Content About Stocktwits
No Image Available

Drumbeats of AI Progress March On: ARD #120

Author
Michael Parekh AI: Reset to Zero
18 hours ago
15 MIN READ

Today’s theme: the drumbeats of progress march on — one of my favorite phrases from literature. The day-to-day drumbeat of AI isn’t standing still; it keeps marching, across borders and across companies large and small. Three events make the point — TSMC, the ‘Fed of the AI Tech Wave‘; China’s Moonshot; and 1Password — each with my Take, then my Overall Take. Plus a Gadget AI on how the revamped Siri AI is already changing how I use my iPhone, and two questions. Let’s get started.

(1) The ‘Fed of the AI Tech Wave’ Reports Strong Numbers and Capacity Hikes — TSMC

MP TAKE: I’ve talked for a while about how TSMC — the key gating company to the AI Tech Wave — plays a ‘Fed’-like role, carefully tempering how fast it grows capacity against the extraordinary global demand for chip fabs. Especially from its two biggest customers, Nvidia and Apple in that order — it flipped in the last year), along with everyone else. This latest report shows management leaning in a bit more than of late. The core words from CFO Wendell Huang should resonate around the world: ‘Our conviction in the AI megatrend is very strong.’

They now guide capex of $60-to-$64 billion for 2026 — about $4 billion above the earlier plan — and revenue growth above 40%, up from the 30%-plus predicted earlier, with Q2 profit up 77%. Alongside the results, TSMC announced another $100 billion for the US, taking its total US investment north of $265 billion — directionally correct in this geopolitically sensitive world, and welcome in DC ahead of the second Xi-Trump meeting in September. Taiwan sits at the crucible of the most complex supply chain humanity has built; Nvidia is threading the two halves of the globe. The ‘AI Fed’ has spoken — better late than never — and for global tech and AI, at least short term, that’s a positive step.

Sources, in narrative order: Bloomberg — TSMC Hikes Sales, Spending Outlook to Catch AI ‘Megatrend’. CNBC — TSMC to Invest Additional $100 Billion in Arizona After Q2 Profit Soars 77%. Bloomberg — TSMC to Spend $265 Billion on US Buildout in Key Trump Deal. For longtime readers: ‘TSMC, Fed of the Global Chip Economy’ in AI-RTZ #1108; and ‘TSMC, The Fed of the Tech Economy’ in ARD #57.

(2) China’s Moonshot AI Nears Anthropic’s Opus 4.8 — the Kimi K3 Model

MP TAKE: Moonshot remains one of the relatively unheralded China AI startups, but its open-source Kimi K models have been watched the world over. This new Kimi K3 — a 2.8-trillion-parameter model just released (open weights to be released July 27) — is a big iteration, and it looks close to Anthropic’s Opus 4.8 (itself below the flagship Fable 5 / Mythos 5). Crucially, it’s open-weights — the world can inspect it, host it anywhere and use it as it wants, independent of China — and Moonshot could come in at a third of the price or less of Anthropic Claude Opus 4.8.

As I’ve written lately, US companies and customers are leaning into alternatives to the frontier models — not as binary replacements, but as supplements in a portfolio approach, balancing cost, privacy and capability. Note GLM 5.2 from Chinese lab Z.ai too. Between Moonshot, DeepSeek, Alibaba and others, the perceived gap between the two countries looks narrower than one might think. This shouldn’t be viewed as a race: it’s a global undertaking where bottom-up cooperation between US and Chinese companies grows the pie and makes horizontal platforms possible — while geopolitically-driven separation shrinks it. Microsoft and others are already leaning into Chinese models, as they should. And the frontier labs like Anthropic will of course push back — on geopolitical fears, ‘distillation’ charges, and the general bogeyman of open-source AI — all antithetical to where their bread is buttered.

Sources, in narrative order: FT — ‘Chinese AI Start-Up Moonshot to Launch Model Challenging Anthropic’s Lead’ (FT link — confirm URL). For longtime readers: ‘China’s Post-Meta/Manus Changes for Foreign Investors’ in AI-RTZ; ‘Looking Beyond Space and AI Races’ in AI-RTZ #1138; ‘Anthropic’s Blip 2.0 in New Form as China AI Ramps’ in AI-RTZ #1131; and ‘Nvidia’s Jensen Rides Air Force One to China After All’ in AI-RTZ #1085.

