Government & Corporate AI Stakes get Bigger. ARD #89

Michael Parekh AI: Reset to Zero

23 hours ago

12 MIN READ

Today’s theme: government and corporate AI stakes get bigger — and soon public investors will have to weigh the consequences, not just the presumably more sophisticated private ones to date. The White House hits a compromise on US AI regulation. China is still figuring out the right level of Nvidia in its AI infrastructure as Huawei falls behind in AI chips. And the costs and delays of gigawatts of AI infrastructure go up. Three Takes today, each with my Take — and my Overall Take.

(1) White House Settles on a 30-Day Review Period for Powerful AI Models

The Wall Street Journal had it — “Trump signs AI Executive Order to increase government oversight” — the White House settling on a 30-day review period for the most powerful AI models. The long-running fight — the original 60-to-90-day model-submission rule, softened to 14 days, then deferred — lands at 30. All this against a backdrop where nearly every US state has its own AI-regulation push that the federal government hasn’t been able to stop, running through and past the November elections. The back-and-forth is in AI-RTZ #935, “Battle Over AI Regulatory Executive Order.”

MP Take: A deal that makes no one happy on all sides — so likely a good deal. And a necessary step given the cybersecurity concerns being raised by Anthropic and OpenAI, especially ahead of their IPOs — with Mythos, Anthropic’s latest model, still hand-released to governments and major partners (and very expensive, as I wrote yesterday). Industry seeking cover yet clear runways on what they want to do with AI ahead. For now they have it — versus overseas markets and governments.

(2) China’s Huawei Struggles With AI Chip-Design Expansion

China’s Nvidia calibration: let them in through the front door, or deal with them via the back door.

The Wall Street Journal ran a deep read — “China’s chip ambitions run into a global tech wall” — without access to the latest lithography from the likes of ASML, Huawei (and SMIC, China’s TSMC-equivalent) could trail rivals by six to eight years by 2031 despite current innovations. The calibration question sharpens ahead of President Xi’s September 24 White House visit. The counter-frame is in AI-RTZ #1076, “Huawei’s AI Gains in China Over Nvidia.”

MP Take: A race for both sides to lose by not threading their needle over tech and trade, despite broader geopolitical goals. Nvidia, Apple and Tesla remain the key US companies to help both countries balance mutual interests — neither side can get what they want by going it alone. Especially since the physical-world AI of cars, robots and drones requires the full global supply chain. Nvidia remains a player to make one plus one equal three in the longer term. Tim Cook and Elon Musk have a lot of wood to chop on both sides — as does Jensen Huang of course, especially with Nvidia’s growing footprint in Taiwan (Nvidia Taipei headcount going from 1,000 to over 4,000) and its deepening TSMC relationship, where Nvidia is becoming the biggest customer, supplanting Apple — the two together already close to 50% of TSMC’s capacity. And TSMC remains the Federal Reserve of the tech/AI industry — its fab decisions govern how fast the whole industry grows.

(3) US AI Data-Center Build-Out Schedules Are Slipping

The Wall Street Journal had the schedule read — “America’s data-center build-out is falling way behind schedule” — regulatory, power and water pushbacks slowing the Googles, Microsofts, Amazons (and Elon and Meta) despite hundreds of billions in spend heading north of $2 trillion over the next year or two. And the spending read — “Google is daring rivals to keep up in AI spending race” — fresh off yesterday’s $80B+ equity raise. The delays are already pushing AI pricing from all-you-can-eat subscriptions toward a-la-carte, with real sticker shock for big business customers as AI compute bills rocket up. The investment FOMO around AI supply chains is in AI-RTZ #1083.

MP Take: The capital is there (sometimes barely). The investors are there (mostly). But things always take longer to build than expected — and what’s $50 billion per Gigawatt today will still be the going rate plus or minus 20% on a variety of factors. That gap of timing — between building plans and executed plans — remains the principal volatility risk to the AI Tech Wave this year and next, especially through the critical mega-AI IPOs. And this is going to accelerate, not decelerate.

MP OVERALL TAKE

The US and China remain in pole position on AI infrastructure build-outs despite regulatory and geopolitical headwinds in both places.

MP Take: The US has the edge on AI chips and supply chains. China on the Power front — its power deployment is far ahead of the US — and sheer innovation driven by the constraints on its companies, plus manufacturing ecosystems and skilled workforces measured in the tens and hundreds of thousands, beyond what the US can field alone. It’s not just about money; it’s about capable people. Both countries should be able to move the collective AI ball forward globally. But watch the volatility — especially with all these mega-AI IPOs being floated around.

