Big Tech Goldilocks AI Strategies: Large, Medium, Small & 'Just Right'. ARD #85

Michael Parekh AI: Reset to Zero

16 hours ago

17 MIN READ

Today’s theme: Big Tech rolling out their latest AI models and infrastructure strategies — large, medium, and small. All the Goldilocks AI bowls are ‘just right’ for now. Three Takes today on the three Goldilocks bowls at the frontier-model, middle-tier AWS-style, and edge / local AI ends of the spectrum.

Each of these strategies makes sense given what’s going on around those specific arenas. The frontier-model companies are on a very rapid clip of product iteration that customers haven’t seen in most tech waves before. The middle-tier — Meta and Elon’s SpaceX/xAI — are leveraging their multi-hundred-billion-dollar AI Data Center build-outs into AWS-style enterprise plays. And Apple — leveraging its unparalleled global supply chain across its 2B+ device base — is doubling down on its ‘mainframe’ to ‘local’ AI edge strategy.

(1) Anthropic + OpenAI — Rolling Out Latest Frontier Models

The Large (LLM) frontier-model and AI Infrastructure bowl. Both Anthropic and OpenAI shipping their latest flagship models this week.

The Information laid out the Anthropic detail — “Anthropic releases new flagship AI model” — confirming Opus 4.8, the latest iteration of the Claude flagship line, just 41 days after 4.7. Mythos-class model expected in a few weeks.

TechCrunch ran the product-feature read — “Anthropic releases Opus 4.8 with new dynamic workflow tool” — covering the new dynamic-workflow-tool capability that allows big enterprises to implement multi-hundred-agent workflows and manage them all. These are exactly the kinds of things that all of these companies will have to do as they really get ready with these products for enterprises — because all of them have found their product-market fit in AI coding, then workflow applications on top of that.

Axios carried the Mythos timing detail — “Anthropic Opus release: Mythos out in a few weeks” — confirming the Mythos-class model is next-up. Mythos was held back earlier because of cybersecurity concerns, with Anthropic discussing the mitigations in earnest with governments, agencies, and companies at the highest levels.

OpenAI shipped GPT 5.5 recently — the flagship iteration ahead of the September IPO filing. With more ahead before then.

MP TAKE 1

The frontier-model companies keep their own on the latest iteration of their flagship models. Along with the top AI coding versions of the AI apps based on those models.

The middle-tier companies — Meta and Elon’s SpaceX — in the meantime are rolling out their AI models — Muse Spark for Meta, and Grok 5 for Elon soon — along with a more AWS-type AI Data Center strategy to help defray the expense on both the model and infrastructure fronts.

In the meantime, Apple is leveraging its strongest assets — over two billion local devices with an integrated tech stack chips on up — with smaller AI models leveraging larger models in the cloud with privacy and security.

These are trends in their early phases, and will take through 2030 to play out.

(2) Meta — Focusing on the AWS-Type Opportunity

The Middle AI bowl. Meta explicitly puts a cloud computing business on the table — pursuing an AWS-style enterprise strategy I’d actually written about over three years ago. Good things take time.

The Information laid out the the enterprise-AI angle — “Meta launches new enterprise push to boost business adoption of AI tools” — Meta kickstarting an enterprise AI business layer on top of its multi-hundred-billion-dollar AI Data Center build-out.

CNBC carried the Zuckerberg quote — “Mark Zuckerberg says a Meta cloud computing business is ‘definitely on the table’” — the explicit on-the-record signal that Meta is looking at the AWS-style monetization-of-infrastructure play. Recognize that Meta will spend over $145 billion this year on AI Data Center infrastructure — a lot of which it will use across its 3.5B+ user properties (Facebook + Instagram + WhatsApp) and to serve sophisticated AI-driven ads. But there’s excess capacity. Defray the costs by getting business customers — that is the strategy.

We saw an echo of this from Elon Musk, who literally — ahead of his SpaceX/xAI mega-IPO later this month — signed a $40+ billion contract with Anthropic, his arch rival, to let Anthropic use Grok/Colossus data center infrastructure to serve their models. Lots of frenemies action here.

