On Tuesday, Anthropic provided an IPO range of $70-$78 per share, valuing the business at up to $1.4 trillion based on the fully diluted share count. The number is well below the $2 trillion valuation that the firm eyed back in September. Pricing is expected to occur on Nov. 18.
Why it matters: We see Anthropic’s IPO range, and downward revision from September targets, as a reflection of the current market sentiment vis-à-vis artificial intelligence, not the firm’s fundamentals. The deal is being priced during a period of high market volatility, with investors increasingly souring on the AI trade.
The bottom line: Our fair value estimate for narrow-moat Anthropic is $92.0 per share, equal to $1.7 trillion in equity value. Using the midpoint of Anthropic’s range, we believe there is 24% upside to the firm’s shares, and we view the stock as undervalued.
Bulls say: Our bull case for Anthropic stock, which envisions rapid adoption of Claude and material share gains against rival model builders such as OpenAI, has a fair value of $188, more than twice as high as our base case, implying more than 150% upside to the IPO range midpoint.
Bears say: Our bear case, which factors in not only slower enterprise adoption but also robust competition stemming from the open-weight model providers, has a fair value for the stock of $25, implying a 66% downside to the midpoint of the IPO price range.
As part of its IPO documentation, Anthropic has made the development capabilities of its AI models to improve themselves, a process known as recursive self-improvement, a clear part of the investment case for the firm.
We are seeing clear progress on this front for the firm. First, Claude, Anthropic’s AI model, now writes more than 90% of contributions to the firm’s codebase. Second, Claude now leads around a quarter of its model research tasks, up from around 1% in March of this year.
Third, Anthropic’s amended prospectus claims that the firm’s internal estimates indicate that the firm may soon meet its own threshold for accelerated, AI-driven progress.
We have some corroboration of this acceleration, with METR, a safety-focused third-party group that reviewed Anthropic’s internal data, estimating this month that Anthropic’s pace of AI progress has roughly doubled over the past year, with METR attributing loosely 80% of this acceleration to AI itself.
While investors may dismiss these claims as dubious or overblown, we are seeing outside AI researchers take them seriously, and we do believe that they merit further attention.
Outside AI researchers now see an increasing likelihood that AI can begin outperforming Anthropic’s best human researchers sometime in 2027, which would dramatically accelerate the firm’s product capability unlock and budget capture among enterprises.
We flag this “fast takeoff” scenario, as it further widens the range of outcomes in both directions. On the upside, if the claims made by outside researchers and METR prove accurate and model improvements generalize beyond certain domains, including coding, then our base case will likely prove far too conservative.
We could see an argument that even our bull case, which values the firm’s equity at $3.4 trillion, would be too conservative under a true recursive self-improvement loop scenario.
On the downside, however, a misaligned “super AI researcher” may justify an industrywide pause, leading to slowdowns in both model capability improvements and enterprise AI adoption while…
