The number is so big that it hardly has any meaning. Forty billion dollars. This is what Alphabet has committed to investing in Anthropic, the four-year-old company that created Claude and was, until recently, thought to be Google’s most unsettling competitor in the competition to define artificial intelligence for the general public. Now, ten billion comes in cash. Thirty more dangles are available, provided Anthropic meets performance targets that the two businesses have mutually agreed upon. The check is real, but it has an odd shape that combines elements of an investment, a credit line, and something the industry hasn’t yet given a name to.
It’s important to keep in mind how recent everything is. Google paid about a tenth of Anthropic $300 million in 2023. Two more billion was added months later. The position had risen above three billion and a fourteen percent stake by the beginning of this year. We are currently discussing something valued at $350 billion, which is more than ten times that amount. In one version of this tale, Google is just increasing its winnings. In a different, less comforting version, Mountain View has acknowledged that, regardless of Gemini’s next move, it cannot afford to lose Anthropic’s clientele.
These days, if you walk through any developer conference, you’ll hear Claude Code mentioned with the same casual reverence that used to be exclusive to Stack Overflow. From about $9 billion at the end of the previous year, Anthropic’s annualized revenue has increased to over $30 billion, a curve that even seasoned investors describe with a small, weary shake of the head. The business is running out of processing power. OpenAI has been publicly making fun of its competitor for not obtaining enough of it. Thus, the peculiar logic of this situation begins to make sense: Google sells servers, Anthropic needs servers, and becoming the company’s largest shareholder is the simplest way to ensure a customer.
The majority of analysts dance around this part. Similar to the loop Microsoft established with OpenAI and Azure, a sizable portion of the forty billion will probably go back to Google in the form of cloud credits and purchases of Tensor Processing Units. Revenue out, money in. If you squint, it’s elegant accounting. This week, a commenter referred to the entire arrangement as a “rat king,” with tails entangled and no clear distinction between one company and another. It is difficult to get rid of the image.
Naturally, there is a significant strategic reward here. Google is adamant about Anthropic using TPUs instead of Nvidia GPUs. A tiny, continuous victory over Jensen Huang’s near-monopoly on AI silicon is achieved with each Claude query that is routed through Google’s chips. In a recent agreement with Google and Broadcom, Anthropic was able to secure 5 gigawatts of computing capacity; additional capacity may be added next year. Not to be outdone, Amazon came in a few days earlier with commitments to Trainium and AWS worth up to $20 billion in addition to its own $5 billion. It is impossible to overlook the pattern.
It’s difficult not to question how long-lasting any of this is as you watch it develop. Servers lose value. Every few months, models advance over one another. Google can purchase a seat at Anthropic’s table for forty billion dollars, but it cannot purchase certainty. Former OpenAI researchers who left the company in 2021 due to concerns about speed and safety founded Anthropic. These same researchers, who were once concerned that the companies might move too quickly, now sit on a balance sheet bigger than some national treasuries. The industry seems to be writing checks against a future that no one has fully envisioned, and by the time they do, the loop will be too tightly wound to untangle.