(3) A Timely AI-Agent Accommodation — 1Password Lets Claude Sign In Without Seeing Your Passwords

MP TAKE: As AI agents become the core thing everyone wants to do with AI — in the enterprise, with developers using them by the gallon of compute — they’re increasingly sent out onto the internet to access websites and tools on their users’ behalf. Problem is, the internet is largely designed to keep them out: most sites try to block ‘bots.’ So it has to be rebuilt for agents — not overnight, not top-to-bottom, but incrementally.

This move by 1Password — one of the top password managers — is exactly the kind of technical innovation needed: they’ve re-architected the front and back end so Claude Cowork can sign into your websites without ever being given the password. The WSJ’s tech analyst tried it, with some trepidation, on her retirement accounts — and it went as planned. It’s a small, incremental step, but the kind we’ll need across thousands of companies over the next few years. The internet for AI agents will be built side by side with the internet for humans — and it’s a great, widely-used case study for what’s needed, under proper user supervision and approvals.

Sources, in narrative order: 9to5Mac — 1Password Now Lets Claude Sign Into Websites Without Seeing Your Passwords. WSJ — I Gave an AI Agent Access to My Passwords. Here’s What Happened. For longtime readers: ‘AI Agents Increasingly Need an Internet of Their Own’ in AI-RTZ #1023; ‘A Better Internet for Both AI and Humans Ahead’ in AI-RTZ #823; ‘To a New AI-Powered Internet’ in AI-RTZ #739; and ‘AI Agents Not Designed for Our Current Internet’ in AI-RTZ #702.

MP OVERALL TAKE

Each of today’s events highlights how the drumbeat of progress marches on — especially with AI — across borders and across companies large and small. From the minute detail of how 1Password accommodates AI agents, to the next generation of Chinese models catching up pari passu with the best from the US despite the geopolitical headwinds, to the granddaddy of chip manufacturing, TSMC, expanding capacity and fabs around the world where it must.

Despite all the geopolitical and mainstream-user fears, and the fair questions about ultimate utility, the day-to-day progress continues — and it’s far more widely dispersed than just the two frontier AI companies that soak up most of the attention lately. It runs from the bottom level to the top, all for a good cause: the AI Tech Wave moving on. The drumbeat doesn’t depend on any one player, and everyone is marching on around the world— and that, to me, is the reassuring part.

GADGET AI

The Revamped “Siri AI Is Already Changing How I Use My iPhone”

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

An echo of yesterday’s show: I talked about the strong review of Siri AI on the Apple Watch; today The Verge has another hands-on — this time on Siri AI on the iPhone — and how it’s fundamentally changing how the reviewer uses their phone. That’s a big step, because Siri has been the relative joke of AI assistants for a decade, especially against Google’s Gemini and Amazon’s Alexa. But Siri AI is holding its own — because architecturally it’s different: local models on your phone, working on your own data, communicating with your watch and, over time, your AirPods, in an ecosystem.

And remember, this is still the beta — there’s now a public beta of iOS 27 for early adopters (the developer beta is the riskier one; try it on a backup iPhone if you’re curious). The final version ships in September. We may end up with a decidedly different mainstream take on consumer AI by year-end — as millions of regular folks get to try this AI stuff, Apple style, for themselves. With safety, security and privacy to boot.

Sources, in narrative order: The Verge — Siri AI Is Already Changing How I Use My iPhone. For longtime readers: ‘Apple Intelligence & Siri AI Hum with Google & Nvidia Inside’ in AI-RTZ #1113; ‘Siri AI Ready for Takeoff This Fall’ in AI-RTZ #1112; ‘Apple’s Ambitious App Intents for Siri AI’ in AI-RTZ #811; and ‘Siri AI Gets More Useful on the Apple Watch’ in ARD #119.

QUESTIONS

Q1 — What’s Michael’s favorite Siri AI feature on the iPhone, using the developer (now public) beta?

Answer: Something subtle. When you install the beta, the system indexes your phone — all the Apple apps: Notes, Mail, Photos, Messages, Reminders. My word for that is training — it’s training on your local data, right on your device, not for years in a data center in the sky. Right now it’s the first-party apps; when the final software ships in September, App Intents will open it to third-party apps too. That’s when consumer AI gets far more interesting — a real glimpse of what AI can do for you when it’s trained on your own local data, with attention to safety, privacy and trust.

