Here is the number that should stop you mid-scroll: Google is prepared to write Anthropic a check worth up to $40 billion. That is not a typo, and it is not a lifetime commitment — yet. The initial tranche is $10 billion, with the remainder unlocking only if Anthropic hits certain performance benchmarks. Still, even the floor here is staggering.
This comes just days after Amazon quietly dropped $5 billion into the same company, with a similar performance-linked structure attached. Both deals peg Anthropic's valuation at $350 billion — a figure that would have seemed delusional for a five-year-old AI startup roughly two years ago.
So what is actually driving this? Anthropic's Claude models have been on a genuine tear. Claude Code, its software development tool, has become a serious productivity weapon for engineering teams. It is not magic — results depend heavily on the project, the team, and how the tool is deployed — but enough companies have seen real gains that demand has exploded far faster than Anthropic can handle it. The company has resorted to throttling usage during peak hours and is reportedly considering restricting its most compute-hungry features from lower-tier plans.
That supply-demand crunch is precisely where Google and Amazon come in. These are not just financial investments in the traditional sense. Both tech giants are also providing AI chips and cloud infrastructure directly to Anthropic, meaning the money Anthropic receives will flow right back to them as compute spend. It is a closed loop that benefits everyone involved — and has become the standard playbook for big tech backing AI startups. Microsoft pioneered this model with OpenAI, and now it is being copied wholesale.
What makes this particularly interesting is that Google is not exactly a neutral party here. The company has its own frontier AI ambitions through Google DeepMind and its Gemini model family. Funding a direct competitor might look strange on the surface, but it makes more sense when you zoom out. Google wants to ensure it is the cloud provider of choice for whoever wins the AI race, and betting on multiple horses is cheaper than losing the whole pot.
Anthropics rise has also been turbocharged by something Google had nothing to do with: OpenAI's stumbles. A stretch of controversies around ChatGPT and its underlying models pushed developers and enterprise customers to look elsewhere, and Anthropic was well-positioned to catch them. Products like Claude Code and the newer Claude Cowork — which brings similar productivity tools to general knowledge work — gave those switchers something genuinely compelling to land on.
The broader takeaway is that the AI infrastructure race is accelerating in ways that make earlier funding rounds look quaint. Billions are being deployed not just to develop smarter models, but to build the physical compute backbone needed to run them at scale. Whoever controls that infrastructure has enormous leverage over the next decade of tech.
POLICY
Palantir staff openly discuss the company's descent into fascism
When two former Palantir employees reconnected by phone last year, one of them opened the conversation with a question that was not really a question: "Are you tracking Palantir's descent into fascism?" That was the greeting. No small talk, no preamble — just a blunt moral reckoning delivered as a hello.
That exchange captures something that is now spilling out of the company more broadly. Current and former Palantir employees are openly wrestling with what their employer has become, and the language they are using is not the careful corporate hedging you might expect. Words like "wrong" keep coming up — not "complicated" or "controversial," but wrong.
Palantir was built on a specific origin story. Founded in the shadow of September 11, with early backing from the CIA, the company positioned itself as a data intelligence firm that could help governments fight terrorism without trampling civil liberties. The tension was always baked in — you cannot build targeting software for the military and claim total innocence — but the founding narrative gave employees a framework for tolerating the discomfort. They were the guardrails, not the machine.
That story is getting much harder to tell. Over the past several months, Palantir's software has become a central tool in the Trump administration's immigration enforcement operations, helping identify, track, and facilitate deportations through the Department of Homeland Security. For employees who signed up to prevent surveillance overreach, watching their work become the engine of mass deportation has created something that one former staffer described plainly as an identity crisis.
The concern is not just about immigration. Palantir's deepening ties to an administration that many of its own workers view as a democratic threat have forced a broader reckoning. Company-released manifestos, a war in Iran, and the general direction of federal power have all contributed to a growing sense among some employees that Palantir has crossed from enabling defense into enabling something else entirely.
Palantir's official response leans heavily on the idea of internal debate as a cultural virtue. A company spokesperson noted that Palantir has never been a monolith of belief and that fierce internal disagreement has been part of its culture since the beginning. That may be true. But there is a meaningful difference between healthy friction and employees publicly invoking fascism to describe their employer's trajectory.
For a company that has historically enforced silence — requiring non-disparagement agreements from departing employees and discouraging press contact — the fact that these conversations are happening out loud is itself significant. Something has shifted.
The uncomfortable reality is that Palantir sits at the intersection of two stories that define this political moment: the extraordinary power of data aggregation tools, and a federal government increasingly willing to use them against its own residents. That combination is why these employee concerns deserve to be taken seriously, not just as internal HR drama, but as an early warning signal about where surveillance infrastructure and political will can lead.
Source: Ars Technica
That exchange captures something that is now spilling out of the company more broadly. Current and former Palantir employees are openly wrestling with what their employer has become, and the language they are using is not the careful corporate hedging you might expect. Words like "wrong" keep coming up — not "complicated" or "controversial," but wrong.
Palantir was built on a specific origin story. Founded in the shadow of September 11, with early backing from the CIA, the company positioned itself as a data intelligence firm that could help governments fight terrorism without trampling civil liberties. The tension was always baked in — you cannot build targeting software for the military and claim total innocence — but the founding narrative gave employees a framework for tolerating the discomfort. They were the guardrails, not the machine.
That story is getting much harder to tell. Over the past several months, Palantir's software has become a central tool in the Trump administration's immigration enforcement operations, helping identify, track, and facilitate deportations through the Department of Homeland Security. For employees who signed up to prevent surveillance overreach, watching their work become the engine of mass deportation has created something that one former staffer described plainly as an identity crisis.
The concern is not just about immigration. Palantir's deepening ties to an administration that many of its own workers view as a democratic threat have forced a broader reckoning. Company-released manifestos, a war in Iran, and the general direction of federal power have all contributed to a growing sense among some employees that Palantir has crossed from enabling defense into enabling something else entirely.
Palantir's official response leans heavily on the idea of internal debate as a cultural virtue. A company spokesperson noted that Palantir has never been a monolith of belief and that fierce internal disagreement has been part of its culture since the beginning. That may be true. But there is a meaningful difference between healthy friction and employees publicly invoking fascism to describe their employer's trajectory.
For a company that has historically enforced silence — requiring non-disparagement agreements from departing employees and discouraging press contact — the fact that these conversations are happening out loud is itself significant. Something has shifted.
The uncomfortable reality is that Palantir sits at the intersection of two stories that define this political moment: the extraordinary power of data aggregation tools, and a federal government increasingly willing to use them against its own residents. That combination is why these employee concerns deserve to be taken seriously, not just as internal HR drama, but as an early warning signal about where surveillance infrastructure and political will can lead.
Enjoyed this?
Get stories like this delivered every Tuesday — free.