Tom Blomfield, cofounder of Monzo and GoCardless, is taking a leave of absence from Y Combinator to join the compute team at Anthropic, a move he announced on July 13, 2026. Blomfield has been a general partner at Y Combinator since 2023, and he now shifts from backing founders to working on the computing infrastructure that decides how fast frontier AI models can be trained and served.
His path had pointed the other way. Blomfield is best known as the cofounder of Monzo and GoCardless, two companies that helped define the modern UK fintech scene, and the obvious next step would have been another consumer product or a continued run funding startups at YC. At Anthropic he joins the compute team under cofounder Tom Brown, the company’s Chief Compute Officer. The compute function is the engine room of a frontier AI lab. It covers the chips, the data centers, the power, and the orchestration that decide how much a model can learn and how quickly it can serve users.
That function rarely draws the attention new model releases get, yet it increasingly sets the ceiling on what a lab can ship. Blomfield spent years building and scaling Monzo through banking systems, regulators, and rapid user growth, so the work of making a large and complicated machine run reliably at scale is familiar territory for him. For someone with that record, the pull of the compute problem is not hard to understand, and it helps explain why he would leave a marquee investing seat to take it on.
Blomfield has framed the move around compute becoming one of the most important problems to solve as the industry enters what he describes as the early stages of recursive self improvement in AI. The phrase refers to a scenario in which AI systems help design and improve the next generation of AI systems, compounding progress over time. In his account, if that loop is genuinely beginning, the availability of raw computing capacity becomes the decisive constraint, and working on it is among the highest leverage things a person can do right now.
That framing is Blomfield’s stated view and a motivating thesis rather than an established fact, and plenty of researchers are skeptical that the industry is anywhere near such a loop. Some see the idea as premature, and others question whether current systems improve themselves in any meaningful sense. His reasoning reads best as a bet by an experienced operator about where the field is heading. What is not in dispute is that his decision reflects a strong conviction that compute has become the chokepoint.
That conviction is the part worth watching for readers following the AI business. When a celebrated founder chooses infrastructure over models or consumer products, it hints that the scarce resource has shifted. For much of the past few years the prestige sat with the researchers shipping new models and the teams building applications on top of them. Blomfield’s move suggests the harder and more valuable problem may now be securing and scaling the compute those models depend on. If the person who built Monzo sees the frontier in the data center, that says something about where value and difficulty are now concentrating.
His arrival also extends a run of high profile hires at Anthropic through 2026. Earlier in the year, around May, Andrej Karpathy, a founding member of OpenAI and former Tesla AI director, joined the company on the pre training team. The two hires show a lab steadily pulling in well known names across different parts of the stack, from the science of training models to the infrastructure that makes training possible. As TechFundingNews reported, Blomfield’s leave from YC is the latest example of that recruiting momentum. For a company competing with much larger rivals over a small pool of talent, attracting operators and researchers of this profile is itself a form of leverage.
Some questions remain open. A leave of absence is not a permanent departure, so Blomfield could return to Y Combinator, and it is not yet clear how long he intends to stay or what exactly he will own inside the compute team. For now, one of fintech’s most accomplished builders has staked his next chapter on the view that compute is the constraint that matters most, and the market will be watching to see whether he is right.
*Disclosure: Entrelligence uses Anthropic’s Claude tooling in its editorial workflow.*
Your First 10 AI Skills
10 practical AI skills, copy-paste prompts and a 7-day plan to start using AI with confidence.
Download the guide →