Q2 — What’s the biggest negative thus far?

Answer: Every knife has two edges — and the same indexing/training is the downside. Right now, in the beta, there’s no indicator of how long it will take (in a couple of my cases it ran for days, sometimes a week or longer, because the index is built locally on each device). I’m hoping that compresses by September — but what I’d really like is a dashboard telling me the progress, made continual, since your data keeps changing and the model needs to keep re-indexing. It uses your phone (and can affect battery), so you should be able to see it, know about it, and maybe control pieces of it. Solve that, and this local training on your personal device is a very big step for consumer AI — and AI in general — after four years of one dominant form of it.

WRAP

Today’s AI-RTZ #1149 — Thinking Machines Joins the US Open-Weight AI-Model Companies List — on Thinking Machines (founded by ex-OpenAI senior executives, with Nvidia an investor) releasing its open-weight model ‘Inkling.’ It doesn’t compete with the big boys, but it’s a flexible tool with interesting quirks — and it extends the thread of Nvidia and Reflection AI as US open-source bellwethers, alongside Apple.

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

Tomorrow — ARD 121 on AI-RTZ 1150.

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

— MP

Full Source Reading —

For the broader context, see the canonical sources for ARD 120 — in today’s narrative order:

Event 1 — TSMC’s Strong Numbers + Capacity Hikes

Event 2 — China’s Moonshot / Kimi K3

Event 3 — 1Password’s AI-Agent Accommodation

Gadget AI — Siri AI on the iPhone

Clips from today

Clip 1 — Local AI Training on Personal Devices

Watch on YouTube Shorts

Siri AI’s most unique feature is the local ‘training’ — the on-device indexing of your own data (Notes, Mail, Photos, Maps, Reminders) that happens right on your iPhone, not in a data center in the sky.

MP Take: This local training on your personal data is the big step for consumer AI. Most models we know, like ChatGPT, train for years on huge data centers; here it happens on your device, privately. Right now it’s first-party apps; with App Intents this fall it opens to third parties — and that’s when consumer AI gets far more interesting.

Clip 2 — AI Agents: Redesigning the Internet

Watch on YouTube Shorts

AI agents are increasingly sent out to access websites and tools on our behalf — in a world where the internet was built to keep bots out. 1Password’s new move lets Claude sign into your sites without ever seeing your passwords.

MP Take: Exactly the kind of technical innovation we need — accommodating AI agents with proper user supervision and approvals, not summarily blocking them. A widely-used password manager is a great case study. The internet for AI agents will be built side by side with the internet for humans, incrementally, over the next few years.

Clip 3 — Siri AI on iPhone: A Game Changer?

Watch on YouTube Shorts

The Verge has a hands-on with Siri AI on the iPhone — and says it’s fundamentally changing how the reviewer uses their phone. It’s still the beta; the final version ships in September.

MP Take: Siri has been the relative joke of AI assistants for a decade, but Siri AI is holding its own — because architecturally it’s different: local models on your phone that work with your own data, communicating with your watch and, over time, your AirPods, in an ecosystem. That’s the key, and why it’s a big step for consumer AI.

Clip 4 — Apple’s Local AI Training: A Game Changer

Watch on YouTube Shorts

When you install the Siri AI beta, your iPhone starts indexing your data locally — Apple calls it indexing; I call it training. It’s the longest, most important step, and it happens privately on your device.

MP Take: The one negative right now: there’s no indicator of how long the indexing takes, and it needs to keep re-indexing as your data changes. I’d love a dashboard showing progress, plus some controls — it uses your phone and can affect battery. Solve that, and this local training on your own data is a very big step for consumer AI.

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 #1149.

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 #120.

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 — TSMC’s Strong Numbers + Capacity Hikes:

Take 2 — China’s Moonshot / Kimi K3:

Take 3 — 1Password’s AI-Agent Accommodation:

Gadget AI — Siri AI on the iPhone:

Companion text:

(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.





Want the latest?

Sign up for Michael Parekh's Newsletter below:


Subscribe Here

More like this

BROWSE ALL ARTICLES