MP TAKE 2

Timely move as Meta ramps up its multi-hundred-billion-dollar AI Data Centers. Logical move, and follows the pattern of Elon Musk going on a similar path with SpaceX/xAI/Tesla AI Infrastructure.

This expands players and competition against the top three — Amazon AWS, Microsoft Azure, and Google Cloud. Not to mention the neoclouds like Oracle, CoreWeave, and others globally.

This segment is one of the most essential in the AI Tech Stack, as AI Compute is built up globally through the decade. Margins and growth remain high for now, and should maintain this pace in the near-term.

(3) Apple — Focusing on Its Edge, Local AI Compute

The Small AI bowl (small language models, SLMs) at the Edge. Apple’s renewed push for AI that runs on devices, not in the cloud. Their WWDC is coming up on June 8 — imminent. And they’ve been long expected to partner with Gemini to power their AI version of Siri, which has been delayed for months.

The Information ran the strategy piece — “Apple to renew push for AI that runs on devices, instead of the cloud” — laying out Apple’s edge-first repositioning.

Bloomberg carried the WWDC June 8 AI overhaul preview — “Apple’s iOS 27 + Revamped Siri + Pro Camera App + New AI Features at WWDC 2026” — covering the on-device AI features landing on iPhone, Photos, Screenshots, and Siri across the iOS 27 platform overhaul.

Apple expected to leverage their really unique vertical tech stack — all the way from Apple Silicon to the devices that sell to over 2.5+ billion users across computers, laptops, desktops, smartphones, and wearables. Especially now on September 1, when Tim Cook is elevated to Chairman on the board, and the two senior executives — John Ternus and Johny Srouji — will basically be running the company on the hardware and software side. Both have very deep hardware supply chain backgrounds, thanks of course to Tim Cook’s scale.

MP TAKE 3

This has been the path for a while now, but needed sufficient development of small language models — SLMs — that could be far more competitive versus LLMs.

That is starting to happen, along with local AI GPU/CPU chips and memory to run them locally with privacy and low latency. That is happening now and will take the next few years to 2030 to play out.

From my perspective, Apple remains the best positioned in this ‘mainframe’ to ‘local’ AI opportunity — particularly given its unparalleled global supply chain at a time of global geopolitics related trade and tariff issues. PC volumes already down 10, 15, 20 percent in 2026 alone on higher prices for computers and laptops — Apple’s getting more competitive on that front.

Even the bigger tech companies lacking their own hardware/supply chains will not be able to emulate Apple’s strategies here at scale.

Gadget AI — Oura Ring 5: 40% Smaller, 15%+ More Expensive, Ahead of 2026 IPO

Oura is the AI healthcare company with one of the strongest brands for AI smart rings — and a potential 2026 IPO candidate. They just announced Ring 5, which is 40% smaller than the previous ring and about 15% more expensive, now closer to $400, ahead of their 2026 IPO.

Bloomberg broke the product detail — “Oura Ring 5: 40% smaller, more like a regular ring, hypertension and sleep apnea” — confirming the new ring is meant for early adopters into exercise, athletics, healthcare, and continuous monitoring. They’re cramming more patented healthcare features into a simple ring — including focus on detecting blood pressure changes, hypertension, sleep apnea, and lots more.

That will of course complement other things being developed by other companies — like the wristbands I talked about yesterday from Google’s Fitbit Air and the Whoop Band (also looking at an IPO). Apple has ambitions here, Meta has ambitions here, Amazon, etc. Everyone’s going to be involved — the elephants and the smaller companies — but it’s a very active area, and very early days.

MP TAKE — Gadget AI

A category just beginning — complementing AI wrist bands, smart glasses, and a lot more to come.

Questions

Q1 — Does MP wear an AI Smart Ring?

Answer: Not yet.

I continue to rely on my Apple Watch Ultra. It’s good enough with three rings. But I am kind of tempted by these newer gizmos — if anything, just to try them. So I might be tempted with an Oura Ring.

