01The frontier, gated
The best model isn’t for you.
Opus 4.7 is better at coding. It sees images more clearly. It has stronger thinking modes. And it costs the same. Normally that would be the whole story. But on the same page, Anthropic showed us Mythos — their private model — and said it’s better on every conceivable metric.
There’s no waitlist. There’s no public date. There’s no way to try it. Only a small group of companies get access now. On launch day, Anthropic made a different kind of announcement: the frontier moves behind a door, and most of us are on the wrong side of it.
Opus 4.7 — what you get
A faster, sharper model at the same price. The visible release.
- CodingStronger benchmarks, longer thinking modes.
- VisionClearer image understanding.
- PriceUnchanged from Opus 4.6.
- AccessPublic — API and consumer tiers.
- TrialAvailable today.
Mythos — what you don’t
Better on every conceivable metric. The invisible release.
- AccessSmall group of companies.
- WaitlistNone disclosed.
- DateNot announced.
- TrialNo path for users.
- Framing“Shared slowly, if at all.”
It’s not a race about who builds the best model anymore. It’s about who gets access to it, who gets checked, and who gets left out.
02The identity layer
KYC for the model.
Anthropic also started asking some users for a government ID and a live selfie to unlock features. It’s not clear which features require it, and it’s not clear how long the data is kept. This is the part that should get more attention than the model upgrade.
A model upgrade is a product decision. An identity requirement is an infrastructure decision. It reshapes who can walk in the door before you ever get to the model behind it.
Jim’s take
This is a strategy, and of course it’s marketing. Anthropic is showing the most powerful AI can be held back and shared slowly — that the frontier is something you unlock, not something you ship.
At the same time, they’re putting identity checks around parts of the product. That changes the shape of the market. The fight stops being “who builds the best model” and becomes who gets access, who gets checked, and who gets left out.
04Silicon, booked to 2027
TSMC is the cleanest read on whether the boom is real.
Nvidia designs the chips. TSMC makes them. The latest results came in and profit hit $18 billion — up 58% from last year. Most of TSMC’s chip business is now advanced chips. Orders are already booked through 2027.
+58%
Profit, year over year
This isn’t talk. This isn’t theory. It’s real money, booked out two years in advance. But the same numbers also show how fragile the whole thing is: so much of modern AI runs through one company, in one place. That place is Taiwan.
AI is not just a software story anymore. It’s a supply story. It’s a factory story. It’s a geopolitical story.
05A memo, aimed
A leaked OpenAI memo praised Amazon — and blamed Microsoft.
A private memo surfaced this week. It praised a new deal with Amazon. It said Microsoft had limited OpenAI’s ability to reach some big customers. It landed right after both companies agreed to renegotiate the famous partnership.
OpenAI is still moving toward a public offering — more than $25 billion in yearly revenue now. The leak feels intentional. OpenAI is reshaping the story in advance. The message is simple: Microsoft slowed us down; now we want room to grow faster.
The Memo Said
“Microsoft has limited OpenAI’s ability to reach some big customers.”
Microsoft Did
Quietly took the Norway data-center capacity OpenAI walked away from.
Jim’s take
Microsoft’s calm response matters too. When a partner steps away from data-center capacity and you quietly absorb it, that’s not how companies act when everything is fine.
If you use OpenAI tools, watch this one closely.
06Distribution wins
Meta is spending at a scale that’s hard to believe.
Between $115B and $135B on AI this year — roughly double last year’s spend. Meta also released a smaller model called Muse Spark, which it says can do work close to the bigger ones on less compute.
1B+
Distribution endpoints
3
Surfaces — WhatsApp, IG, Ray-Ban
The bet isn’t just better models. It’s distribution. Meta already has the audience — a billion-plus users across products people open every day. It’s pushing AI deeper into WhatsApp, Instagram, and Ray-Ban glasses, and speeding up data-center construction to feed them.
Meta doesn’t need to find an audience. It already has one.
The real question isn’t whether Meta is serious — obviously they are; they’re spending $135 billion. The question is whether anyone without a billion users can actually compete.
07Where the money went
Four of every five venture dollars went to AI. Everything else is starving.
$242 billion went into AI companies in just the first quarter. About 80% of all global venture funding. In many other sectors, the funding picture is the weakest it’s been since 2009.
Q1 2026 · Global venture allocation
AI · 80% · $242B
Everything else · 20%
A flow this lopsided changes behavior fast. It changes what founders build. What investors want. What boards approve. Every new product you’ll see is shaped by it. Even companies that aren’t AI companies now have to explain their AI story — and most are tacking the letters onto the end of their name. Not because it’s trendy. Because it’s required.
Jim’s take
AI isn’t one big trend among many right now. It’s the only trend.
