Rune raises $40M to turn idle solar power into AI compute
The AI buildout's binding constraint keeps moving — off GPUs, past memory, onto the grid connection that takes years to arrive. One startup's answer: don't wait for the grid. Use the power the grid is already throwing away.
Rune raised a $40 million Series A led by Spark Capital to install modular data centers directly at solar farms, powered by electricity the plants generate but never sell. Alongside the round the company launched RELIC (Renewable Energy Linked Intelligent Compute), a compute unit that sits behind the meter at an operating generation site — no grid connection, no construction, no interconnection queue. Rune says a unit can come online in as little as six weeks and deploy in 60 minutes, and that its model cuts non-compute infrastructure costs by 85% versus a conventional AI data center. Union Square Ventures, Lowercarbon Capital, Activate Capital, Committed Capital, Timeless Partners and Logos Fund joined, taking total funding to $53.5 million.
The numbers behind the pitch are the interesting part. Solar plants curtail up to 20% of the energy they produce — power generated but never sold or delivered — an estimated 50-plus terawatt hours a year in the US alone. Rune has a RELIC installation live at a 200MW solar facility in Texas, built on an existing renewable asset with no site modifications and no grid work. "Every solar plant is a latent data center," CEO William Layden said in the announcement. Layden previously ran a division at Cube Hydro that used stranded hydro power for bitcoin mining and later worked across SoftBank Energy's solar portfolio; co-founder and CTO Varun Palivela is an ex-NUVIA chip architect whose core work fed into Qualcomm's Snapdragon X Elite. That combination — power origination plus silicon design — is why this is worth watching rather than filing away as another data center round. If a 200MW farm can host inference capacity, the deployment surface for AI compute is every operating renewable asset in the country, not just the handful of sites that can win an interconnection slot.
The OpenAI Foundation is paying to create the biomedical training data that AI labs say they can't find. The nonprofit parent of OpenAI is funding an effort called Data for Public Health, starting with a $500,000 grant to 1Day Sooner, an advocacy group for clinical trial volunteers, to buy the scientific files of bankrupt biotech companies — "biotech's lost archive," in policy analyst Ruxandra Teslo's phrase, whose idea it was. The targets are common technical documents: the back-and-forth between a company and regulators plus detailed scientific, manufacturing and safety measurements — essentially everything known about a drug, and exactly the negative-result data that never reaches a journal.
1Day Sooner's president, Josh Morrison, says nonexclusive copies might be acquired for only "a few tens of thousands of dollars" each, and that two attempts to obtain company files this year already failed when bids weren't accepted. The OpenAI Foundation says it hopes to give away $1 billion by the end of the year. The effort lands in a widening fight over who gets to buy dying companies' data — Google's bid for Spirit Airlines' records prompted outcry over a "new land grab."
The bet here is cheap: a few hundred thousand dollars against the hundreds of billions going into compute. Whether it pays off depends on something labs haven't proven yet — that regulatory filings from failed programs improve model predictions more than they improve data-volume slides.
Should bankrupt companies' scientific filings be a public asset instead of a purchasable one? Tell us in the comments.
Sources: Fast Company · Rune (Business Wire) · MIT Technology Review · Biotech's Lost Archive (IFP) · Techmeme