Q2 — What’s holding MP back?

Answer: Another siloed device and app to maintain, track, and coordinate with all the AI health apps.

To me, the bigger issue is less about the device itself — rings are rings, they’re pretty straightforward. People have preferences (design, style, how many rings, wedding ring, etc.). But the bigger issue across these devices is that this is another siloed device and app to maintain and track and coordinate with all the AI health apps. And one of the issues across this category is that all these companies — especially the large ones — are siloed. They want everything in their walled garden, and then it becomes more cognitive load on the user to manage what my ring app says versus my wrist app versus my watch app.

MKBHD just had a review across just three AI fitness wristbands to watches — the Apple Watch, the Whoop Band, and the Google Fitbit Air — and the metrics were different across all three. If you’re really into this stuff, that becomes additional cognitive load. There’s very little out there that helps people support this stuff and make sense of it horizontally.

MP Take: I would expect at some point we will have third-party AI apps that — with permission — get the data horizontally from these various silos and help users make heads or tails of it. So they can actually get a lot more signal from the walled silos and the noise across these devices. A friction point potentially solved by industry consolidation over time.

Source Reading — For the Full Context

For the full context, see the canonical sources:

Shorts Clips from today

Clip 1 — Oura Ring 5: AI Healthcare IPO Candidate

Watch on YouTube Shorts

Oura Ring 5 just launched — 40% smaller, 15%+ more expensive, now closer to $400, with new hypertension and sleep-apnea monitoring features. Oura is also one of the most-talked-about AI healthcare IPO candidates for 2026, alongside Whoop. A category really meant for early adopters into exercise, athletics, and continuous health monitoring.

MP Take: A category just beginning — complementing AI wrist bands, smart glasses, and a lot more to come from Apple, Meta, Amazon, and others ahead.

Clip 2 — Apple’s Goldilocks AI Strategy

Watch on YouTube Shorts

Apple is focused on its side of the AI Goldilocks spectrum — local edge devices that run local small models and leverage the larger models from Gemini and presumably others in the cloud as needed, with the appropriate privacy, trust, and security that Apple is much more focused on than all the other competitors. They don’t have as big a business to protect on the advertising side, where Google and Meta do in spades. They do have very large ambitions on subscription services, which can run off blended local + cloud services.

MP Take: Apple remains uniquely positioned with its supply chain versus everybody else. Hardware is going to be one of the toughest things to scale at hundreds of millions of devices per year — like clockwork — with all the geopolitical issues around supply chains.

Clip 3 — AI Smart Rings: Tempting, But Siloed

Watch on YouTube Shorts

Do I wear an AI smart ring? No — I still rely on my Apple Watch Ultra. It’s good enough with three rings. But I’m tempted by these newer gizmos. The bigger issue holding me back isn’t the device itself — rings are straightforward. It’s that this is another siloed device and app to maintain, track, and coordinate with all the other AI health apps. All these companies — especially the large ones — want everything in their walled garden.

MP Take: It becomes more cognitive load on the user to manage what my ring app says versus my wrist app versus my watch app. At some point we’ll have third-party AI apps that — with permission — get the data from these various silos and help users make heads or tails of it. The industry needs to figure this out horizontally.

Clip 4 — Apple’s Supply Chain Edge AI Moat

Watch on YouTube Shorts

Apple remains uniquely positioned with its supply chain. Hardware is the toughest thing to scale. It’s straightforward to do millions of devices like Google has with Pixel — but consistently scaling hundreds of millions of devices per year like clockwork, with geopolitical issues around supply chains, especially in China — there are very few companies that can manage that at scale. Especially when memory prices and supply constraints are going up across almost any category. PC volumes already down 10-20% on higher prices this year.

MP Take: That is Apple’s big advantage. They’re focused on local edge devices that run small models locally, with appropriate privacy, trust, and security. Even the bigger tech companies lacking their own hardware/supply chains will not be able to emulate Apple’s strategies here at scale.

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

Tomorrow — ARD 86 on AI-RTZ #1101.

Thanks for joining, 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.