PLUS: Claude's new chemistry skills, turmoil at Meta's AI unit, and a police AI evidence scandal
Happy reading
Peter Thiel is backing a bold new venture to take AI data centers offshore. A startup has raised $140 million to build wave-powered, floating platforms that can run AI models completely off-grid in the middle of the ocean.
This creative approach aims to sidestep the immense energy and land constraints throttling the industry's growth. Could deploying AI compute at sea truly unlock the ocean as a vast, untapped power source for the future of the industry?
In today’s Next in AI:
Peter Thiel’s floating AI data centers
Claude’s new chemistry analysis skills
Turmoil inside Meta’s AI unit
A police AI evidence scandal in the UK
The Floating AI Brains
Next in AI: A startup called Panthalassa has raised $140 million in a round led by Peter Thiel to build wave-powered, floating data centers. The goal is to run AI models completely off-grid in the open ocean, sidestepping the energy crunch on land.
Explained:
The platform is an untethered, lollipop-shaped buoy that generates its own electricity. As waves lift the buoy, the motion drives an internal turbine, a design that solves the cost and maintenance issues of subsea power cables that plagued earlier wave-energy projects. You can see how the system works in this recent breakdown.
Instead of selling electricity, Panthalassa is selling AI compute. The power is used on-site to run AI inference chips, while a Starlink connection handles all communication with the shore, neatly sidestepping the grid entirely.
The company's most ambitious claim is that it can eventually achieve costs as low as $0.02 per kWh for generating power. However, the reliance on satellite internet likely limits its use to AI inference tasks rather than the data-heavy process of training new models.
Why It Matters: This approach offers a creative way to bypass the immense energy and land constraints currently throttling AI's growth. If successful, deploying AI compute at sea could unlock the ocean as a vast, untapped power source for the future of the industry.
Claude the Chemist
Next in AI: Anthropic has demonstrated its Claude Opus 4.7 model can perform complex chemistry analysis, matching or even outperforming specialized software in a major leap for AI in scientific research.
Explained:
Claude accurately predicted molecular structures from NMR spectra, proving more precise than established tools on key hydrogen analysis metrics.
The model also worked in reverse, successfully identifying molecules from their spectral data alone—a difficult task that specialized software typically leaves to human chemists.
This performance comes from a general-purpose model without chemistry-specific fine-tuning, demonstrating the power of large models to tackle highly specialized scientific problems.
Why It Matters: This signals that general AI models can begin to automate and accelerate highly specialized scientific workflows, saving researchers significant time. The demonstration opens the door for AI to become a genuine partner in research, moving beyond data analysis to assist in complex scientific discovery.
Meta's AI Gulag
Next in AI: Meta’s aggressive push to dominate the AI space is facing a significant internal revolt. Engineers in its new 6,500-person Applied AI unit report being drafted into tedious work, sparking widespread discontent and raising questions about the company's strategy.
Explained:
Engineers reassigned to the unit describe their new tasks—like generating puzzles to test models—as soul-crushing drudgework, with one employee comparing the environment to a "gulag" where they feel they have "zero purpose."
Frustration boiled over when an employee interrupted a company-wide livestream, launching an expletive-filled rant and asking presenters to tell a specific Meta AI executive he’s a “piece of shit.”
The turmoil extends beyond a single team, reflecting broader morale issues from recent restructuring and a controversial plan to monitor employee activity, which prompted a petition against employee monitoring signed by over 1,600 workers.
Why It Matters: This internal backlash reveals the intense human pressure and cultural challenges hidden behind the AI arms race. It underscores the critical test for tech leaders to balance the urgent demand for AI progress with the need to keep their most talented people engaged in meaningful work.
AI in the Evidence Room
Next in AI: A UK police officer is facing a criminal investigation for allegedly using AI to create evidence, marking the first known case of its kind and spotlighting the high-stakes risks of deploying AI in the justice system.
Explained:
The Derbyshire Constabulary officer was removed from frontline duties while being investigated for perverting the course of justice by allegedly using AI systems to generate evidential material for multiple cases.
This incident comes just as the UK government launches PoliceAI, a new national unit designed to help forces safely adopt AI tools to speed up investigations and summarize digital evidence.
The investigation follows prior warnings from UK police leadership, who cautioned forces against using generative AI for preparing court statements due to significant concerns over accuracy and reliability.
Why It Matters: This case serves as a critical stress test for integrating AI into sensitive public sectors like law enforcement. It underscores the urgent need for clear governance and robust guardrails to ensure these powerful tools support, rather than compromise, the integrity of the justice system.
AI Pulse
Sundar Pichai skipped mentioning AI in his Stanford commencement address, a notable omission amid a trend of students booing pro-AI graduation speeches.
Meta faces renewed privacy scrutiny after a hands-on review of its Ray-Ban AI glasses highlighted their potential for covert recording alongside factual inaccuracies during a test in Paris.
Yat Siu countered AI risk warnings, with the Animoca co-founder predicting that human creativity will become the most in-demand skill as AI handles more routine work.