Saudi Aramco's corporate venture arm led an $800M round in Together AI — a signal that sovereign state capital is now pricing open source AI compute as a strategic infrastructure category.
The story today is less the dollar figure than who wrote the check. Together AI, the open-source AI cloud startup, has closed an $800 million Series C at an $8.3 billion valuation, with Saudi Aramco's corporate venture arm leading the round (SiliconANGLE). That combination, sovereign oil money backing a community-built AI infrastructure layer at near-hyperscaler prices, is the structural signal.
Together AI runs what the industry calls a neocloud: a public cloud platform purpose-built to host and serve open-source AI models, distinct from the general-purpose hyperscalers (AWS, Azure, Google Cloud) that host them as one workload among many (TechCrunch). The stack sits on Nvidia GPUs and bundles a proprietary engine called ATLAS, which uses speculative decoding, a technique where a small, fast "draft" model pre-generates tokens that a larger target model then accepts or rejects, and which Together AI claims speeds up certain inference workloads by up to 400% and gives its serverless tier roughly 2x the throughput of the fastest alternative in the market (SiliconANGLE).
The company sells that capability in four shapes: a serverless inference endpoint for low-friction, pay-as-you-go access; two dedicated-inference tiers for reserved capacity, reliability, and customization; and a Batch Inference service priced at up to 50% below standard rates, aimed at asynchronous jobs like evaluation sweeps or dataset enrichment (TechCrunch). For training, the same platform can stitch together clusters of up to thousands of Nvidia GPUs for fine-tuning open-source models, scheduled through Kubernetes or Slurm (SiliconANGLE).
Aramco Ventures, the corporate venture arm of state-owned oil major Saudi Aramco, led the round, with Nvidia, Vista Equity Partners, and General Catalyst joining (MarketScreener). That a state-owned oil major is now leading an open-source AI infrastructure round at neocloud-scale prices is itself a signal of how the category is being priced: less as developer tooling, more as a strategic compute layer.
The $8.3 billion valuation puts Together AI in the same neighborhood as CoreWeave, Lambda, and Crusoe, the neoclouds that began as crypto-mining operations and repositioned as AI compute providers, and which have since traded at multi-billion-dollar multiples on the strength of GPU-rich balance sheets (CityBiz). Together AI's pitch is narrower: it bets that the workload itself, running community-built models at scale rather than just renting GPUs, is the defensible layer. The new capital is supposed to fund exactly that: more clusters, more ATLAS engineering, and broader coverage of the open-source model catalog (Entarabi).
The framing has limits. Together AI's performance claims originate with the company itself, and the public source set does not include independent benchmarks that confirm the 2x or 400% figures (SiliconANGLE). The Nvidia-chip dependency is structural: the platform runs on Nvidia hardware, which means the economics of the open-source AI cloud and the economics of Nvidia's hardware cycle move together. A reader taking the open-source framing seriously should hold benchmark skepticism and supplier concentration in mind at once.
The next legible test is execution against the open-source model catalog: whether the ATLAS-driven stack can keep pace as community models grow larger, more multimodal, and more expensive to serve, and whether more sovereign or strategic capital follows Aramco Ventures into the category before the next round resets the valuation.