A figure like $800 billion attached to a company that most people outside Silicon Valley couldn’t have named until a few years ago seems a little ridiculous. However, investors have been subtly speculating about that amount in recent weeks for Anthropic, the company that makes Claude, which is more than twice as much as the valuation it secured only a few months prior. According to several people familiar with the offers, the discussions are taking place quickly, frequently informally, and occasionally during dinners in Palo Alto where the wine is excellent and the numbers seem to keep growing.
It remains to be seen if Anthropic accepts any of these offers. Hot startups frequently receive preemptive bids and graciously reject them, particularly when an IPO is rumored to be imminent. It seems as though the company is purposefully letting the heat build, waiting to see what the market will bear before making any commitments. In this industry, it usually means that everyone is talking but no one wants their name on it when a spokesperson declines to comment.
The background is important. Anthropic closed a $380 billion round led by GIC and Coatue in February. At the time, that number seemed big. It feels like a discount now. Anthropic appears to have decided at some point that it wants to be evaluated on the same scale as OpenAI, the older and more prominent sibling in this race, which closed its own round last month at $852 billion. Anthropic is currently valued at $688 billion on Caplight, the secondary exchange where investors who are eager for an IPO exchange private shares. increased by 75% in just three months. It’s still unclear which is winning, but that kind of move typically indicates either collective delusion or true breakout.
The revenue is evident. Last week, Anthropic announced that its run-rate had increased from $9 billion at the end of the previous year to $30 billion. It’s the type of growth curve that has traditionally only been found in pitch decks, and it’s not a typo. In less than two months, the number of business clients who spend over a million dollars annually on Claude has doubled to over 1,000. Claude Code, the coding assistant that has subtly emerged as the preferred tool for engineering teams fed up with waiting for the competition to catch up, is largely responsible for that demand.
Anthropic was the talk of the town at HumanX last week, the current major event on the AI conference circuit. Jared Quincy Davis, the owner of the AI cloud platform Mithril, remarked, “They’re crushing it,” with the weary admiration of someone witnessing a competitor retreat. Founders gathered in groups outside the conference room, comparing notes about which Claude model they had switched to and when. Wearing hoodies, it resembled a small religious revival.
Mythos followed. Anthropic’s latest model, which was unveiled last week, is so potent that cybersecurity concerns prevent it from being made public just yet. That type of framing is very effective. It simultaneously conveys capacity, accountability, and scarcity. It’s a big deal at HumanX, according to Theory Ventures’ Tomasz Tunguz. “There’s a tremendous amount of excitement,” he said, which is about the same as standing on a chair for a venture capitalist.
Here, it’s difficult to ignore the pattern. The money comes after one of the frontier labs releases something that starts a new dialogue every few months. The tone of Mythos feels different. Anthropic isn’t merely asserting superiority. It is asserting a model that has ventured into an area where prudence is the key. That’s either an accurate assessment of AGI’s future direction or the best valuation theater the sector has produced in a long time. Maybe both.
The offers continue to come in for the time being. The race has obviously changed, regardless of whether Anthropic stays private long enough to test the $800 billion mark or jumps to public markets and lets the rest of us find out in real time. The pace is currently being set by two companies, not just one. Additionally, the gap is beginning to appear smaller than anyone anticipated, depending on the model and the week.
