Anthropic files to go public, and the money declares the AI race a two-horse sprint

Information Technology 4 min 11 sources

Anthropic confidentially filed for a US IPO, beating OpenAI to the public markets — while a CNBC investigation found the same AI wave is quietly killing the generation of startups built just before ChatGPT.

Key takeaways

Anthropic, the maker of Claude, confidentially filed to go public — beating OpenAI to the markets and signalling the money sees the AI race as a two-horse sprint.
The same wave is quietly killing the generation of startups built just before ChatGPT: disrupted or dead.
Chips remain the wall, Brussels is drawing a line around the cloud, and an AI forecast beating the weather service deserves more attention.

Anthropic heads for the public markets

Anthropic, the AI company behind the Claude models, confidentially filed for a US initial public offering on Monday — getting there ahead of its rival OpenAI [2][8]. A confidential filing lets a company start the IPO process privately before its finances become public, so the timing is the signal: an analyst quoted by Reuters read it as Anthropic trying to reach the public markets first for strategic advantage [2]. The company recently raised $65 billion at a reported $965 billion valuation, and its backers now include Blackstone and Insight Partners [2][5].

The move is a test of a single question: will ordinary investors pay for the AI boom at the prices private investors have set? Reuters called the filing a moment that “ratifies Wall Street’s AI obsession” — tech stocks are leading a record run on the Nasdaq and S&P 500, Elon Musk’s SpaceX is set to price its own IPO this month, and OpenAI is “waiting on deck” [5].

For anyone in or near tech: when the two frontier labs both move to cash in at once, that’s less a vote of confidence in the future than a read on the present — the people closest to the technology are choosing to lock in today’s valuation.

The other side of the boom: “disrupted or dead”

The same wave is drowning an earlier generation. A CNBC investigation found that startups built in the years just before ChatGPT — companies whose whole product was a narrow slice of language or document handling — are being crushed, because the thing they sold is now a free feature inside a general AI model [3]. Their valuations, set in the old world, no longer hold in the new one [3].

Capital is not leaving the sector — it is concentrating. DriveNets, which makes networking software for AI data centres, raised $410 million, with the chipmaker AMD joining as an investor [4]. Defence-tech firm Mach Industries hit a $1.8 billion valuation, quadrupling in a year [9]. The money is flowing past the old startups and into the layer beneath the models — the chips, the networks, the infrastructure the boom physically runs on.

Chips: the wall, and the holes in it

That infrastructure is also where the US-China contest is fought. On Sunday the Commerce Department issued guidance to close a loophole that may have let companies ship the most advanced American chips — Nvidia’s top Blackwell processors — to subsidiaries of Chinese firms based outside China [10]. On Monday, Democratic Senators Elizabeth Warren and Andy Kim said it wasn’t enough and called for Commerce Secretary Howard Lutnick to testify, arguing the administration had allowed the leak in the first place [10].

The pressure is having a second effect the controls were meant to prevent: CNBC reports that China is learning to build without Nvidia, designing around the restrictions rather than waiting them out [11]. Export controls buy time; they also force the blocked side to build its own version. Both things are true at once.

Microsoft tries to win developers back

Microsoft holds its Build conference in San Francisco this week, in a smaller venue than usual — a deliberate, more intimate pitch [1]. The plan, per The Verge, includes a Copilot “super app,” a new reasoning AI model, and a batch of Windows improvements [1]. The framing is telling: a 20-year Microsoft watcher called it the most pivotal Build he can remember, with the company’s standing among developers — its trust in Windows and GitHub — the thing actually at stake [1]. The platform owners are not above the anxiety; they’re managing their own version of it.

Brussels draws a line around the cloud

The European Union is drafting rules that would curb Amazon and Google’s access to “strategic” public-sector cloud contracts, according to a draft document seen by Reuters [7]. The aim is digital sovereignty — keeping the most sensitive government workloads from depending entirely on a handful of American providers [7]. It’s the cloud version of a debate Europe has already had over chips and energy: how much of the thing your society runs on can sit in someone else’s hands.

