Together AI Just Raised $800M at an $8.3B Valuation — Aramco Ventures Is Betting Sovereign Capital on Open-Source AI Infrastructure

An $800 Million Round, and a Valuation That Tripled in 16 Months

On July 1, 2026, Together AI closed an $800 million Series C at an $8.3 billion post-money valuation, led by Aramco Ventures — the venture arm of Saudi Arabia's state oil company, Saudi Aramco — with Nvidia, Vista Equity Partners, General Catalyst, Emergence Capital, Schneider Electric's SE Ventures, March Capital, Pegatron, Salesforce Ventures, and SentinelOne's S Ventures also participating. The new valuation is more than 2.5x the $3.3 billion mark Together AI set just 16 months earlier in its $305 million Series B. For a company whose entire pitch is running open-weight models cheaper than the closed-model labs, that is a striking amount of capital to raise in a single round.

The Number Investors Actually Care About: $1.15 Billion in Bookings

Together AI says its annual bookings now exceed $1.15 billion, and that usage of open-source models on its platform has tripled over the past year. The pitch to enterprises is concrete: run models like Llama, DeepSeek, Qwen, Mixtral, Nemotron, MiniMax, and Kimi on Together's infrastructure and cut inference costs by as much as 60x compared with closed-model APIs, while getting what the company claims is the fastest output speed among GPU-based providers — up to 2x faster than competing platforms on some models. Those are vendor-reported numbers, but the underlying trend they point to — enterprises treating the model layer as swappable commodity infrastructure rather than a single-vendor relationship — lines up with what GitHub, Apple, and OpenCode have all been building toward this year.

Why Aramco Ventures, of All Investors

The most telling detail in the round isn't the size, it's the lead investor. Abhishek Shukla, Managing Director of Prosperity7 Ventures US — Aramco Ventures' diversified venturing program — put it directly: 'Building AI infrastructure over the next decade will be the biggest infrastructure project in human history... Together has built the platform that makes open source models genuinely usable at enterprise scale.' That framing matters: Gulf sovereign and sovereign-adjacent capital has increasingly treated AI compute not as a financial bet on one lab winning, but as a strategic national resource — and a neutral, model-agnostic inference layer is a safer place to park that kind of money than a wager on any single frontier lab's next release.

The Elephant in the Room: You're Still Renting Someone Else's Cloud

Open weights do not mean no lock-in. Every workload an enterprise runs through Together AI still depends on one company's uptime, pricing decisions, and GPU allocation — the same shape of dependency as a closed-model API, just with a portable escape hatch if it goes wrong. Nvidia is both an investor in this round and the GPU supplier the entire platform runs on, which aligns incentives but also concentrates risk: a supply shock, an export-control change, or a pricing shift from Nvidia touches Together AI's entire cost structure at once. And for any organization with export-control-sensitive contracts, a state oil company's venture arm now sitting on the cap table of a core infrastructure vendor is a fact worth knowing, not a reason to panic.

What to Check Before Your Next Model-Hosting Decision

Three concrete steps for engineering leads evaluating this shift: first, benchmark the 'up to 60x cheaper' claim against your own workload, not the vendor's reference numbers — inference cost comparisons are notoriously sensitive to batch size, context length, and caching strategy. Second, confirm actual portability: because the models are open-weight, verify today, before you build on top of Together AI, that you can export and redeploy on a second provider or your own GPUs without an architecture rewrite — that portability is the real hedge against lock-in, not the 'open source' label by itself. Third, add investor and ownership structure to your vendor risk review alongside model provenance, particularly if your contracts or compliance posture already treat that kind of thing as a factor.

Bottom Line

Together AI's raise is the clearest sign yet that sovereign capital sees the AI infrastructure layer — not any single model — as the durable asset worth owning a piece of. For engineering teams, the practical takeaway isn't 'switch to open-source models' or 'avoid Saudi-backed vendors' — it's that the vendor you pick to run open-weight models is making just as many decisions on your behalf as a closed-model API would, and deserves the same level of scrutiny.