Together AI raised $800M renting out the GPUs you're not buying
On July 1, 2026 Together AI raised an $800M Series C at $8.3B — up from $3.3B sixteen months earlier — on $1.15B of bookings and 500MW of committed compute. The neocloud layer is a bet that most teams should rent, not build. I think it's mostly right.
On July 1, 2026, Together AI — a "neocloud" that rents out Nvidia GPU clusters — raised an $800 million Series C at an $8.3 billion valuation, up from $3.3B just sixteen months earlier. Aramco Ventures led the round; Nvidia, Vista Equity Partners, and General Catalyst joined. Behind the raise: $1.15 billion in bookings last quarter and 500+ megawatts of committed compute, funding roughly 50x capacity growth over five years.
What I take from it
Strip away the valuation and the neocloud is a bet on one asymmetry: owning GPUs is a capital-and-operations problem most software teams have no business taking on. Buying the hardware is the easy part. Powering it, cooling it, replacing it on Nvidia's cadence, and — the killer — keeping it busy is a full-time infrastructure company. Most of us are not that company.
Which is why this raise reads to me as the flip side of the utilization story I keep coming back to: if the average enterprise runs its GPUs at 5%, then renting a right-sized slice by the hour beats owning a rack you idle 95% of the time. Together's whole pitch is "let someone whose job is utilization run the metal." That's a real argument, not just a financing one.
How I actually weigh rent vs build:
- Utilization is the entire argument. If I can't keep an accelerator busy, I have no business owning it — I should be renting someone else's by the hour and letting them eat the idle time.
- Lock-in hides in the tooling, not the invoice. The bill is easy to compare. The cost of coupling your pipeline to one provider's runtime, quirks, and APIs is the number that bites later. I keep the model-serving layer swappable on purpose.
- The middle layer earns its margin on convenience, and convenience is real. Neoclouds sit between you and the hyperscaler and charge for it. That spread is worth paying right up until your volume is large and steady enough that the math flips — and not one day before.
Renting isn't the unambitious choice. At 5% utilization, it's usually the correct one — and it stays correct until your workload is big and predictable enough to make a rack pay for itself.
Sources: Together AI raises $800M Series C at $8.3B valuation.
