John Ternus, Claude 5.1, Physical Superintelligence

For those of you that are new here. Welcome to the party, pal.

The format I’m working on: one prediction at the top, the stories that earned their spot and the vision to close it out.

The people building the future are getting older, the models are getting smarter, and the physics ChatGPT moment is coming soon.

Anthropic just released Claude Fable 5.1 and Claude Mythos 5.1, claiming the world's top benchmark for coding and knowledge work.

Physical Superintelligence raised a $58M seed round to commercialize fundamental physics breakthroughs, not software, not models, physics itself.

Tim Cook's last day is done. John Ternus, Apple's hardware chief, inherits a $3 trillion company at the exact moment AI and spatial computing collide.

These are not separate stories. They are three readings of the same pressure: the next decade of compounding leverage comes from whoever commands the physical world, not just the digital one.

My bet is that within three years, physics-grounded research labs will produce at least one commercially viable breakthrough that no scaling-law AI lab predicted or could have generated. The architecture matters as much as the compute.

My bet is that within 3 years we will see breakthroughs in physics and physics-adjacent scientific fields that will accelerate our ability tackle real world problems like energy production and distribution or the creation of novel materials to build previously unimaginable products.

Below: a VC walks out of Neuralink stunned, Grok controls your Tesla hands-free, and hardware founders are getting older for good reason.

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 this week, claiming the top position for coding and knowledge work. That is a direct challenge to OpenAI's o3 and Google's Gemini Ultra landing at the same time the frontier labs are all sprinting.

If you are building a dev tool, a coding assistant, or anything where model quality is a differentiator, benchmark these now. Defaults shift fast and the gap between first and second place has real consequences for user retention.

Alex Wawro, co-founder of Physical Superintelligence PBC, just raised a $58M seed round to build a lab dedicated to discovering and commercializing transformative physics, not models, not software. Their launch video explains the mission with unusual clarity: physics research itself as a scalable, commercial enterprise.

I'll put something personal on record here: this is the kind of work I would genuinely want to contribute to someday. The same acceleration curve that hit math and coding benchmarks is coming for physics.

My bet: within five years, at least one physics-first lab produces a commercially viable result that scales faster than any model trained purely on existing data could have reached.

Separately from PSI, two researchers declined a multi-billion dollar offer from Bezos to build Accelerated Understanding (AU), a physics-grounded AI architecture that runs on principles traditional LLMs ignore entirely. The model reportedly cannot run on a standard PC.

The "more compute, more data" orthodoxy has held for a decade. AU is a direct structural challenge to it. If the architecture delivers, the entire stack gets rebuilt from scratch. Worth watching closely.

Sam Altman is signaling that the next major OpenAI release will operate computers autonomously at superhuman levels and, more boldly, generate genuinely novel scientific discoveries. That is the line: from tool to researcher.

Watch whether early access users report outputs that could not have been retrieved or remixed from training data. That is the signal worth waiting for, not the benchmark score.

Get revenue agents, workflows, and automations across every stage of your motion. Access customer data in real time through Attio's web app, MCP, API, and SDK.

Then Ask Attio anything about your business and get instant answers.

It's the CRM that runs the work behind every win.

Try Attio Now

Tim Cook signed off on his last day as Apple CEO this week. His successor is John Ternus, Apple's head of hardware engineering, the person responsible for Apple Silicon and the physical product line. A hardware chief taking the top seat at the exact moment AI and spatial computing converge is not an accident.

The incoming CEO's instinct will be to build the physical layer of whatever AI becomes. Watch the first product decision Ternus makes as CEO. It will tell you everything about Apple's next decade.

A Tesla owner tested 170 voice commands using Grok via "Hey Grok" and got 116 executed hands-free, covering climate, navigation, and vehicle settings. That is a 68% success rate on a comprehensive test, not a cherry-picked demo.

Gavin Baker called Grok's coding bot a "ChatGPT moment" this week too, a useful footnote: the same agent capability that compresses hours of coding work into seconds could, long term, become the thing actually driving the car while you talk to it.

Pair that with Tesla's Robotaxis beginning to flood streets in testing and the vehicle-as-AI-platform thesis is no longer theoretical. If you are building anything in the mobility or fleet management space, the AI-native car interface just got a concrete benchmark to beat.

Ben Horowitz and Martin Casado at a16z backed Patrick Collison's observation: there are fewer young founders today because hardware, deep tech, and physical infrastructure require domain expertise, regulatory navigation, and capital patience that takes years to build. Elon at 22 was building software. SpaceX and Tesla came later.

If you are a founder in this space and you feel like you are "too late," you are probably right on time.

Dyson's CameraJet is a camera-equipped electric toothbrush that detects gaps between teeth and fires mouthwash to floss automatically, no string required, for $499. The same playbook that worked on vacuums and hair dryers: take something mundane, over-engineer it, charge accordingly.

The bet for Dyson is that oral care has a Peloton-style premium consumer waiting. The bet for anyone building hardware products is that the mundane daily routine with a measurable health outcome is underpriced real estate.

Ars Technica reports that Meta is deploying robots inside its data centers to handle server swaps, cable management, and physical infrastructure tasks. The loop is closing: the AI that needs massive compute is now automating the physical machines that provide it.

Data center operations is not a glamorous category, but it is a massive one. If you are building in robotics or physical automation, this is your proof-of-concept at scale from one of the best-resourced operators in the world.

Luke Bugbee, Head of Design at 8VC, visited Neuralink for the first time and called it stunning. 8VC is one of Silicon Valley's most connected firms. That reaction, from a design-trained eye inside the building, signals Neuralink is past science project territory and into something that reads as real to serious people.

The same week, Elon Musk claimed MRI scans can reveal a brain's computational abilities, framing the brain as hardware with readable specs. Most neuroscientists would push back hard on that. But as a philosophical foundation for Neuralink's roadmap, it is clarifying. The company is building toward a world where the brain is upgradeable in the same way a chip is. Straight out of a Black Mirror episode, except the VC visit happened last Thursday.

This space is early and speculative. But the gap between "science fiction" and "VC visiting the lab and walking out wide-eyed" just narrowed considerably. Keep watching the pace of clinical results.

The counterargument worth sitting with is this: every era thinks it is at the inflection point.

Deep tech has promised to reshape the physical world before, and the timelines have almost always been wrong. Physics labs, brain-computer interfaces, humanoid robots in data centers: every one of these categories has a decade-long graveyard of well-funded companies that did not make it.

What is different now is the compounding. The tools available to a physics researcher today, the simulation capacity, the model-assisted hypothesis generation, the speed of iteration, are categorically different from what existed five years ago.

The same acceleration that made software eat the world is now being pointed at the physical layer. That does not guarantee the timelines are right, but it does change the base rate.

The thing to watch is not whether any single company in this edition succeeds. It is whether the category of "people who deeply understand physical systems and can now use AI-grade tools to work on them" starts producing results faster than anyone expected.

If that happens, the companies that positioned at that intersection in 2025 will look prescient. The companies that did not will spend the following decade catching up.

John Ternus running Apple's hardware instincts into the AI era, a $58M seed round for a physics research lab, and Neuralink turning serious skeptics into believers one facility tour at a time: the week's evidence points in one direction.

The physical world is the next surface area. Build toward it.

Looking for more?

Ready to transform artificial intelligence from a buzzword into your personal revenue generator?

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get The Free Guide

Get revenue agents, workflows, and automations across every stage of your motion. Access customer data in real time through Attio's web app, MCP, API, and SDK.

Then Ask Attio anything about your business and get instant answers.

It's the CRM that runs the work behind every win.

Try Attio Now