Today’s theme: they’re ‘setting the stage’ again — more theater ahead in US-China AI, in this AI Tech Wave at its most geopolitical. The two countries are prepping for their second presidential meeting at the White House on September 24, and the big issues remain wide open: open vs closed AI models and their ‘distillation,’ and how US AI data centers and their power ramp up here versus there — a lot of cross-winds for the AI industry across the two largest AI markets in the world. Three events, each with my Take, then my Overall Take. Plus a Gadget AI on IMAX’s biggest moment in decades, and two questions. Let’s get started.

(1) Prep Underway — Preliminary Negotiating Positions Being Set by Each Party

MP TAKE: As the daily posturing around open vs closed AI models from China accelerates, it’s worth remembering these are the preliminary chess moves ahead of the second top-level US-China meeting in September. Reuters confirms the two sides are set to hold AI talks in September; Axios reports a ‘secret’ administration battle to fight Chinese AI — centered on China’s open models like Kimi K3 (Moonshot AI) & Qwen 3.7 Max (Alibaba). The first meeting had its own dramas, culminating in Nvidia’s Jensen Huang being picked up in Alaska on Air Force One — almost a Home Alone scenario — to join the President’s trip to China. Expect a similar set of machinations this time.

On China’s side, President Xi just wrapped a big AI conference in Shanghai, where the country’s big and small companies put out their best work. Moonshot’s Kimi K3 — at 2.8 trillion parameters — goes toe-to-toe with the best US frontier models, and Alibaba released its 2.4-trillion-parameter Qwen 3.7 Max around the same keynote. The geeky details matter less than the posturing: China is telling the world to build on its open models rather than the frontier models the US is guarding in a protective manner. It’s a compelling pitch for a global audience — and that’s exactly what’s different now.

What’s also different: more of the bigger US tech companies — Microsoft, Meta, Amazon — are leaning into open-source models from China than they were just a few months ago. Even mainstream opinion is shifting: the Washington Post now argues that powerful AI models being given away free is good competition — and was inevitable. These US players, along with Nvidia, Apple and Elon’s SpaceX-AI/Tesla interests in China, are the ones to watch ahead of the DC meetings.

Sources, in narrative order: Reuters — US, China to Hold AI Talks in September; Washington Post — Powerful AI Models Are Being Given Away for Free. It Was Inevitable; Axios — The Secret Trump Administration Battle to Fight Chinese AI. For longtime readers: ‘China ramping up in AI vs US, open source and beyond’ in AI-RTZ #1153; and ‘Nvidia’s Jensen rides Air Force One to China after all’ in AI-RTZ #1085.

(2) The Key AI Companies Posture and Position Themselves — Both Here and There

MP TAKE: Why are US tech giants leaning in? Because today, in this AI tech wave, the market for US tech is global. Unlike the internet wave — when yours truly was head of internet research at Goldman Sachs, and 70-80% of the overall market was in the US, with a little in Europe and almost nothing in China — the world has changed. Microsoft, Amazon and Meta are now going to the White House saying they’d support working with more open Chinese AI models, because it gives customers — most of them global companies across every industry — a wider set of choices: large models and small, open and closed, plus all the harnesses and applications being built around them. Playing the game as if it’s a US-only context “sets us up more for failure than success.”

Expect strong words from the administration in the months ahead, pro and con on open source. On the hawkish side, CNBC reports Treasury’s Bessent floating that the US could sanction China over AI-model ‘theft’ — the ‘distillation’ charge in another guise. On the other side, the domestic open-weights ecosystem keeps building anyway — Thinking Machines just entered as the next US open-weights model maker, with Nvidia’s backing. Ironically, Thinking Machines is said to have ‘distilled’ its Inkling model with Chinese AI models. With pro-open-source voices recently departing the administration, US policy will carry a thematic inconsistency — hawkish sanctions talk pulling one way, a homegrown open-weights ecosystem pulling the other — right through the September meetings. Watching which voices win each week is the next global AI football game to watch.

