Anthropic’s IPO and Macro
Busy giving Kilby access to TEK2day Finance these past two weeks, so I haven’t written much, but what a mess the world is. Let’s start with macro because it could pop the AI bubble once and for all.
Treasury Yields: I’ve sounded like a broken record on this, but it was not a difficult call over the past year to say Treasury yields would move higher. When you print money because you run $2-3 Trillion annual fiscal deficits, yields will move higher even if not in a perfectly straight line. HERE.
Iran War & Oil: When you squeeze supply, prices are going to move higher assuming demand is stable. My view is that Iran will continue to squeeze the supply of oil and thereby squeeze Trump ahead of the mid-term elections, thus ruining the GOP in the process and in doing so making the Dems hesitant to pick up the war fighting mantle. There is no card for the U.S. to play. Drop tactical nukes? I think Iran would welcome that. There is no scenario in which the U.S. wins anything. Iran will continue to choke off oil supply and will push oil above $100 by October. This will drive Treasury yields higher, even if Treasury Secretary Bessent says the U.S. will increase Treasury buybacks at the long-end of the curve. Bessent is doing that because there is upward pressure on yields and it will persist: HERE
Japan: The Yen is weak and will remain weak versus the Dollar until Japan addresses the Yen at the policy level. If Japan continues to do nothing, the U.S. will be back in Japan with another bailout. One bailout will not fix anything nor will two or three or four. HERE
Anthropic IPO: “GPUs are the new Dollars.” “Tokens are the new Dollars.” These comments are actual comments I’ve come across and they feel like 1999 all over again. GPUs will remain scarce until they are not. The numbers I hear on Anthropic are $85 billion annualized run rate (ARR) now, and north of $100 billion ARR exiting 2026, and more than $400 billion ARR exiting 2027. What does that translate into actual full year revenue I am not sure. Will Anthropic 4x Revenue from YE 2026 to YE 2027? See below:
Outside of the macro elements I have outlined, much depends on the uptake of agentic AI, the open source threat, and access to compute.
I use an agentic platform to write code - Claude Code. I’ve also used Codex (OpenAI), Qwen Code (Alibaba) and Kimi Code (Moonshot AI). I believe that Qwen Code with Qwen 3.8 Max is the best value model/harness on the market for writing software code. Qwen and Kimi and GLM and DeepSeek are real threats to Anthropic and OpenAI, especially with Enterprise customers as they wish to control the model weights and to not expose their IP to the model companies. The Enterprise can protect its IP by running models locally. You can’t run Anthropic and OpenAI locally.
Software development is different from other functions as it is verifiable and therefore lends itself well to automation. I’m not sure that agentic AI will enjoy the uptake it has enjoyed within the Software Development function across other disciplines. Other disciplines are not as easily automated as they are generally not as verifiable as Software Development. Two verticals that come to mind are Investment Management and Corporate Finance given that numbers are verifiable (which is why Kilby will roll out a Corporate Finance plugin later in Q4, probably with Netsuite, to augment our Investment Management capabilities).
Add to the above the fact that most organizations are not AI-ready. Their data is not prepped, they are not staffed correctly, they can’t run AI projects at scale if they wanted to. If oil moves back above $100, if the 10YR moves above 5%, no Enterprise will be taking on AI projects and data center finance will be even more difficult to execute.
My guess is that the majority of Anthropic’s 2027 growth will come from Meta, Google, Amazon, Microsoft and Cursor and that the rate of growth in 2027 and 2028 will slow from 2025 and 2026, which was 10x in each of the two earlier periods.