Infinity raised $15 million led by Touring Capital, TechCrunch reported on July 20, with an investor base that stands out more than the round size: individual researchers from both OpenAI and Anthropic participated personally, alongside the institutional lead. Researchers from two competing frontier labs backing the same early-stage inference infrastructure startup is a stronger signal of genuine technical conviction than most seed rounds get, since these are people who understand the inference-cost problem from the inside and have no obvious reason to co-sign a mediocre bet.
Inference cost and latency optimization has become one of the most crowded and technically demanding infrastructure categories in AI over the past year, as the industry's cost center has shifted from training runs to the ongoing, compounding cost of serving models at scale to millions of users and, increasingly, autonomous agents making multiple model calls per task. Infinity is entering a field that already includes well-funded players attacking the same problem from different angles -- this same week, storage company Weka launched a platform that caches 100% of a model's pre-calculated tokens specifically to reduce GPU load, and Google shipped Gemini 3.6 Flash with claims of up to 65% lower token costs on long-horizon agent tasks.
That's the real competitive context for Infinity's raise: inference efficiency is being attacked simultaneously at the model layer (Google, and implicitly every other frontier lab), the storage layer (Weka), and now at whatever specific technical layer Infinity is targeting -- the round doesn't disclose full technical detail, but the researcher-investor base suggests something differentiated enough to draw attention from people building competing solutions at OpenAI and Anthropic themselves.
$15 million is a modest seed check by 2026 AI-infra standards, and Infinity will need to prove out real customer traction against much better-capitalized rivals and against the frontier labs' own first-party cost optimizations, which are effectively free improvements for any customer already using their APIs.
For infra-focused seed investors, the takeaway is that researcher-led angel participation from rival labs is becoming a meaningful signal worth weighting in inference-infrastructure diligence -- it's a proxy for technical credibility that's hard to fake and harder to buy.