Protocol Labs

BitRobot Network advances embodied AI through real-world robotics data collected at scale, using crypto-powered subnets to run robotic deployments and generate diverse data across the embodied AI stack. 🔗 https://lnkd.in/enz3sRfk BitRobot Foundation

Ahead of COP17, Protocol Labs Research & Development, Funding the Commons, and David Dao are convening a workshop on AI and decision sovereignty for nature-related public goods. As AI becomes more central to biodiversity governance, questions of authority, equity, and decision sovereignty are becoming increasingly important. Researchers, technologists, funders, and policymakers will come together to help develop a joint declaration and policy brief. Learn more and register: https://lnkd.in/efYHHG8y

In a San Francisco Chronicle front page story about the PRIMA retinal implant, our Medical Director for Vision Dr. Frank Brodie, MD MBA discusses how, in clinical trials, it helped restore meaningful vision to people living with geographic atrophy due to age-related macular degeneration. As Jason Menzo, CEO of the Foundation Fighting Blindness, shared: “There … is hope. Just knowing there’s something out there, just the promise of the technology, whether an individual even gets the device, is lifting people’s spirits.”

AI has become woven into the fabric of modern life in remarkably little time, with increasing utilization reflecting its entrenchment. However, every innovation has a cost, and AI’s is energy, as noted by Lacey-Ann Wisdom. Training OpenAI’s GPT-4 is estimated to have used enough electricity to power the city of San Francisco for three days, and every AI prompt adds to growing energy demand. As AI adoption accelerates, energy is emerging as one of the defining investment themes alongside it. Learn more. 👇

AI's hidden cost isn't the models. It's the electricity. ChatGPT hit 1B monthly users in May 2026 and every single query burns energy. At PL Capital we don't see a compute crisis coming. We see an energy one. Training GPT-4 alone used ~50 GWh of power. That's enough to run San Francisco for 3 days. And that's before a single user types a prompt. Here's the twist: training is now mostly a fixed cost. Inference, the point at which you ask a question, now eats 80-90% of AI's total compute power. Usage, not training, is the real energy sink. A typical AI data center now uses as much power as 100,000 homes. New ones under construction? Built to consume up to 20x more than today's centers. Data centers are already ~1.5% of global electricity use. By 2030, they could drive 20%+ of all demand growth in the US and Europe The grid can't keep up. New data centers wait up to 10 YEARS just to connect to power. -> Northern Virginia: 7-year queue -> Netherlands: 10-year average -> Dublin: literally freezing new connections Even the hardware to fix this is stuck. Transformer backlogs: 2.5–3 years Gas turbine backlogs: 3+ years GE Vernova's order backlog: ~100 GW Big Tech's workaround (investing in their own plants) is now bottlenecked too. Some ideas being tested to solve this: ->Space-based data centers (free solar, natural cooling) — Starcloud, Axiom Space + IBM already testing prototypes on the ISS ->Decentralized solar networks like Glow International Inc. & Daylight funding new solar farms & local solar networks via crypto incentives -> Fusion: Helion just raised $465M at a $15.5B valuation Meanwhile: only 2% of all equity funding into energy startups has an AI focus. Compute is getting billions. Power is getting pennies. The market is pricing AI's compute layer and ignoring its power layer entirely. AI didn't just create a new kind of software, it created a new kind of infrastructure demand. We believe the next big AI investment might not be a chip or a model company. It might be an energy company solving AI's problems. Check out our full thesis below:

With a $500M war chest at a $2.5B valuation, Flourish is on a mission to crack the brain’s core algorithm and usher in an era of low-power general intelligence inspired by biology. If anyone can pull it off, it’s this team. Proud to be an early backer. Read the full WIRED announcement: https://lnkd.in/enRHq-8P

“We found that a single neuron can solve MNIST.” Dr. Konrad Kording's lab built a computational model of a single neuron that can recognize handwritten numbers — a hint that neurons can do more than we assumed. Full conversation on the Juan Benet Podcast >>> https://lnkd.in/eSXDpKHs

Recording from 100 neurons at once used to be heroic. Now one device reads thousands, and that number doubles every six years. Dr. Konrad Kording explains how new technologies are turning the brain into something scientists can observe — and even “play like a piano.” Full conversation with Konrad Kording on the Juan Benet Podcast >>> https://lnkd.in/eSXDpKHs