Together AI Raises $800M. NVIDIA and a Saudi Oil Giant Are Betting Against Closed Models.

Together AI announced Wednesday that it closed an $800 million Series C at an $8.3 billion valuation — more than double the $3.3 billion the company was worth in early 2025. The round was led by Aramco Ventures, the investment arm of Saudi Arabia's state oil company, with NVIDIA, Vista Equity Partners, General Catalyst, and Emergence Capital also writing checks. That combination of investors, a sovereign energy fund and the world's dominant chipmaker, is not a coincidence. It is a bet that the next phase of the AI buildout runs through infrastructure companies that make open-source models fast and cheap, not through the handful of labs building closed ones.

Together AI's pitch has stayed consistent since it launched: instead of building a frontier model and charging for access to it, run the best open-source models — DeepSeek, Nemotron, MiniMax, Kimi — on optimized infrastructure and sell the inference. The company says customers save between 6x and 60x compared to running equivalent workloads on proprietary APIs from OpenAI or Anthropic, with comparable or better performance on many tasks. That is a serious claim, and the business results back it up in scale terms at least: annualized bookings for the last quarter topped $1.15 billion, up from a fraction of that a year ago.

The Arms Race Has a Second Track

Every headline out of AI in 2026 has been about frontier labs raising absurd sums to train the next model — OpenAI's trillion-dollar IPO filing, Anthropic's export fights, xAI's compute hunger. Together AI's raise is a reminder that a second, quieter arms race is happening underneath that one: the fight to own the infrastructure layer that makes AI cheap enough for a normal company to actually deploy. Enterprises don't need the absolute best model for most workloads. They need something good enough, running reliably, at a price that doesn't blow up their unit economics. That is the gap Together AI is built to fill, and $1.15 billion in annualized bookings suggests the gap is real and growing.

NVIDIA's participation is the more interesting signal. NVIDIA has invested in nearly every layer of the AI stack over the past two years, but backing Together AI specifically means backing a company whose entire business model depends on commoditizing the model layer while keeping compute demand high. If enterprises shift from closed frontier APIs to open-source models running on optimized inference stacks, GPU demand doesn't shrink — it possibly grows, because the total addressable market of "companies that can afford to run AI" expands. NVIDIA doesn't need OpenAI or Anthropic to win the model war. It needs someone to win the inference war, on any chip that happens to be sitting in a data center.

Why a Saudi Oil Fund Leads This Round

Aramco Ventures leading an $800 million round for an AI infrastructure company is its own story. Sovereign wealth capital has been circling AI infrastructure for two years, but leading a round of this size signals a shift from passive participation to active positioning. Saudi Arabia has been building out data center capacity domestically and wants AI compute to be a pillar of its post-oil economy; owning a stake in the company that lets any enterprise run frontier-quality AI without needing an OpenAI-scale training budget fits that strategy directly. It's a hedge against two futures at once — one where closed labs stay dominant, and one where open models eat their lunch — with the fund positioned to profit either way as long as inference demand keeps climbing.

The Real Test Is Retention, Not the Round

Raising $800 million and hitting $1.15 billion in annualized bookings is a strong headline, but the number that will actually determine whether Together AI is a durable business is retention once open-source model quality plateaus or closed labs cut their own prices. OpenAI and Anthropic have both been dropping API prices aggressively as compute costs fall, narrowing the cost gap that Together AI's entire pitch depends on. If that gap closes, the "6x to 60x cheaper" argument gets a lot less compelling, and infrastructure companies get squeezed from both sides — undercut on price by closed labs and undifferentiated on the open-source side by competitors offering the same models. Together AI just bought itself five years of runway to prove the open-model bet is structural rather than a temporary pricing anomaly. That is the wager $800 million and an oil fund's credibility are now riding on.