Wafer

Wafer is an inference provider focused on serving open models quickly. Its pitch is continual inference: the platform observes how a customer's workload behaves and keeps tuning the serving stack, covering the engine, fused kernels and traffic patterns, to improve latency, reliability and efficiency over time. The home page illustrates this with throughput rising from 230 to 400 tokens per second as tuning steps are applied, and reports more than 10 trillion tokens of inference served. The product is aimed at teams that run open weight language models in production and care about response time under load, such as those building real time agents or applications that depend on low latency completions. The site offers a hosted app, a section of customer cases, and a dedicated deployment option for teams that want reserved capacity. Customer quotes on the site focus on low latency that holds up as request volume rises, and on lower inference costs. One case describes work with DigitalOcean to make a Kimi model run faster on AMD hardware. Wafer has announced a $40 million Series A round. Pricing is not published. The site has no pricing page, and the home page invites visitors to start building through the app rather than listing rates for models or dedicated capacity.