US & China at Cross-Purposes on own interests in Global AI. ARD #115

Today’s theme: the US and China continue to work at cross-purposes on AI — both within and outside their borders. Almost every day, each country seems to reach for kryptonite against its own AI superpowers. Politicians and regulators on both sides keep taking steps that further balkanize global AI — even as their leading companies aggressively try to find ways to balance the conflicting demands and thread the needle on cross-border AI business. Three events for the AI Tech Wave — each with my Take first, then my Overall Take.

(1) The US Is Exploring Ways to Slow US Adoption of Chinese AI Models — Open or Closed

MP TAKE: This would be another shot in the US’s own foot if it’s implemented. There’s a bipartisan effort in the House — a number of lawmakers exploring slowing down or even banning the import and use, by US companies and customers, of Chinese open- or closed-source models. It would limit competition for the frontier AI labs in the US — specifically Anthropic and OpenAI — both of whom have, in various ways, been lobbying the US government for exactly these kinds of protections in the guise of global safety and geopolitics. Dario Amodei, Anthropic’s co-founder and CEO, has talked a lot against open-source models as insecure — even though, ironically, much of the world’s technology, from the PC to the internet, runs on 80% or more open-source software. That’s the reality of the substrate, the tech stack, of every major technology wave for the last fifty years.

And yet this political dialogue is now more real in the US than it has ever been this early — year four after ChatGPT first showed the world what large language models could do. It also, conveniently, helps the incumbents hold the price umbrella I talked about yesterday, on their latest and greatest models. That’s the crux: it’s regulatory capture dressed as national security, and it shrinks the global pool of the best models for US companies just as prices are climbing.

Sources, in narrative order: CNBC — Lawmakers probe growing use of Chinese AI models in US companies. For longtime readers, in narrative order: ‘Looking beyond Space & AI Races’ in AI-RTZ #1138; and yesterday’s ‘AI Price Umbrellas Getting Larger’ in ARD #114.

(2) China Is Exploring Curbs on Overseas Access to Its Top Models — While Permitting Some Nvidia GPU Purchases

MP TAKE: China, for now, seems careful in executing ‘tit-for-tat’ responses to match US politicians in both the Executive and Legislative branches. These look like carefully calibrated moves on either side. China is weighing curbs on overseas access to its top models, whether open or closed — even as, after weeks of delay, it looks set to permit some Nvidia GPU purchases by its own companies who desperately need them: the Alibabas, ByteDances and DeepSeeks of the world. Those firms are being encouraged to use homegrown chips from Huawei and others, but remain very eager to buy Nvidia — even though those chips are relatively ‘nerfed’ by US regulatory orders (a limited amount of H200s, per the reporting).

There’s a lot of tit-for-tat positioning here by both governments, ahead of the second meeting between Presidents Xi and Trump at the White House this September — now only weeks away. There are rumblings both sides want some sort of grand deal. But we know how these things turn: we’ve watched the US-Iran ceasefire negotiations happen, not happen, happen — and then not happen again. We may see that scenario play out here too. Either way, the signal is the same: rising political headwinds for AI companies in both countries.

Sources, in narrative order: Reuters — Beijing is looking at curbing overseas access to China’s top AI models. The Information — China plans to let top AI firms buy a limited amount of Nvidia H200 chips. For longtime readers: ‘Nvidia’s Jensen rides Air Force One to China after all’ in AI-RTZ #1085.

(3) Apple Taps China’s Memory Supply for Its China-Bound iPhones

MP TAKE: It’s good to see Apple continuing to explore its options to alleviate the global ‘RAMageddon’ pressures — the limited memory-chip supply and rising prices — even if these moves only help Apple’s iPhone supply and prices inside China. Apple looks set to tap memory from Chinese suppliers like CXMT. I wrote about this a few days ago as a potential near-term valve the US could turn to give US companies additional memory supply from China — while the three big oligopolists (SK Hynix and Samsung in Korea, and Micron here), who hold 80%-plus of global memory supply, are vigorously raising prices — 30, 40, 50% every few months, creating this RAMageddon situation for phones, computers and all manner of gadgets.

The catch: the memory Apple gets from CXMT would only go into iPhones made in China for the Chinese market — not the US. So it threads the needle for Apple, but the broader imperative remains for US tech at large: relief beyond the oligopoly — even despite Micron’s famous spending of tens of millions on lobbying on both sides of the aisle in Washington, in favor of curbs on RAM and chips from China. The same balkanization impulse cuts across the memory layer too.

Sources, in narrative order: FT — Apple interest thrusts China’s CXMT into the memory-chip spotlight. For longtime readers: ‘US hamstrung on memory chip supply via China’ in AI-RTZ #1128.

MP OVERALL TAKE

The political headwinds in both the US and China keep heightening the risk of truly balkanizing the AI tech wave — something we’ve never seen before across the PC, internet, or mobile waves. Yes, we’ve had a China Great Firewall that restricts a lot of US companies — but that was primarily balkanized within China, for China’s political purposes. It didn’t stop the spread of US technology across the rest of the world.

