$800M for a Company That Sells Picks to the Gold Rush - Together AI's Series C Is the Whole Story

The Number That Shouldn't Surprise You (But Will)

$800 million. Series C. For a company most retail investors couldn't name last Tuesday.

Together AI closed that round on July 1, led by Aramco Ventures alongside Vista Equity, General Catalyst, Nvidia, and Salesforce Ventures. The post-money sits at $8.3 billion. Together AI builds cloud infrastructure that lets enterprises train and run AI on open-source models: the unsexy, load-bearing wall behind every AI deployment that isn't running on a closed hyperscaler. Boring? Absolutely. Lucrative? Ask the nine-firm syndicate that just wrote eight-figure cheques.

Let's Do the Maths

The Crunchbase weekly roundup covering June 27 through July 2 put Together AI at number two by deal size, behind only Joulent's $1.75 billion energy infrastructure financing. That alone tells you something: in a week that included AI software, biotech, and professional lacrosse, an open-source model infrastructure company was the second-largest ticket in the room.

At $8.3 billion on an $800 million raise, investors paid roughly 10.4x the round size for the valuation. Compare that to Databricks at $134 billion on $5.4 billion ARR (approximately 25x revenue) or Anthropic at $965 billion on revenues that remain, charitably, still climbing toward justification. Together AI's multiple looks positively parsimonious by the standards of this market. That's either a bargain or a warning that the market is beginning to price infrastructure differently from frontier labs. Probably both.

Concentration Is the Only Story

Zoom out and the picture sharpens uncomfortably. Q1 2026 saw $300 billion poured into 6,000 startups globally, an all-time quarterly record, up over 150% year-on-year, per Crunchbase. Four rounds alone (OpenAI at $122 billion, Anthropic at $30 billion, xAI at $20 billion, Waymo at $16 billion) absorbed $188 billion, or 65% of the entire quarter's global venture capital. AI as a sector captured 80% of total global venture in Q1.

Nearly 88% of AI-related startup funding in 2026 has gone to U.S.-headquartered companies, per Crunchbase data: $319 billion, the overwhelming bulk concentrated in two recipients. That's not a market. That's a pipeline with two very large nozzles and everyone else trying to hold a bucket underneath.

Charlie Munger once called diversification "protection against ignorance." The AI funding market has decided it is not ignorant. It might be right. It has been wrong before.

The Hollow Core of the New Formation Stack

Here's the detail that should give fintech and financial services investors pause. AIBD's analysis of Companies House data shows zero new SIC 64.20 (holding company) registrations in Q3 2026 to date: a 100% decline versus the prior period. A hard stop. The vehicle layer that private capital traditionally uses to warehouse AI investment positions isn't forming. The money is moving, but the corporate formation activity underpinning it in the UK has gone quiet.

In a normal cycle, funding surges create downstream company formation: holding vehicles, SPVs, advisory entities. The absence here either reflects capital routing directly through U.S. structures (consistent with that 88% concentration figure) or a structural hesitation in the financial services formation layer. Neither reading is comfortable for London's fintech ecosystem, which has pulled in $16.5 billion so far in 2026 but remains a fraction of the U.S. total.

Pre-ChatGPT Darlings, Meet the Guillotine

While infrastructure players like Together AI feast, a quieter reckoning is running in parallel. Valuations for companies that peaked before the ChatGPT moment have compressed roughly sixfold from the 2021 peak of 50x forward revenues, per Ryan Falvey of Restive Ventures: the same revenue buys 85% less valuation today than five years ago. Pre-ChatGPT unicorns without a credible AI pivot face a binary: acquisition at a steep discount, or slow fade. The assumption that a startup's engineering headcount provides a valuation floor (historically around $2 million per coder) evaporated once AI coding tools allowed far smaller teams to ship equivalent products.

The headline numbers are staggering. The footnotes are brutal.

The IPO Queue Is Getting Impatient

Anthropic filed its draft S-1 on June 1 at a $965 billion valuation, technically now the world's most valuable standalone AI startup, having overtaken OpenAI's $852 billion last private-round figure. SpaceX went public on June 12 under SPCX at $150 per share, rising to over $2 trillion in market cap in what became the largest IPO in history. Cerebras IPO'd May 14, raising $5.55 billion and gaining 68% on day one.

Databricks CEO Ali Ghodsi told Bloomberg TV in June that 2026 was a "terrible" year to IPO, citing SpaceX, OpenAI, and Anthropic. He has found a reason not to IPO in every year since 2021. At some point the market stops laughing.

OpenAI's CFO Sarah Friar has reportedly expressed concerns about the company's public-reporting readiness and cash burn rate, introducing the possibility of a 2027 delay. Polymarket traders are pricing 73% odds on a December 2026 OpenAI IPO. Three percent are betting on July 31. That's not scepticism. That's a write-off.

The Prediction

Together AI's $800 million round is not an outlier. It's the template. As foundation-model economics commoditise and margin compresses at the model layer, the infrastructure layer (the firms that run the pipes regardless of which model wins) will attract increasingly rational capital. Aramco's lead here is not accidental: energy majors understand infrastructure economics better than any Silicon Valley GP.

Expect two more $500 million-plus open-source AI infrastructure rounds before Q4. Expect Databricks to still not have IPO'd. And expect the Companies House formation data to stay cold until U.S.-domiciled holding structures stop being cheaper, faster, and 88% of the action.

The gold rush is real. The shovels are where the money is. They always were.