Sources, in narrative order: CNBC — Bessent Says US Could Sanction China Over AI-Model ‘Theft’. For longtime readers: ‘Thinking Machines next US open weights AI model entrant’ in AI-RTZ #1149.

(3) The Regulatory Fight — Open vs Closed AI, and Data-Center Build-Outs

MP TAKE: The regulatory fight over open vs closed AI, and over AI data-center build-outs, remain the two biggest issues being fought by both parties — at the Federal, State and local levels — with some big unintended consequences for the US if taken too far.

For now, it’s OpenAI and Anthropic against open source — under the guise of cybersecurity, geopolitically-driven ‘national interests,’ and US leadership in the US-China ‘AI Space Race.’ But the ground under that argument is contested: Axios reports a top Pentagon official publicly blasting OpenAI’s Dean Ball — a fault line inside the government over the closed-lab position — while Senator Mark Warner is putting forward his own AI plan, a bipartisan approach to regulating the build-out of AI data centers. That’s the other half of this political football: as I covered yesterday, Elon’s Memphis build-out became a catalyst for nationwide headwinds — around water, power, and, heading into the midterms, rising utility and electric prices. And Stratechery’s Ben Thompson makes the case directly — ‘Who’s Afraid of Chinese Models?’ — that the fear is overblown. The open question is whether the broader US stakeholders — big and small tech, and businesses large and small — make their voices heard for a global diversity of open and closed AI models, rather than ceding the frame to a handful of closed labs. Uber VC Bill Gurley makes a similar point in this Washington Post op-ed.

Sources, in narrative order: Axios — Top Pentagon Official Blasts OpenAI’s Dean Ball; Axios — Inside Senator Mark Warner’s AI Plan; Stratechery — Who’s Afraid of Chinese Models?. For longtime readers: ‘How Nvidia and Apple can be the global, US open source AI champions vs China’ in AI-RTZ #1089; and ‘Haste makes Waste in super-sizing AI data centers’ in ARD #122.

MP OVERALL TAKE

For now, open vs closed AI — along with the fights over building US AI data centers and their power at scale — remain the two biggest political footballs in US tech this year, full stop. The September 24 White House meeting is the near-term forcing function that concentrates all of it. And China’s real structural advantage in the data-center race, worth underlining, isn’t chips or chip-making machines — it’s power, and the ability to use eminent domain more aggressively there than here.

A word on ‘distillation,’ since it’s become central to the geopolitical AI kerfuffles. There’s a position being pushed — by the Anthropics and OpenAIs — that distillation is somehow Chinese ‘stealing,’ unfair and terrible. Recognize that distillation is just a technique as old as the large language models themselves. The very first models were trained by distilling all the data we’ve put on the internet over decades — that’s how GPT models were built. Those same US companies then went on to distill other US models and data stores; OpenAI, you can look it up, distilled YouTube videos from Google to train its multimodal capabilities. US labs have used these techniques against each other from the very beginning. So to wield the distillation charge as a stealing charge against China’s models is a theatrical technique to gain political advantage — not a charge with substance. There’s plenty of technical detail people can debate until the sun goes down, but that’s the technically factual take.

Much depends on the long-term opportunity for the US to build and keep its global tech leadership — and, to my mind, the path to that larger prosperity does not run through a handful of closed frontier models. It runs through a diverse, competitive ecosystem of open and closed models, US and global, with American champions like Nvidia and Apple helping anchor the open-source side. Even the Washington Post and Stratechery are now making versions of that argument in the mainstream. The theater in the weeks ahead will be loud — but the strategic question underneath it is the one that matters.

GADGET AI

Going Big and Not Going Home to watch a cool movie — IMAX’s Biggest Moment in Decades

MP TAKE — Gadget AI (take-first, Ep 103+)

The medium, as Marshall McLuhan famously said, ‘is the message.’ But often it’s the message that can re-make the medium. In this case, it was Christopher Nolan’s ‘The Odyssey’ — filmed on custom-built IMAX cameras — that turned a multi-decade-old format, IMAX, into the movie theater place to experience the visionary director’s message. IMAX was introduced to the world in 1971 but didn’t go mainstream in the US until around 2008; now, after years of people staying home to stream, the Odyssey has had a blockbuster opening — over a quarter of a billion dollars — and theaters are sold out for days and weeks as everyone wants the full IMAX experience. The Verge’s reviewer went in a skeptic and came out an IMAX believer.

