Together AI Pursues Gigawatt-Scale AI Factories

This hire marks Together AI moving from renting and routing other people’s GPUs to designing the kind of power, hardware, and site strategy needed to win very large enterprise and lab contracts. That means less emphasis on lightweight developer cloud workflows alone, and more emphasis on building data center capacity, supply relationships, and reliable long term deployments that look closer to CoreWeave, Lambda, and Crusoe than a typical model API startup.1 2 3 4 5 6

• Alon Gavrielov’s background is directly relevant to this shift. Together AI said in February 2026 that he joined as VP of Infrastructure Strategy, and Data Center Dynamics describes his Cloudflare role as managing global infrastructure growth, planning, hardware partnerships, site selection, logistics, and real estate across 250 plus locations. That is operating muscle for footprint expansion, not just product packaging.5 6
• The scale target is large enough that infrastructure strategy becomes a core company function. Together AI said in its July 1, 2026 Series C announcement that it had secured over 500 MW of compute capacity commitments, while its company profile says capacity could grow roughly 50 fold over five years. At that size, the hard problems are power access, cooling, financing, procurement, and uptime guarantees for big customers.1 5
• The closest comparables are neoclouds graduating into AI factory builders. Crusoe is building the first Stargate data center in Abilene with 700 plus MW expected live by December 2026, while Lambda shifted away from smaller self serve products toward multi year supercluster contracts with NVIDIA and Microsoft. The pattern is that scale winners stop selling only convenience and start selling reserved capacity at industrial scale.2 3 4

Going forward, Together AI is likely to look more like a hybrid of an open model platform and an AI infrastructure prime contractor. If it executes, the company can use its software layer to pull in startups early, then graduate the biggest customers onto dedicated clusters and sovereign or enterprise deployments that carry much larger contract values.1 2 3 4