推論スタートアップのInfinityがTouring CapitalやOpenAI・Anthropic研究者から1500万ドルを調達Inference startup Infinity raises $15M from Touring Capital, OpenAI and Anthropic researchers

Infinity, a startup focused on optimizing artificial intelligence inference, has raised $15 million in a seed round, according to a report from TechCrunch. The financing was backed by Touring Capital along with individual researchers affiliated with OpenAI and Anthropic. The involvement of technical staff from two of the most prominent AI labs suggests that the challenge of running models efficiently, rather than simply training larger ones, is drawing attention from inside the organizations building frontier systems.

Inference refers to the stage at which a trained model actually generates outputs, such as answering a query, producing code, or returning a classification. It stands in contrast to training, the compute-intensive process of building a model in the first place. While training tends to attract headlines because of its enormous one-time costs, inference is the recurring expense that accumulates every time a model is used in production. For companies deploying AI at scale, inference can become the dominant line item, which is why efficiency at this layer has grown into a distinct area of engineering and investment.

The demand pressure is straightforward. As chatbots, coding assistants, and agent-based applications handle rising volumes of requests, the cost and latency of each response directly affect margins and user experience. Newer reasoning models, which generate long chains of intermediate tokens before producing a final answer, consume substantially more compute per query than earlier systems. That trend has intensified interest in techniques that reduce cost without materially degrading quality, including quantization, speculative decoding, continuous batching, and optimization of the key-value cache that models rely on during generation.

TechCrunch's report does not detail the specifics of Infinity's technical approach, and the company appears to be at an early stage typical of a seed-funded venture. What is clear is that it is positioning itself within a crowded and fast-moving segment. A number of well-capitalized companies already compete on inference, including Together AI, Fireworks AI, Baseten, and Groq, the last of which designs custom chips it calls language processing units to accelerate model serving. Hardware challengers such as Cerebras and SambaNova are pursuing similar goals with alternative architectures, while open-source frameworks like vLLM and NVIDIA's TensorRT-LLM have become common building blocks for teams that serve models themselves. Against that backdrop, a new entrant must demonstrate a meaningful advantage in throughput, latency, cost, or ease of deployment.

The participation of OpenAI and Anthropic researchers is notable for reasons beyond capital. Angel investment from working scientists at leading labs can serve as a form of technical validation, signaling that people close to the underlying problems see promise in a given approach. It is worth emphasizing, however, that such backing typically reflects individual decisions rather than any institutional endorsement from the labs themselves, and the size of these contributions within a $15 million round is generally modest. Touring Capital, as the named backer, is associated with investments in AI infrastructure and enterprise software, which aligns with Infinity's stated focus.

The broader context is an industry increasingly attentive to the economics of deploying AI rather than only the frontier of model capability. Much of 2024 and 2025 saw heavy spending on GPUs and data center capacity, and the constraint of chip availability has pushed both startups and large providers to extract more performance from existing hardware. Efficiency gains at the inference layer are one way to ease that pressure, and they are likely to remain commercially relevant regardless of how quickly raw compute supply expands. Enterprises evaluating where to run workloads, whether through hyperscale cloud providers, specialized serving platforms, or their own infrastructure, are weighing these trade-offs more carefully as AI moves from experimentation into sustained production use.

For now, the funding gives Infinity resources to build out its product and team, but the company will need to prove its technology in a market where incumbents and open-source tools are advancing quickly. The round is best read as an early signal of investor and researcher confidence in the inference-optimization thesis rather than evidence of a settled outcome. How Infinity differentiates itself, and whether it can convert technical claims into measurable savings for customers, will determine its trajectory from here.