Back to archive

A billion free protein structures, AI on your laptop, and the quiet walk-back on job loss

Mon, Jun 1, 2026 · 8 stories

This week everyone refreshed the same headline: Anthropic crossed $965 billion and passed OpenAI as the most valuable AI startup on the planet. The more interesting AI was free. An open-source model predicted the shapes of roughly one billion proteins and released them to anyone who wants them, far past what AlphaFold (DeepMind's protein-structure model) ever opened up. No frontier lab required, just a laptop.

Sam Altman and Dario Amodei spent the last year warning that AI would erase whole categories of jobs. This week they both walked it back, softening the timeline right as their companies line up for the public markets. Funny how the apocalypse gets rescheduled the moment there are retail investors to reassure. Doom was a useful story when it sold urgency and made the technology feel inevitable. It is a much worse story when you are asking Wall Street to believe your business is stable enough to go public.

Liquid AI tied the week together with an 8 billion parameter model that only fires about 1 billion of them per token (a mixture-of-experts design) and runs on your laptop, not in a data center. The frontier spent the same week proving how expensive it has gotten: a $36 billion debt deal floated just to buy Anthropic its chips, SK Hynix and Micron both past a trillion dollars on memory alone. Capital is concentrating at the top while capable, cheap-to-run AI keeps leaking out the bottom. If you are building, the bottom is where the room is.


Top Stories

"Move Over, AlphaFold": An Open-Source Model Predicts the Shape of 1 Billion Proteins

Nature

A newly released open-source structure-prediction model published predicted shapes for roughly one billion proteins, expanding the public structural universe far beyond what AlphaFold made available and putting frontier protein prediction in researchers' hands for free.

Why this matters:

The most capable result of the week was also the cheapest to access. Open-sourcing a billion predicted protein structures removes a cost and access barrier that used to favor well-funded labs, and hands every AI-bio team a shared base layer to build on. For founders, it reinforces that durable advantage rarely lives in the model itself. It lives in the proprietary data, workflow, and domain expertise layered on top of an increasingly commoditized foundation.

Sam Altman and Dario Amodei Both Walk Back Their AI Jobs-Apocalypse Predictions

Fortune

Both Sam Altman and Dario Amodei have publicly softened their earlier predictions that AI would cause mass unemployment, reframing the labor impact in more measured terms as both companies move toward IPOs.

Why this matters:

Executive forecasts about AI and jobs now function as investor communication as much as technical assessment. The move from imminent-disruption warnings to measured language tracks the approach to public markets, where durability sells better than upheaval. The underlying model capabilities did not change this week, only the framing around them. Operators making hiring and workforce decisions should weight observed capability over leadership rhetoric in either direction.

Liquid AI Ships an 8B-A1B MoE Trained on 38T Tokens

Liquid AI

Liquid AI released LFM2.5-8B-A1B, a mixture-of-experts model with 8 billion total parameters but only about 1 billion active per token, trained on 38 trillion tokens and tuned to run on local and edge devices rather than in the data center.

Why this matters:

The week placed AI's two diverging cost curves side by side. At one end, a $36 billion debt facility to finance chips for a single frontier lab. At the other, a sparse mixture-of-experts model capable enough to run on consumer hardware. For founders without hyperscaler backing, efficient open-weights models are the more relevant trend, lowering the capital needed to ship real products as frontier training costs keep climbing.


Quick Hits

Share:
Daniel Ryan

Daniel Ryan

Founder of Stratavore

Daniel Ryan has been shipping software for 15 years and is now building Stratavore. He reads too much AI news so you don't have to.

Get this in your inbox

Subscribe for free weekly AI intelligence digests.