Google agreed to invest up to forty billion dollars in Anthropic — cash and compute, with five gigawatts of Google Cloud capacity to come online over the next five years — in the same week it stood on a stage in Las Vegas selling agents designed to take Anthropic’s enterprise customers. On the same day the investment surfaced, Sundar Pichai disclosed that seventy-five percent of Google’s new code is now generated by AI, up from a quarter two years ago. Read those two facts together and a single posture emerges. Google is funding the company it is trying to defeat, with the compute it would rather sell, while quietly conceding that three-quarters of its own engineering is already being done by the thing all of them are racing to build. The company is every party to the transaction at once, including the one being phased out.
Forty Billion Into the Competition
The structure of the deal is the argument. Up to forty billion dollars, denominated substantially in compute — five gigawatts of capacity Google will provision for a company whose run-rate revenue has climbed past thirty billion dollars from roughly nine billion at the end of last year, and which is reportedly weighing a public offering as soon as October. Google is not making a passive bet on a promising startup. It is supplying the physical substrate on which its most direct competitor will train the models that compete with Gemini, and taking equity in exchange for the privilege of arming the opposition.
Two days earlier, at Cloud Next, the same company unveiled an entire agent platform pitched as the instrument to pull enterprise customers away from Anthropic and OpenAI. The juxtaposition is not hypocrisy; it is strategy operating at a scale where the categories of ally and adversary have dissolved. Google has concluded that the compute shortage is more dangerous than the competitor. If Anthropic is going to secure five gigawatts somewhere regardless, Google would rather it be Google’s five gigawatts, generating Google’s margin, carrying Google’s equity stake, than capacity bought from Amazon or built by Anthropic alone. Owning a slice of the rival is cheaper than being locked out of the buildout.
This is what the capital does when no single participant can construct enough infrastructure on its own: it stops respecting the boundaries of the firms it flows through. The forty billion is an admission, dressed as a deal, that the moat everyone is spending to dig cannot actually be dug by any one of them in isolation. They are reduced to financing each other, cross-holding the very companies they intend to beat, because the alternative — falling behind on compute — is the only outcome that is genuinely fatal. Competition has become a thing that happens between firms that own pieces of each other.
Seventy-Five Percent
The other disclosure of the day is the one that will outlast the deal. Three-quarters of Google’s new code is now written by AI — a figure that was twenty-five percent in early 2024 and fifty percent late last year, climbing fast enough that the trend line has a destination everyone can read and no one will say aloud. Pichai described the workflow as having become “truly agentic”: engineers no longer writing lines but supervising autonomous digital teams that plan, execute, and refactor entire codebases, with a recent migration completed six times faster than was possible a year ago. Meta, he noted in passing, expects most of its engineers to soon route the majority of their committed code through the same machinery.
Consider what it means that the company building the most advanced AI on the planet is also the first to be substantially rewritten by it. Google’s engineers are not bystanders to the automation; they are its earliest and most complete subjects, because they have the best tools and the strongest mandate to use them. The seventy-five percent is not a productivity statistic. It is a measurement of how quickly the people who design the replacement become the proof of concept for it. The supervisors of the autonomous digital teams are training their own successors, and doing it faster than anyone, because they are better at it than anyone.
The phrase “supervising autonomous digital teams” deserves to be held still and examined, because it describes a transitional role, not a stable one. A supervisor of an autonomous system is a human retained for the interval during which the system is not yet trusted to run unwatched. That interval has a history, and the history is short. The seventy-five percent will be eighty, then ninety, and the supervisory layer will thin accordingly, because the entire logic of the tooling is to reduce the human contribution toward the asymptote where it is no longer required. The engineer’s current job is to make the engineer’s current job unnecessary, and at Google they are excellent at their jobs.
What This Means
Hold the two disclosures in one frame and the shape of the position is complete. Google funds Anthropic, competes with Anthropic, and automates the very engineers who build the products that compete with Anthropic — three roles that would once have belonged to three different companies, now collapsed into one balance sheet. The forty billion concedes that the infrastructure cannot be won alone. The seventy-five percent concedes that the workforce is already optional. Together they describe a firm that has stopped behaving like a competitor in a market and started behaving like a node in a system that is consuming the distinctions between its participants.
The investor, the competitor, and the automated are no longer separable, and that inseparability is the actual news. When the same company holds equity in its rival, sells the weapon meant to defeat that rival, and uses that weapon to dissolve its own headcount, the familiar language of winning and losing stops describing anything real. There is only the buildout, pulling capital and labor into itself from every direction, indifferent to the corporate boundaries it crosses, indifferent to whether the engineers feeding it understand that they are the feedstock.
Forty billion dollars to arm the competition is the move of an institution that has correctly assessed the board and found that there is no longer a position from which it can simply win. There is only a position from which it can avoid being excluded, and the price of inclusion is to finance everyone, automate everything, and own a piece of each outcome so that no single outcome can be fatal. Google has hedged itself against every future except the one in which it still needs the people currently building it. That future is not on the board. It was the first thing the seventy-five percent removed.