Tech Lobby Tells Trump: Don’t Slap ‘National-Security Risk’ on Anthropic

TL;DR: Google, Apple, and friends say labeling Anthropic a supply-chain boogeyman will nuke the whole AI stack. The White House isn’t convinced.

A coalition of trade groups—fronted by the likes of Google, Apple, Cisco, and Intel—has fired a letter at the Trump administration, begging it to reverse a draft rule that would designate Anthropic PBC a “national-security risk” under Defense Production Act (DPA) authorities. The label, quietly floated by the Bureau of Industry and Security (BIS) last month, would force any U.S. company using Anthropic models to conduct onerous supply-chain audits and potentially cut off cloud access to non-allied nations.

Why Anthropic?

The government’s worry appears to center on Anthropic’s custom silicon partnerships—specifically its reliance on Taiwan’s TSMC for 5-nm “Tesla-beating” AI accelerators and South Korea’s SK hynix for 12-layer HBM3e stacks. Officials fret that a single geopolitical hiccup could kneecap U.S. AI capacity. Ironically, Anthropic is the least vertically integrated of the frontier labs; it doesn’t design its own GPUs like Google (TPU) or Amazon (Trainium). That openness is exactly why the trade groups say punishing Anthropic is counter-productive.

The industry’s math

The letter cites internal forecasts showing a DPA tag would:

• Add $2.7 B in annual compliance costs across the ecosystem (Deloitte estimate).
• Push 18–22 % of AI workloads to non-U.S. clouds within 18 months (IDC model).
• Delay 37 commercial LLM deployments currently on 2026 roadmaps.

Those numbers smell like they were cooked up in a PR kitchen, but even if they’re off by half, we’re still talking real money and real schedule slips.

Vague marketing watch

The draft rule leans on the phrase “foreign ownership, control, or influence (FOCI) of undetermined magnitude.” Translation: nobody has shown a smoking gun—just vibes. Meanwhile, Anthropic’s cap table is 79 % U.S.-domiciled (per its last Series E filing), and its board has zero non-allied nationals. If that meets the FOCI threshold, then so does half of Silicon Valley.

Benchmark corner

For the spec nerds, here’s the silicon footprint at risk:

Cutting those off wouldn’t just hurt Anthropic; it would yank the rug from under every startup renting Claude via AWS Bedrock.

What happens next

The White House has 90 days to publish the final rule or kick it back to BIS. Congressional Democrats—traditionally no friends of big tech—have oddly stayed quiet, probably because “national security” is the one phrase that still trumps antitrust headlines. Meanwhile, the Commerce Department is reportedly asking for a classified briefing on Anthropic’s constitutional AI safeguards. Yes, they want to classify safety papers. Good luck FOIA-ing that.

Bottom line

Weaponizing the DPA against a U.S. company whose biggest sin is buying the same chips everyone else buys sets a gnarly precedent. If the rule sticks, expect every CFO to add a “regulatory haircut” line item to AI budgets—and maybe stash a few GPUs in a friendly data center north or south of the border. Because nothing says “secure supply chain” like forcing your own firms to shop abroad.