This is different. If the world starts using a lot more Chinese chips, Chinese open-source software and frameworks — against Nvidia’s CUDA — then in the next five or ten years you could have two big substrates of AI infrastructure, software and data centers, working with much more friction with each other. And that friction runs all the way up to the application layer serving over eight billion people — and, by then, 80 billion or more AI agents working for all those people. All of these moves also limit the opportunity to innovate and build for the US companies that want to do more, but are held back by the higher prices of the incumbents lobbying in DC. The firms keep finding ways to do business across the border. The open question is whether the politics lets them.

Gadget AI — Meta Explores ‘Always-On’ Recording on Its AI Smart Glasses

MP Take: I’ve discussed the social issues around AI Wearables and AI Smart Glasses a lot. Here’s the pattern: tech companies — and Meta especially — always lead first with the technology, with the capabilities they want and an eye to customer acquisition and monetization, ahead of any negative side effects for mainstream users, with a plan to take corrective action later, after regulatory pushback. Now Meta is testing ‘super-sensing,’ always-on glasses that can capture every moment — with a wrinkle: it says it’ll disable the camera if you tamper with or cover the indicator light.

Meta has a harder road here than most, because its model is advertising — where the users are the product. The latest Muse Spark model has entire advertising toolsets built around it. To its credit, Meta says it’s trying a privacy-first approach — remember it bought Limitless, which makes AI pendants. But society is a little creeped out by glasses: over a decade ago, early Google Glass adopters got called ‘glassholes’ because the camera was on and lit. And this reaches beyond glasses — it touches all the AI wearables coming from Apple, Google, Amazon, and OpenAI with Jony Ive. Recognize the regulatory headwinds too: at least twelve to fifteen US states restrict multi-party recording without permission. All of that has to be addressed before these things reach mainstream adoption.

Sources, in narrative order: FT — Meta tests ‘super-sensing’ AI glasses that can capture every moment. Digital Trends — Meta will disable the camera on its AI smart glasses if you tamper with or cover the indicator light. For longtime readers, in narrative order: ‘Meta leans in on AI Smart Glasses “Science Projects”’ in AI-RTZ #849; ‘Meta’s new $299 AI smart glasses’ in ARD #103; and ‘The Privacy We Haven’t Yet Defined for AI Smart Glasses’ in ARD #113.

Questions

Q1 — What’s MP’s reaction to that ‘always-on’ recording capability?

We had a bit of this with Microsoft’s ‘Recall’ feature on Windows PCs two or three years ago — constantly taking pictures of everything on your screen, so you could later ask, “show me that blue shirt I looked at on that e-commerce site four months ago.” Always-on recording is essentially an effort to train models on your local data — your local life, work and business. The smart-glass effort runs along the same lines. It’s a feature that would be attractive — but the privacy issues are gnarly, to say the least. On a relative basis, as I’ve said before, I’d trust Apple and probably Google a little ahead of Meta here. But all of them are recognizing the privacy hurdles. Meta also seems to be trying to get ahead of Apple and its rumored AirPods with cameras next year — though that’s a distinctly different use of cameras from what Meta appears to be pressing for.

Q2 — What’s the core difference between Apple’s and Meta’s approaches to AI Wearables?

It comes down to a top-down difference in how each company thinks about privacy and trust. Meta has shown, time and again, that where there’s an opportunity to build something quickly for market share, it will do it first and ask forgiveness later — if at all. That’s been its modus operandi. Its bread is buttered on advertising, so there’s a natural pull to use audio, picture and video data for those purposes — now supercharged by AI, and sometimes at cross-purposes with its other objectives (exactly today’s theme), ahead of user permission. Not to mention the multi-party recording restrictions in over a dozen US states. It’ll be interesting to see how that plays out when we finally have mainstream AI smart glasses from Apple, Google, Amazon and Meta all at once.

Wrapping up

Today’s AI-RTZ #1142 — the US frees OpenAI’s GPT-5.6 for wide, global release — now three weeks in, the US has green-lit OpenAI’s latest and greatest GPT-5.6, up against Anthropic’s 5.6, which was freed earlier this week. Both companies are now unshackled to offer their best models to the rest of the world — presumably at rising a la carte prices — helping each accelerate revenue ahead of their planned mega-AI IPOs (Anthropic shooting for this year; OpenAI now looking to next). Recommended as today’s reading.

Tomorrow — ARD 116 on AI-RTZ 1143.

Thanks for joining us today, AI Curious Folk. Stay tuned.

