Anthropic Q3 26 – The Tokenmax Effect

Price cuts put Anthropic back in the red. 

• Anthropic has updated its S1 (see here) to include Q3 2026, where the end of tokenmaxxing has slowed growth, competition with a resurgent OpenAI has put it back in the red and there is more to come.
• This will call into question the market’s willingness to pay a premium valuation, which I think is beginning to look optimistic at $2tn.  
• Q3 26 revenues / operating profits were $16.96bn / LOSS $567m (page 160) compared to $11.56bn / $974m in Q2 2026.
• Q3 revenues grew a very healthy 46.6% QoQ, but crucially this is way below the 144.4% they grew QoQ in Q2 26.
• This is a result of the end of the tokenmaxxing trend at the end of Q2 26, when financial controllers in companies realised that tokens were costing a fortune and put in place cost controls to prevent massive bills.
• This is a perfectly normal state of affairs, but it shows that the predictions being made earlier this year that Anthropic would top $400bn in revenues in 2027 were nothing more than a fantasy.
• In fact, it appears that Anthropic itself is targeting $190bn-$200bn by 2028 (see here), well adrift of what was circulating in Silicon Valley just a few months ago.
• This means that 2026 revenues could be in the $60bn range, with 2027 something like $120bn.
• Anthropic has also given a much clearer view of its spending commitments, which now total $895bn, with $448bn occurring before the end of 2030 and $70bn in the next 12 months.
• This does not look unreasonable, as 2027 could be $120bn of revenue with something like $90bn in compute costs, leaving the company $30bn for other things, including, perhaps, breaking even.
• This is supported by the fact that although Anthropic was able to maintain gross margins at 85.7% despite having to give away $600m in price reductions and moves to new model variants with cheaper token costs.
• However, Q3 26 compute costs grew to 58.2% of revenue from 48.1% of revenue in Q2 26, which is where I suspect some of these discounts have been accounted for.
• For example, US GAAP allows a discount that is used as an incentive to make a sale to be accounted for as sales and marketing rather than in gross costs.
• This is commonly abused by software companies to flatter gross margins when facing competition, and I suspect, given the pressure that the valuation is coming under, there may be some of this going on here.
• This is also because a resurgent OpenAI has seen its growth reaccelerate, and its latest models are proper competition for Anthropic for the first time in a while.
• Q3 26 is also the first quarter when the much higher price that Anthropic is paying to xAI for compute at $37.7bn / GW hits the profit and loss account for the first time.
• Hence, it is going to take at least a few more quarters for Anthropic to grow its way back into profitability, but given the figures, this remains quite possible in 2027.
• Hence, the big risk remains competition, as the league tables show that there is no moat in models, meaning that a price war with OpenAI in 2027 is likely.
• RFM Research has estimates that OpenAI is in a far better position when it comes to the cost of compute than Anthropic is, meaning that it can make good money at prices where Anthropic is in the red.
• For me, this is the key risk to the Anthropic proposition as a public company, and as such I would want a large discount to the valuation to compensate me for bearing this risk.
• This is something that Anthropic and its backers will be unwilling to do, and so I suspect that the IPO will come at somewhere around $1.5tn.
• Hence, while the numbers still look fairly good even with the revenue slowdown, risks are clearly rising, and so I am still sitting this one out.