For a movie-and-technology nerd, how Nolan made it is the fascinating part. IMAX effectively reinvented its cameras for him — he needed specialized, more soundproof rigs the actors had to work around — with very little CGI and casts of hundreds and thousands in the major scenes. It’s a story as old as time — literally, Homer’s Odyssey — introducing a new generation to the oldest of stories, and a vivid case that technology adoption makes a real difference when it’s paired with the right message. Here, a new message remade the medium. A format that’s been around for decades (IMAX’s long history) is suddenly the must-have experience again.

Sources, in narrative order: The Verge — ‘The Odyssey’ Turned Me Into an IMAX Believer; Wikipedia — IMAX (history). For longtime readers: ‘Three personal “Wow” moments with Apple Vision Pro’ in AI-RTZ.

QUESTIONS

Q1 — What’s Michael’s last similar ‘Wow’ moment watching a movie with new technology?

Answer: The Apple Vision Pro still delights me. Introduced two or three years ago at $3,400, it’s been all but written off as a commercial disappointment — but I said back then, and I’ll say again, that the IMAX experience at home is available through it, in a manner of speaking. I watch almost every movie, show and podcast on it. Just recently I watched an old favorite — Jaws — and it was a visceral experience, in both the video and the spatial audio. Whether it’s a real IMAX theater or a computer-and-display strapped to your head, the huge screen in front of you remains the wow.

Q2 — What’s Michael’s biggest issue with the Vision Pro?

Answer: Long-term comfort. You’ve got a computer with big displays on your head, so it runs a little heavy — but that’s manageable, and all the more excuse for a couple of ‘mini intermissions’ to refill a beverage in a two-hour-plus movie. The real fix comes later, when the lighter, more capable smart glasses I’ve written a lot about finally arrive as a more comfortable alternative. Until then, it still feels like an IMAX at home.

WRAP

Today’s AI-RTZ #1154 — AMD’s Helios Up Against Nvidia’s Vera Rubin AI Infrastructure — AMD’s new Helios system stepping up against Nvidia’s Vera Rubin at the AI-infrastructure level. A credible new competitor, with big customers like Microsoft and Meta already using Helios; it may take a few points of share, but the market keeps growing for everyone amid the shortage of AI GPU and data-center infrastructure.

AI Ramblings Daily on AI-RTZ is here to think through AI and reset. Together.

Tomorrow — back with ARD 124 on AI-RTZ 1155.

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

— MP

Full Source Reading —

For the broader context, see the canonical sources for ARD 123 — in today’s narrative order:

Event 1 — Prep Underway / Preliminary Negotiating Positions

• Reuters — US, China to Hold AI Talks in September · Washington Post — Powerful AI Models Are Being Given Away for Free. It Was Inevitable · Axios — The Secret Trump Administration Battle to Fight Chinese AI

• AI-RTZ #1153 — China ramping up in AI vs US, open source and beyond · AI-RTZ #1085 — Nvidia’s Jensen rides Air Force One to China after all

Event 2 — The Key AI Companies Posture and Position (Here and There)

• CNBC — Bessent Says US Could Sanction China Over AI-Model ‘Theft’

• AI-RTZ #1149 — Thinking Machines next US open weights AI model entrant

Event 3 — The Regulatory Fight (Open vs Closed + Data-Center Build-Outs)

• Axios — Top Pentagon Official Blasts OpenAI’s Dean Ball · Axios — Inside Senator Mark Warner’s AI Plan · Stratechery — Who’s Afraid of Chinese Models?

