Together AI Raises $800M at $8.3B Valuation

The money chasing AI compute keeps finding new homes. On July 1, 2026, Together AI announced an $800 million Series C at an $8.3 billion post-money valuation, one of the largest raises yet for a company whose entire business is renting out infrastructure to run open-source AI models. The round was led by Aramco Ventures, the corporate venture arm of Saudi Aramco, and the valuation marks a sharp step up from where the company sat just over a year ago.

The raise lands in the middle of an unusually busy stretch for the AI compute market — the same week SoftBank unveiled its SB Neo neocloud and days after a sharp sell-off in AI chip stocks rattled the sector. Together AI’s pitch cuts against that anxiety: it is betting that the shift toward cheaper open-weight models is durable, and that the picks-and-shovels layer beneath them is where a lasting business lives.

The numbers

Together AI last raised a $305 million Series B at a $3.3 billion valuation roughly 16 months ago. The new round therefore represents about a 2.5x increase in valuation in just over a year — steep even by the standards of the current AI cycle. The syndicate is broad: alongside lead investor Aramco Ventures, participants include Vista Equity Partners, General Catalyst, Emergence Capital, NVIDIA, March Capital, Pegatron, and S Ventures, the venture arm of security firm SentinelOne.

The company backed the valuation with traction figures. Together AI reports annual bookings above $1.15 billion and says industry-wide usage of open-source models has roughly tripled over the past year. It plans to put the new capital toward expanding public cloud capacity by a factor of 50 over the next five years, plus improvements to its training and inference stack.

NVIDIA’s presence on the cap table is worth noting. The chipmaker has been backing the very neoclouds that buy its GPUs in volume — a circularity that boosters read as ecosystem-building and skeptics read as demand engineered from the supply side. Either way, it means the company selling the accelerators, the company renting them out, and the enterprises running models on top are increasingly financially entangled.

What Together AI actually does

Together AI sits in the category the industry now calls a “neocloud” — a provider built around one job: renting GPU clusters and AI-specific infrastructure at scale, rather than offering the sprawling menu of services a general-purpose hyperscaler does. But Together AI has carved out a narrower position than a pure capacity reseller. Its core value proposition is running open-source and open-weight models — the likes of DeepSeek, Nemotron, MiniMax, and Kimi — cheaply, at scale, and with performance the company claims is comparable to or better than closed systems.

That focus is a bet on a trend we’ve tracked closely: open-source AI models are closing the gap with the best proprietary systems on a growing list of real-world tasks. As they do, the calculus for enterprises changes. A closed frontier model accessed through a metered API is convenient, but it is also a recurring cost that scales with usage and a dependency a company does not control. An open-weight model of comparable quality, run on rented infrastructure, turns that variable cost into something a business can shape — and Together AI wants to be the layer that makes running it turnkey.

The company was founded in 2022 by Vipul Ved Prakash, who previously sold social-search startup Topsy to Apple, alongside Stanford professor Percy Liang and researcher Ce Zhang. That academic pedigree is more than decoration: much of Together AI’s engineering advantage comes from inference and training optimizations — the kernel-level work of moving data and math efficiently across clusters — that determine the cost per token an operator can offer.

Why Aramco led

The identity of the lead investor is a story in itself. Aramco Ventures taking the top slot signals how sovereign and energy-linked capital is flowing into AI infrastructure, and why. The binding constraints on this buildout are increasingly power, land, and financing rather than the chips themselves — a dynamic we’ve unpacked in our look at AI data-center economics. An investor whose parent company is one of the world’s largest energy producers is positioned to understand — and potentially supply — the input that matters most as compute demand collides with the grid.

It also continues a pattern of Gulf capital moving aggressively into the AI stack, from chip supply agreements to data-center joint ventures. For Together AI, that money buys more than a valuation bump; it buys a partner with a strategic interest in seeing large-scale compute get built.

The bubble question hanging over it

Together AI’s raise arrives against a jittery backdrop. Just days earlier, memory and equipment names led a semiconductor rout, and a louder debate has taken hold over whether the enormous sums flowing into AI infrastructure will earn a return. NVIDIA’s move to backstop financing for neocloud partners — effectively helping fund the customers who buy its chips — has become a lightning rod for that debate, with critics asking whether demand is as organic as the headline numbers suggest.

Neocloud capacity is, at bottom, fungible: one operator’s GPUs look much like another’s, so the business tends to compete on price and utilization. That is the same commoditization risk that shadows every entrant in the space, from SoftBank’s SB Neo to the sovereign-backed megaprojects. Together AI’s answer is differentiation up the stack — the software that makes open models fast and cheap to serve — rather than raw capacity alone. Whether that moat holds as rivals copy the same optimizations is the central question its investors are underwriting.

What it means

Together AI’s $800 million round is a wager that the center of gravity in AI is drifting toward open models running on flexible, third-party infrastructure — and that owning the layer between the two is a durable business.

Who wins if it’s right. Enterprises gain a credible path off closed-model lock-in, with cost curves they can control. Open-weight model developers get a distribution channel that makes their work usable at scale. And Together AI captures margin on a workload — inference on open models — that is growing faster than almost anything else in the stack.

Who’s exposed if it’s wrong. The bear case is timing and commoditization. If open models plateau relative to the closed frontier, or if a wave of neocloud capacity lands into demand that has cooled, the whole compute-rental layer competes its margins away. A 2.5x valuation step-up in 16 months leaves little room for disappointment, and the same capital enthusiasm that funded this round could reverse quickly — as the recent chip sell-off and the ongoing scrutiny of AI capital spending both showed.

What to watch. First, the bookings-to-revenue conversion: $1.15 billion in bookings is a strong signal, but neocloud economics turn on utilization and cost per token, not headline commitments. Second, whether open-model adoption keeps compounding at the pace Together AI cites, or whether the closed frontier reasserts a durable quality lead. Third, the financing plumbing — with NVIDIA on the cap table and Gulf energy capital leading the round, the flows between chipmaker, cloud, and customer are worth tracking for signs that demand is being manufactured rather than met. The AI-infrastructure boom has minted a lot of paper value in a short time. Together AI is now one of the larger bets that the demand beneath it is real.