The story worth more attention: an AI forecast beating the weather service

Away from the money, a quieter result. A startup using AI to forecast the weather is now out-predicting government meteorological agencies on some measures, TechCrunch reports [6]. It matters beyond weather. National weather services are decades-old public infrastructure, expensive and authoritative; an AI model trained on the same public data starting to beat them is an early, concrete case of a pattern that will repeat across public institutions — where a well-aimed model quietly outperforms the agency whose job it was. Worth watching not because the forecasts are better, but because of what it says about everything else governments do that looks like prediction [6].

02 · Lesson · why it matters

Disrupted doesn't mean you got worse — it means the ground moved

When a general-purpose technology becomes the platform everyone builds on, the companies sitting just above it don't lose a fair fight — they get absorbed, because the floor rose until their product was a feature.

Two headlines, one shape

Two tech stories ran side by side on Monday. Anthropic filed to go public, racing OpenAI to the markets while investors pour money in. And a generation of startups built just before ChatGPT is being, in one founder’s words, “disrupted or dead.”

It’s tempting to read those as opposites — winners and losers. They’re the same story told from two heights. Both are about what happens when a new general-purpose technology stops being a product and becomes a platform. Understanding that shift explains far more than today’s news.

A platform is a floor that keeps rising

Most competition is two companies fighting on level ground. You build a better version of the same thing, win some customers, lose some. The ground stays put.

A general-purpose technology doesn’t compete with you on the ground. It raises the ground. The large language model didn’t out-compete the startup that did clever document summarising — it made summarising a thing any model does for free, as a side effect. The floor came up until that startup’s whole product was standing underwater.

That’s why “disrupted” is the wrong word if you hear it as “outcompeted.” Those companies didn’t get worse. Their work was fine. The baseline of what counts as ordinary moved up past them, and a product that was special last year became a default this year.

Above the line or below it

So the useful question about anything built on a fast-moving platform isn’t “is it good?” It’s “does it sit above the floor, or below it?”

Below the floor: the value you add is something the platform will soon do itself. Document summarising, basic chat, simple classification — features now, not companies. Above the floor: the value comes from something the platform can’t easily absorb — a hard dataset, a regulated relationship, a physical operation, deep workflow no general tool will bother to learn.

The catch is that the floor keeps rising. Above-the-line today is below-the-line in two years. Staying above it isn’t a position you reach; it’s a thing you keep paying for, climbing as fast as the platform does. That’s the quiet anxiety even Microsoft is managing this week — the platform owner, worried about its own footing.

Why the winners are cashing out now

This also explains the rush to go public. If you’re Anthropic or OpenAI, you’re not only the platform — you’re also racing other platforms, and you know the floor rises under you too. Today’s valuation reflects today’s gap between what you can do and what everyone else can. Filing to go public now is a bet that the gap is about as wide as it will get — that the smart move is to convert the lead into capital before the next floor-rise narrows it.

When the people closest to a technology move to lock in its present value, that’s worth reading. Not as proof the future is bright, but as a sign that they think now is the moment the present is worth the most.

Where you’ll see this again

You don’t have to run an AI startup for this to land. The same shape governs a skill, a job, a side business, a tool you depend on. Ask it plainly: is the value here something the platform underneath will soon do for free?

If yes, you’re standing on a floor that’s rising, and the move is to climb — toward the part that’s hard to absorb — or to cash out while the gap still pays. If no, you’re holding something the platform can’t easily reach, and that’s the rarer, steadier place to stand.

The mistake is to assume that because your work is good, the ground will hold. Under a rising platform, good is not the question. Height is.

03 · Lab · your turn

Above the Line

Run a company under a rising platform and feel how standing still lets it absorb you, while every defense costs runway.