Aramco Ventures leads Together AI's $800M round, and the lead is the story

Together AI has raised an $800M Series C at an $8.3bn post-money valuation, led by Aramco Ventures. The headline is the price, 2.5x up from a $3.3bn mark in February 2025, but the structural story is who is holding the pen. A Gulf sovereign-adjacent energy fund is now setting terms on US neocloud infrastructure (specialist AI-focused cloud, distinct from AWS and Azure), and this is the second billion-dollar inference round in a fortnight.

What was actually announced. Together AI, which runs an inference platform (the compute layer that serves already-trained AI models to end users) optimised for open-weight models like Llama, Mistral and DeepSeek, closed $800M at $8.3bn post-money on 1 July. Aramco Ventures, the strategic arm of Saudi Aramco, led. The cap table (the list of who owns what) also includes Nvidia, Vista Equity Partners, General Catalyst, Salesforce Ventures, Emergence Capital, March Capital, Lux Capital, PSP Partners, SE Ventures (Schneider Electric), S Ventures (SentinelOne) and Pegatron. Reuters puts total funding to date at around $1.3bn.

The company's own announcement notes something the coverage mostly buried: alongside the $800M in equity, investors have "committed over 500 MW of compute capacity to be capitalized independently." That is a separate infrastructure financing sitting behind the round, and 500 MW is a serious number — comparable to a mid-sized hyperscaler data-centre campus. The equity round is $800M. The whole financing is larger, and structured.

Aramco leading is not colour, it is the map. Strategic investors participate in growth rounds all the time; leading a round is different. The lead sets the price, takes the largest cheque, negotiates information rights, and typically anchors governance. When the entity doing that is a state-affiliated Saudi energy fund, the question is what US neocloud pricing power looks like when Gulf sovereign capital is the marginal price-setter rather than the tag-along.

This is the MGX pattern extending. MGX, Abu Dhabi's AI vehicle, has taken positions in Anthropic and OpenAI over the past eighteen months. Those were minority strategic cheques into rounds priced by US venture. Aramco leading Together is a step further along the same curve: Gulf capital moving from participant to lead in the layer of US AI infrastructure that actually meters compute to enterprises.

The energy-industrial logic is not subtle. Aramco runs cash flows the size of small states and needs somewhere to deploy them at scale; AI inference infrastructure consumes electricity at data-centre scale and is, at bottom, an energy-conversion business. The 500 MW commitment sitting alongside the equity round is exactly the shape you would expect if the strategic view is "we sell the energy, we own the platform that burns it."

The revenue multiple is load-bearing. Reported annualised bookings are $1.15bn, implying roughly 7.2x forward bookings at the $8.3bn post-money. For a growth-stage software company that number is unremarkable. For an infrastructure business with GPU rental as the dominant COGS (cost of goods sold — the direct cost of delivering the service), it depends entirely on gross margin, and gross margin is where Together's pitch gets interesting.

Together's marketing claim is that its inference runs at one-fifth to one-seventh the cost of closed-model alternatives like GPT-4o or Claude. That is a volume claim, not a margin claim. If the pricing wedge is being funded by passing GPU costs through at near-zero markup to win share, the $1.15bn bookings number flatters the underlying economics. Vista Equity, which built its franchise on high-margin enterprise software buyouts, entering at 7x bookings on a business whose COGS floats with Nvidia's price list is the more curious signature on the term sheet.

Bookings, importantly, are not revenue. Annualised bookings are the run-rate of what customers have committed to spend; recognised revenue can lag materially, particularly on multi-year enterprise contracts. The 7x figure is the generous read.

Inference is being financed as its own asset class. Two weeks ago Baseten closed $1.5bn at $13bn. Together closes $800M at $8.3bn. Same layer, same pitch, same fortnight — roughly $2.3bn of fresh equity into $21bn of combined valuation for pure inference-optimisation businesses running on open weights.

The inference economics frame predicted this. As training-cost curves flatten and open-weight models close the capability gap with the frontier labs, the margin opportunity shifts from training the models to serving them efficiently. Capital has spotted the shift and is pricing the layer accordingly. Whether it is pricing the layer correctly is a different question — there are at least four well-funded competitors (Baseten, Fireworks, Cerebras, Modal) making near-identical claims. Capital concentration in a layer does not resolve competitive differentiation. Sometimes it accelerates the price compression that eventually removes the margin everyone was chasing.

The cap table concentrates a conflicted-buyer problem. Nvidia holds equity and sells the GPUs Together deploys. Salesforce Ventures holds equity and represents a major enterprise buyer class. Aramco holds equity and is a plausible offtake customer for hyperscale data-centre capacity, as well as, via the 500 MW commitment, effectively a landlord. Pegatron holds equity and manufactures the hardware.

Each relationship is defensible individually. Together, they mean the entities negotiating Together's supply contracts, enterprise deals, and infrastructure leases sit on the same cap table as the company negotiating them. I flagged the same pattern on Palantir–Nvidia's Nemotron arrangement earlier this year; Together's table is a denser version of it. The governance question is whether procurement decisions are being made at arm's length or through the equity relationship. There is no way to tell from a funding announcement.

Pegatron is the odd name. A Taiwanese contract manufacturer showing up in a software inference round is either a signal of upcoming custom inference appliances (on-premise Together boxes for enterprises that cannot send data to a public cloud), or Taiwan-capital positioning into US AI infrastructure ahead of tighter export controls. Both are structurally interesting; neither is in the press release.

What to watch. Gross margin disclosure, if it ever comes — a real number, not a bookings figure. Whether the 500 MW commitment materialises as Aramco-sited capacity (in which case the vertical integration is explicit) or gets spread across US colocation providers. Whether MGX or PIF (Saudi Arabia's Public Investment Fund) shows up leading the next US neocloud round; if they do, the pattern is confirmed and the pricing power question stops being theoretical.

The equity round is $800M. The story is who wrote the biggest cheque, and what else they committed alongside it.

Footnotes

¶