Corporate venture arms are writing deep tech's first purchase order

An offtake for a plant that does not exist is not revenue, and Google says something close to that itself, acknowledging that fusion's commercial success is not guaranteed ^[3]. What the 200 MW agreement hands Commonwealth Fusion is a named counterparty for a generating asset that has yet to produce an electron ^[1][2]. My read, and the source does not price it, is that this is worth more to CFS than the same value arriving as equity, because money buys another year of engineering while a contract changes who else will take the Virginia project seriously.

Run the source's own roll call, though, and the pattern is thinner than the framing suggests: across the eight corporate relationships and programmes it names, exactly one carries a contracted purchase of the startup's output ^[15]. The rest supply premises, proximity and introductions. Schneider Electric's accelerator connects young companies to technical experts, business units and potential customers ^[13], which is access to a buying decision rather than the decision itself, and Siemens Energy Ventures is still at the stage of looking at climate technologies across generation, transport, storage and conversion ^[14].

RWE's EUR 25m into Proxima Fusion, which the source dates to July 2026 and which I would want set against RWE's own disclosure before treating the sequence as settled ^[4], is in substance a claim on Gundremmingen, a former nuclear site whose existing infrastructure could support a commercial magnetic-fusion plant and whose regulatory history is itself part of what the utility contributes ^[5]. Founders remember the first purchase order better than the first round because investment buys development time and a customer puts the technology where it has to work ^[19]; a site with permitting scar tissue sits somewhere between the two.

The sequencing in the smaller deals is the part worth copying. ABB invested in OKTO GRID after the Danish startup had already built a pilot for monitoring ageing electrical infrastructure ^[12], so the cheque priced a demonstrated result rather than a promise, and SAP's purchase of Prior Labs, a young specialist in tabular foundation models, is a different instrument again, an acquisition rather than a customer-led investment ^[11]. Amazon's version splits the difference: its Industrial Innovation Fund backed Agility Robotics ^[6], then Amazon said it would test the bipedal Digit at its robotics research site, starting with moving empty totes ^[7]. That is the smallest unit of warehouse work you can hand a robot, and the most informative, because technical progress still has to survive installation, safety reviews and workplace routines ^[18]. Agility, incidentally, is the only startup here sitting in two corporate portfolios ^[16].

The claim being made for all of this is that a factory, a grid connection or a procurement introduction can outweigh a larger cheque from an investor with no route into the industry ^[17]. Reasonable, and testable. The thesis fails if the access turns out to be free: if Agility's next order comes from a warehouse Amazon does not own, and CFS's next round prices like any other fusion round, then the test site and the offtake were courtesy rather than consideration, and the bigger cheque was the better deal all along.