Together AI welcomes Alon Gavrielov as VP of Infrastructure Strategy

Together AI wants you to think of itself less as a chatbot API and more as a factory floor — one that churns out tokens instead of widgets. To keep that factory running at gigawatt scale, the company just brought in Alon Gavrielov as VP of Infrastructure Strategy, poaching him from Cloudflare where he spent years building out a network that eventually spanned hundreds of locations and handled over 100 million requests at peak.

That hire matters more than a typical executive shuffle because of what Together AI is actually trying to build. The company already backs over a million AI engineers and researchers, including fast-scaling names like Cursor, Decagon, and ElevenLabs. Serving that base isn't just about renting GPUs — Together AI is stacking together custom kernels, an inference layer tuned for cost efficiency, and developer tooling, then wrapping the whole thing in power and cooling planning that resembles a utility company more than a software startup.

Gavrielov's résumé fits that pivot. At Cloudflare he wasn't writing code for a CDN; he was solving capacity planning, supply chain logistics, and data center strategy at a company that had to add new locations constantly just to keep pace with traffic. Together AI is making a bet that the same muscle — figuring out where to put compute, how to source it, and how to keep it running — is now the actual bottleneck in AI, not model architecture.

It's a bet with some backing in the numbers. Together AI's pitch has always leaned on speed — fastest inference for popular open-source models — but hitting that speed at the scale its customers now demand requires exactly the kind of physical infrastructure discipline Gavrielov practiced for years. The company says it's now planning AI factories across multiple locations with gigawatt-level capacity, language usually reserved for power grids, not cloud providers.

What this hire really signals is a shift in where the competitive edge in AI is moving. Model quality gaps are shrinking, so the fight increasingly comes down to who can physically deliver compute reliably and cheaply at massive scale. Together AI just bet its future on winning that fight, not the model race.