is a bifurcation that may bother many: some AI developers are building products that probably have low defensibility, while others are building to be acqui-hired in the future. For the thousands of founders who are mid-build, the map has changed. They need to thread a needle in the gaps that major labs cannot be bothered to fill on their own. It’s a different kind of map, one that isn’t on the frontier.

Hi! This is Vidhatri, the producer of Zero Shot.

For as long as I can remember, one version of success around me primarily looked like this: join an engineering college, get recruited by an IT-services company, get trained, work for clients abroad, maybe even go to the US for one of these projects, and settle down.

And for years, this path paid off, changing the lives of many families. But with AI existentially cracking this model, all eyes are on how IT-services companies will reinvent themselves.

The recent news about HCLTech’s $150 million bet on Sarvam is one answer to that major question. As per reports, no VC led this round. But HCLTech, one of India’s largest IT-services firms, is looking to write the biggest check in Sarvam’s $300 million raise, which valued the company at $1.5 billion. This is a 7X markup on Sarvam’s last fundraise from December 2023.

But what is HCL buying and what is Sarvam selling? What does this signal about the IT-services industry?

On this week’s episode of Zero Shot, host Rohin Dharmakumar debated these questions with Abhishek Pathak, the lead IT services and internet analyst from Motilal Oswal, and Kashyap Kompella, the founder of RPA2AI Research, an advisory firm focused on AI governance, risks, due diligence, and policy.

Tune in! You can listen to the episode on Spotify, Apple Podcasts, or The Ken‘s app.