— MP

Full Source Reading —

For the broader context, see the canonical sources for ARD 115 — in today’s narrative order:

Event 1 — US Slowing US Adoption of Chinese AI Models

• CNBC — Lawmakers probe growing use of Chinese AI models in US companies

• AI-RTZ #1138 — Looking beyond Space & AI Races

• ARD #114 — AI Price Umbrellas Getting Larger

Event 2 — China Curbs on Its Top Models + Nvidia H200

• Reuters — Beijing is looking at curbing overseas access to China’s top AI models · The Information — China plans to let top AI firms buy a limited amount of Nvidia H200 chips

• AI-RTZ #1085 — Nvidia’s Jensen rides Air Force One to China after all

Event 3 — Apple Taps China Memory (CXMT)

• FT — Apple interest thrusts China’s CXMT into the memory-chip spotlight

• AI-RTZ #1128 — US hamstrung on memory chip supply via China

Gadget AI — Meta ‘Always-On’ AI Smart Glasses

• FT — Meta tests ‘super-sensing’ AI glasses that can capture every moment · Digital Trends — Meta will disable the camera if you tamper with the indicator light

• AI-RTZ #849 — Meta leans in on AI Smart Glasses ‘Science Projects’ · ARD #103 — Meta’s new $299 AI smart glasses · ARD #113 — The Privacy We Haven’t Yet Defined for AI Smart Glasses

Clips from today

Clip 1 — Meta’s AI Glasses: Always-On Recording

Watch on YouTube Shorts

Meta is testing ‘super-sensing,’ always-on AI glasses that can capture every moment — disabling the camera only if you tamper with or cover the indicator light.

MP Take: Meta especially always leads first with the technology — acquisition and monetization ahead of the side effects, with corrective action promised later. AI Smart Glasses are the latest bleeding-edge category being pushed as hard as it can, ahead of society’s comfort.

Clip 2 — Apple’s Chinese Memory Supply Gambit

Watch on YouTube Shorts

Apple is exploring China’s CXMT for memory in its China-bound iPhones — a way around the ‘RAMageddon’ squeeze from the SK Hynix / Samsung / Micron oligopoly.

MP Take: Good to see Apple exploring options to ease RAMageddon — even if it only helps iPhone supply and prices inside China. The bigger imperative is memory relief for US tech beyond the oligopoly, despite Micron’s heavy DC lobbying to limit Chinese supply.

Clip 3 — AI Balkanization: A New Tech Divide

Watch on YouTube Shorts

Political headwinds in both the US and China keep raising the risk of balkanizing the global AI markets — across supply chain, market access, and investment flows in both directions.

MP Take: That would make this AI Tech Wave a very different one from every tech wave of the last fifty years — more net-negative unintended consequences than positive ones. The firms keep finding ways across the border; the open question is whether the politics lets them.

Clip 4 — US Explores Banning Chinese AI Models

Watch on YouTube Shorts

A bipartisan House effort is probing — and may move to slow or ban — US companies’ use of Chinese AI models, open or closed.

MP Take: Another shot in the US’s own foot. It shrinks the global pool of the best models for US companies, raising prices — and plays into top US frontier labs inciting politicians in the name of national security while accelerating their own regulatory capture. The price umbrella, defended by policy.

About AI Ramblings Daily (ARD), and AI-RTZ

Both are daily. Both are free. Both are about AI. But they’re different mediums carrying different messages.

AI-RTZ is the morning text — a deeper written take on one idea, published by at least 5 AM EST. Today: post #1142.

AI Ramblings Daily is the afternoon video + podcast — my ad hoc takes and perspective on the day’s AI issues & news flow, around 20 minutes, with short 1-2 minute clips for quick topic views. Today: episode #115.

Subscribe to either or both on michaelparekh.substack.com. They run as separate Sections you can opt into or out of.

Links used in today’s show (already embedded inline above; listed here for reference)

Take 1 — US Slowing US Adoption of Chinese AI Models:

• CNBC — Lawmakers probe growing use of Chinese AI models in US companies

• AI-RTZ #1138 — Looking beyond Space & AI Races

• ARD #114 — AI Price Umbrellas Getting Larger

Take 2 — China Curbs on Its Top Models + Nvidia H200:

• Reuters — Beijing is looking at curbing overseas access to China’s top AI models

• The Information — China plans to let top AI firms buy a limited amount of Nvidia H200 chips

• AI-RTZ #1085 — Nvidia’s Jensen rides Air Force One to China after all

Take 3 — Apple Taps China Memory (CXMT):

• FT — Apple interest thrusts China’s CXMT into the memory-chip spotlight

• AI-RTZ #1128 — US hamstrung on memory chip supply via China

Gadget AI — Meta ‘Always-On’ AI Smart Glasses:

• FT — Meta tests ‘super-sensing’ AI glasses that can capture every moment

• Digital Trends — Meta will disable the camera on AI smart glasses if you tamper with the indicator light

• AI-RTZ #849 — Meta leans in on AI Smart Glasses ‘Science Projects’

• ARD #103 — Meta’s new $299 AI smart glasses

• ARD #113 — The Privacy We Haven’t Yet Defined for AI Smart Glasses

Q1 + Q2 — Meta ‘always-on’ reaction + Apple-vs-Meta on wearables:

• Q1 — (MP’s own — Microsoft Recall parallel; trust Apple/Google ahead of Meta)

• Q2 — (MP’s own — Meta’s ad model + ‘do first, ask forgiveness’; multi-party recording rules)

Companion text:

• AI-RTZ #1142 — The US frees OpenAI’s GPT-5.6 for wide global release

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here.)

Subscribe to AI: Reset to Zero for daily AI Ramblings + Sunday Bigger Picture posts.