• AI-RTZ #1089 — How Nvidia and Apple can be the global, US open source AI champions vs China · ARD #122 — Haste makes Waste in super-sizing AI data centers

Gadget AI — IMAX’s Big Moment (Nolan’s ‘The Odyssey’)

• The Verge — ‘The Odyssey’ Turned Me Into an IMAX Believer · Wikipedia — IMAX (history) · AI-RTZ — Three personal ‘Wow’ moments with Apple Vision Pro

Clips from today

Clip 1 — AI Distillation: US vs China

Watch on YouTube Shorts

Amid the US-China AI theater, one charge keeps coming up: that China ‘steals’ via distillation. Here’s why that’s mostly political.

MP Take: Distillation is as old as large language models themselves. The first models distilled all the data we’ve put on the internet for decades; OpenAI distilled YouTube videos from Google to train multimodal; the US labs have distilled each other’s models from the start. Casting ‘distillation = stealing’ at China is a theatrical, political charge — not a real one — ahead of the September 24 White House meeting.

Clip 2 — AI Bill: Bipartisan Approach to Data Centers

Watch on YouTube Shorts

The other half of the US-China AI political football: how AI data centers and their power get built out here. Senator Mark Warner is putting forward a bipartisan AI plan.

MP Take: The regulatory fight over data-center build-outs runs at the Federal, State and local levels — with rising utility and electric prices becoming a midterm-election issue. Warner’s bipartisan approach is one attempt to get ahead of it. The strategic question underneath the theater is the one that matters: how the US scales AI infrastructure without repeating the ‘build fast, break things’ mistakes.

Clip 3 — US Tech Giants Back Open Chinese AI Models

Watch on YouTube Shorts

What’s changed ahead of the September US-China meeting: more of the biggest US tech companies are leaning into open-source models from China.

MP Take: It’s not just Nvidia, Apple and Elon’s Tesla with big China interests anymore. Microsoft, Amazon and Meta are going to the White House saying they’d support working with more open Chinese AI models — because it gives their global customers a wider set of choices, models and tools. Today’s AI market is global, unlike the 1990s internet wave. Playing it as a US-only game sets us up more for failure than success.

Clip 4 — China’s AI Open Source Push

Watch on YouTube Shorts

At President Xi’s big AI conference in Shanghai, China’s companies put out their best open models to court the world.

MP Take: Moonshot’s Kimi K3 (2.8 trillion parameters) goes toe-to-toe with the best US frontier models; Alibaba released its 2.4-trillion-parameter Qwen 3.7 Max around the same keynote. The posturing is the point: China is telling the world to build on its open models rather than the closed US frontier ones. China’s real structural advantage isn’t chips — it’s power, and more aggressive use of eminent domain to build data centers.

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 #1154.

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 #123.

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 — Prep Underway / Preliminary Negotiating Positions:

• Reuters — US, China to Hold AI Talks in September

• Washington Post — Powerful AI Models Are Being Given Away for Free. It Was Inevitable

• Axios — The Secret Trump Administration Battle to Fight Chinese AI

• AI-RTZ #1153 — China ramping up in AI vs US, open source and beyond

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

Take 2 — The Key AI Companies Posture and Position (Here and There):

• CNBC — Bessent Says US Could Sanction China Over AI-Model ‘Theft’

• AI-RTZ #1149 — Thinking Machines next US open weights AI model entrant

Take 3 — The Regulatory Fight (Open vs Closed + Data-Center Build-Outs):

• Axios — Top Pentagon Official Blasts OpenAI’s Dean Ball

• Axios — Inside Senator Mark Warner’s AI Plan

• Stratechery — Who’s Afraid of Chinese Models?

• AI-RTZ #1089 — How Nvidia and Apple can be the global, US open source AI champions vs China

• ARD #122 — Haste makes Waste in super-sizing AI data centers

Gadget AI — IMAX’s Big Moment (Nolan’s ‘The Odyssey’):

• The Verge — ‘The Odyssey’ Turned Me Into an IMAX Believer

• Wikipedia — IMAX (history)

• AI-RTZ — Three personal ‘Wow’ moments with Apple Vision Pro

Companion text:

• AI-RTZ #1154 — AMD’s Helios up against Nvidia’s Vera Rubin AI Infrastructure

(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.