Together AI's $800M Series C at $8.3B — Neocloud Infrastructure and the Open-Source AI Shift

Vipul Ved Prakash sold his last company to Apple for $200 million. His next bet is bigger: a cloud built not for the hyperscaler giants, but for the companies that cannot afford them.

The San Francisco company operates in a category that barely existed three years ago, the "neocloud." It rents NVIDIA GPU clusters and provides inference infrastructure optimized for open-source models. Its customers, Cursor, Cognition and Decagon, report cost savings of six to sixty times compared to running the same workloads on OpenAI or Anthropic APIs.

The company was founded in 2022 by Vipul Ved Prakash, who built the social media search platform Topsy and sold it to Apple for $200 million in 2013, together with Stanford professor Percy Liang and ETH Zurich researcher Ce Zhang.

The Round That Changes the Category

The numbers alone tell the story. It raised $800 million in a Series C led by Aramco Ventures, with participation from Vista Equity Partners, General Catalyst, Emergence Capital, NVIDIA, March Capital, Pegatron, SentinelOne's S Ventures, and Schneider Electric's SE Ventures. The total raised to date is approximately $1.3 billion.

The cap table composition matters more than the round size. Aramco Ventures, the $7 billion venture arm of Saudi Aramco, leading the round means Middle Eastern sovereign wealth has identified AI compute infrastructure as a strategic asset class. Sand Hill Road chases software multiples. Energy capital bets on compute demand the way it once bet on oil demand.

NVIDIA's participation is equally telling. The chipmaker does not invest passively in every customer that rents its GPUs. When NVIDIA writes a check, it is placing a bet on the distribution channel through which its hardware reaches the fastest-growing segment of demand. It is not a customer. It is a strategic pipeline.

Vista Equity Partners, a $100 billion software-focused private equity firm, adds a third signal: neoclouds have graduated from early-stage experiments to infrastructure assets with the recurring-revenue profile of mature enterprise software. SE Ventures, the corporate venture arm of Schneider Electric, ties the round directly to the energy-AI nexus: more efficient inference means less energy per workload, and Schneider wants equity exposure to that efficiency curve.

Open-Source AI Hits an Inflection Point

The engine behind its growth is the sharp deflation in AI inference costs. Enterprise token costs fell 67% year-over-year in Q1 2026, according to data from AI.cc's infrastructure report analyzing 2.4 billion API calls across 8,000 accounts. Open-source and open-weight models now capture 38% of enterprise token volume, up from 11% a year earlier.

It claims annual bookings crossed $1.15 billion as of its most recent quarter. The company serves thousands of paying customers. Its platform combines a token-based inference API, roughly 30 to 40 percent of revenue, with classic GPU rental. Unlike CoreWeave, which relies almost entirely on colocation, it is increasingly building its own data centers. Maryland went live in July 2025. Memphis is in preparation. It has secured commitments for more than 500 MW of additional compute capacity, to be financed separately by new investors, with a target of 50x infrastructure expansion over five years.

The Neocloud Sector Becomes an Asset Class

The company is not an isolated story. Over the last 90 days, the neocloud sector has consolidated into a recognizable category with three of the largest AI infrastructure rounds in history closing within weeks of each other.

At a combined $1.65 billion in new capital across three neocloud companies in 30 days (or $3.15 billion including Baseten), the segment is absorbing more institutional capital than most growth-stage software verticals. The common thread: each builds GPU infrastructure optimized for a specific niche. It targets open-weight models. Upscale focuses on enterprise SLAs for inference. TensorWave differentiates on AMD hardware. Baseten operates the largest independent inference cloud.

The market is segmenting before it scales. A new infrastructure category usually consolidates around a few winners after a shakeout. Here, capital is flowing in before the shakeout, because investors are betting the total addressable market is large enough to sustain multiple independent players, each with a distinct hardware or software moat, for years.

Where the Risk Sits

Three things could break the neocloud thesis.

First, hyperscaler response. AWS, Azure, and Google Cloud have not yet launched dedicated open-source AI cloud products with aggressive pricing. If they do, and they have the balance sheets to sustain a price war, the neocloud margin structure compresses overnight. Its current ~45% gross margins depend on the gap between what hyperscalers charge for general-purpose compute and what neoclouds charge for AI-optimized compute. A dedicated hyperscaler AI product would close that gap.

Second, frontier model pricing. If GPT-5 or Claude 4 delivers a capability jump large enough to justify premium pricing, the economic calculus that drives enterprises to open-source models weakens. Its thesis rests on open-source models reaching and maintaining production parity with closed frontier models. That parity is not guaranteed.

Third, capital intensity. It plans to 50x its infrastructure footprint. That requires access to debt markets at scale. The company has separated compute financing from equity: the 500 MW commitment is backed by new investors, not the Series C proceeds. But the execution risk of building gigawatt-scale AI infrastructure in a supply-constrained market for transformers, cooling, and grid interconnection is real.

What happens to the neocloud market a year from now?

Development scenarios