Gadget AI — Microsoft Tries AI Devices for the Office, With Qualcomm: Project Solara

CEO Satya Nadella showed off Project Solara at Build — AI devices for the office, built with Qualcomm’s chips and design skills.

The Verge had the hardware with videos— “Microsoft shows off two devices at Build: a desk concept and a budget concept” — one a desk gizmo (think Alexa with a small TV), the other a lanyard badge that’s essentially a computer with screen, cameras and sensors. Ars Technica had the architecture read — “Microsoft’s Project Solara is an OS designed for agents instead of apps” — devices that run AI agents seamlessly around you, not just AI search. The Qualcomm AI-devices thread is in AI-RTZ #1015.

MP Take: These are the first genuinely new office-AI form factors we’ve seen beyond the consumer watches, wristbands, smart glasses and camera-AirPods — and for geek-heads like me, very cool prototypes to watch, with companies like Qualcomm and MediaTek important to track.

Questions

Q1 — What intrigues MP most about these office AI devices?

Entirely new form factors beyond the smartphone and other imagined AI devices for consumers— now enterprise sensors and cameras that observe and record for agents running on the devices, seamlessly integrating with laptops and more. A lot of software and hardware work ahead, and much of it is prototype hand-waving — but the possibilities in an office environment are ones we haven’t seen. Could be a differentiated hardware ecosystem with Microsoft’s dedicated support — the most logical party to bring such devices into the enterprise.

Q2 — What would be the chief concern?

Security — off the charts issues. AI agents running not just on your laptop and phone but on devices sensing and recording everything — imagine an office badge on your neck recording as you walk around getting coffee. The societal and regulatory issues of what’s recorded and controlled, plus an unprecedented cybersecurity and hacking surface. As cool as these devices are, it’ll take years to sort out managing them — societally, the creepiness factor, and securing the agents. A brave new world. Will take years to sort out.

Source Reading — For the Full Context

For the full context, see the canonical sources:

Shorts Clips from today

Clip 1 — AI’s Global Supply Chain Complexity

Watch on YouTube Shorts

We can’t manufacture all the AI-related hardware — especially for physical-world AI — just in the US. Not just iPhones and computers: AI-driven cars, robots, and drones all require the global supply chain. And it’s a very complex supply chain — most regulators don’t understand the depth and breadth of that complexity.

MP Take: Both countries need to figure out how to thread the needle. Neither the US nor China can do everything they want with AI without the other — it’s not just money; it’s capable workforces measured in the hundreds of thousands.

Clip 2 — US & China’s AI Infrastructure Race

Watch on YouTube Shorts

The US and China remain in pole position on AI infrastructure build-outs, despite all the regulatory and geopolitical issues. The US has the edge on AI chips, the latest roadmaps, and supply chains. China leads on the power front — its deployment is far ahead of the US — plus far deeper manufacturing ecosystems for the cars, robots and drones that people like Elon point to as trillion-dollar markets.

MP Take: It’s not buildable without China, without Taiwan and TSMC — regardless of how many hundreds of billions are invested in the US. It’s about capable people measured in the tens and hundreds of thousands, not just money.

Clip 3 — AI Devices: Privacy & Security Concerns

Watch on YouTube Shorts

Once you see Microsoft’s Project Solara office-AI devices, it becomes very clear the security concerns are off the charts. AI agents running not just on your laptop and phone, but on devices sensing everything, recording everything — imagine an office badge on your neck recording as you walk around getting coffee. An unprecedented cybersecurity and hacking surface.

MP Take: As cool as these devices are, it’ll take years to sort out how to manage them — from the societal, creepiness, and agent-security points of view. A brave new world.

Clip 4 — AI Data Center Delays & Pricing Surge

Watch on YouTube Shorts

US AI data-center schedules are slipping despite the best efforts of the Googles, Microsofts and Amazons — and Elon and Meta — with spend heading north of $2 trillion in the next year or two. Regulatory impediments are real, and the delays are pushing AI pricing from all-you-can-eat subscriptions to a-la-carte, with serious sticker shock for business customers.

MP Take: The timing gap between building plans and executed plans remains the principal volatility risk to the AI Tech Wave — especially now for public investors, not just the private ones. And this is going to accelerate, not decelerate.

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

Today’s AI-RTZ #1106 — Anthropic Mythos: Earning Mythic Pricing — on Anthropic’s ever-rising AI pricing for its latest model, Mythos. It ties into much of today’s discussion.

Tomorrow — ARD 90 on AI-RTZ #1107.

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

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