🍭 Chip Control Chaos 💾

Good morning. The U.S. might start approving AI chips like nightclub bouncers.

“Sorry sir, that GPU cluster is not on the list.”

Let’s dive in 👇

🍭 What’s Cookin’:

• U.S. wants control over every AI chip sale on Earth

• Anthropic clashes with the Pentagon over Claude restrictions

• Broadcom’s AI boom just doubled its chip revenue

Global Economy
💰️ The U.S. Wants To Control Every AI Chip On The Planet

The Bite:

The U.S. Department of Commerce has drafted rules that would require government approval before Nvidia or AMD can ship AI chips to any country in the world.

Foreign companies buying fewer than 1,000 chips face a light review.

Those seeking 100,000 or more would need government-to-government security assurances.

Anyone buying 200,000 or more would need to commit to investing in U.S.-based AI data centers.

The rules are not final and the White House has already pushed back.

A senior official said the draft does not reflect Trump's direction on chip exports.

Snacks:

• A 129-page draft is circulating through Commerce's Bureau of Industry and Security. It’s the sixth iteration from that office

• Purchases of 200,000+ chips would require foreign buyers to invest in U.S. AI infrastructure or provide security guarantees

• Installations approaching 200,000 chips could trigger on-site visits from U.S. export control officials

• The proposal applies globally

• AMD dropped over 2% and Nvidia over 1% on the news

Why it Bites:

This draft isn't really about China. China is already locked out.

This is about turning AI compute into a diplomatic tool.
It’s the same playbook Trump used with tariffs, now applied to GPUs.

The tiered structure tells the story.

Small purchases: easy.
Large purchases: you owe us a data center.

The U.S. is trying to make sure that when data centers get built somewhere else, America gets something back for it.

The White House pushing back publicly on its own Commerce Department is worth noting.

Not because the rules are dead, but because it reveals that even internally there's no settled answer to a simple question:

Do we want the world buying American AI chips, or do we want to control who gets them?

Right now the answer appears to be both.
That’s a contradiction, not a strategy.

For builders and operators outside the U.S., nothing changes today.
But the direction of travel is clear.

Frontier compute is becoming more a political thing than a commercial one.

Steal This Prompt
🍬 Turn Anything Into Hyperrealistic Gummy Candy

Turn any icon, logo, or object into a glossy, translucent gummy candy. Same shape, same colors, just dipped in sugary chaos.

The prompt preserves the original design while adding jelly texture, shine, and that “don’t lick the screen” realism.

Use it to:

• Turn startup logos into candy brand mockups

• Make gummy versions of famous icons or emojis

• Build oddly satisfying candy-style UI visuals

Workflow:

1. Click this link (Prompt)

2. Paste into your AI model

3. Replace the #s with your image or icon

4. Watch it cook (suddenly your logo looks like it belongs in a Haribo commercial)

   Try the prompt on Snack Prompt

AI Gossip
🧪 Anthropic As The Government’s Difficult Child

The Bite:

The Pentagon has formally designated Anthropic a supply chain risk to national security.

The move bans defense contractors from using Claude in military work and gives Anthropic six months to be phased out.

The dispute centers on two red lines Anthropic refused to drop: no fully autonomous weapons, and no mass surveillance of U.S. citizens.

The Pentagon said it needed Claude available for "all lawful purposes" and wouldn't accept restrictions from a private contractor.

Amodei called the designation legally unsound and said Anthropic will challenge it in court.

Snacks:

• Anthropic was the only AI company cleared for use on the Pentagon's classified networks until last week

• The military is actively using Claude in its ongoing Iran campaign, via Palantir's Maven Smart System, while the ban plays out

• Anthropic's own lawyers argue the designation only restricts Claude's use as a "direct part of" DoW contracts. Not all commercial activity

• Microsoft studied the designation and confirmed Claude remains available across M365, GitHub, and Azure for non-defense work

• OpenAI struck a new Pentagon deal the same week, granting access to classified networks for "all lawful purposes”

Why it Bites:

Here's the truth sitting under this story: Anthropic didn't lose. Not really.

The Pentagon is still running Claude in an active war.

The commercial business is largely intact.
Microsoft confirmed it's keeping Claude across its entire enterprise stack.

What Anthropic lost is the government's favorite-child status.

The coveted position of being the one AI company trusted enough to run on classified networks.

And it lost that position specifically because it refused to be fully obedient.
Which tells you something important about what the government actually wanted.

The tricky part is that every other AI company is now watching this play out and doing the math.

Anthropic held the line and kept most of its business.
OpenAI signed "all lawful purposes" and got the classified keys.

Both outcomes are on the table.
Both come with a price.

What's new here is that an AI company decided its product was consequential enough to have terms. And survived the blowback.

Will it hold as the pressure keeps building? That's the question that matters.

_{Everything Else}
🧠 You Need to Know

🤖 OpenAI drops GPT-5.4 with native computer use
→ Hits 75% on OSWorld-Verified, supports 1M token context, and claims 33% fewer false individual claims than GPT-5.2.

🏛️ Anthropic: the Pentagon ban is narrower than it sounds
→ Microsoft's lawyers confirmed Claude stays available for all non-defense commercial work, including M365, GitHub, and Azure.

🌐 U.S. wants approval rights on every AI chip sold globally
→ Countries buying 200K+ Nvidia or AMD chips would need to invest in U.S. data centers or provide security guarantees.

⚡ White House vs. Commerce: Trump's team fights its own chip draft
→ A senior official called the 129-page proposal "diffusion 2.0". The Biden-era export rule Trump already killed last year.

📈 Broadcom: AI chip revenue up 106%, CEO sees $100B+ by 2027
→ Q2 guidance came in at $22B vs. $20.5B expected. New $10B share buyback approved.