Seattle-based OctoML, an ML acceleration service built on the open source Apache TVM compiler framework, raises $28M Series B, after $15M Series A in April 2020
Frederic Lardinois / TechCrunch :
Context & Ripple Effects
OctoML is scaling fast on an unusually short cadence: less than a year after its $15M Series A led by Amplify in April 2020, it has closed a $28M Series B — and by November 2021 it would follow with an $85M Series C led by Tiger Global. The company's bet is that Apache TVM, the open source compiler framework it was founded around, can be productized into a service enterprises pay for to make ML models run efficiently across different types of hardware.
The round also fits a Seattle pattern in the coverage: the city keeps producing ML infrastructure and developer-tooling companies, from XNOR.ai's device-deployment platform to Uplevel's engineering-analytics play, giving OctoML a local talent and investor base rather than a Bay Area dependency.
First-order effects
- OctoML gains the capital to move beyond research-grade compiler work and staff out the enterprise-facing optimization-and-deployment service described in its own positioning.
- Apache TVM gets a funded commercial steward, which typically accelerates feature development and gives the open source project a supported enterprise path.
Second-order effects
- The funding lands in the same window as Comet's $50M Series B for data-scientist tooling, signaling that investors are pricing the ML workflow layer broadly and pushing adjacent MLOps vendors toward competing on the optimization step, not just training and tracking.
- A compiler-driven acceleration layer sits between frameworks and silicon, so chip vendors' performance claims become testable against TVM-optimized deployments — pressure on hardware makers whose edge depends on proprietary software stacks.
Third-order effects
- If the capital cadence holds — A, B, and C inside roughly eighteen months — it points toward the ML stack consolidating into distinct paid layers, with compilation and hardware-portability emerging as a standalone market rather than a feature of cloud providers or chip companies.
The trend: ML infrastructure is splitting into independently capitalized layers, and the compiler-and-optimization tier — where OctoML plays via Apache TVM — is becoming one of them.