Content About Stocktwits
No Image Available

'Blind Ambitions Abound': Tech Incumbents in Unbound AI ‘Founder Mode’. ARD #110

Author
Michael Parekh AI: Reset to Zero
3 hours ago
19 MIN READ

Today’s theme: ‘Blind Ambitions Abound’— and Unbound. A special edition going into the 250th July 4th weekend of the founder-led country, the USA:

This AI Era tech incumbents are running in ‘Founder Mode’ — Y Combinator founder Paul Graham’s idea, now practically a religion in Silicon Valley — with the ferocity of the tech startups of prior waves, but the resources of established giants. Three events around Anthropic, Elon’s SpaceX/xAI, and Apple for the AI Tech Wave — each with my Take first, then my Overall Take.

(1) Anthropic’s Fable 5 À La Carte Pricing Ambitions

MP TAKE: The big bet by the frontier-model AI companies — especially by Anthropic, even more than OpenAI — is that they can move their core developer and enterprise users away from ‘all-you-can-eat’ subscription pricing to à la carte, usage-credit pricing for their top models. Look in the depths of the Fable 5 redeployment announcement and it’s right there in the fine print: as of July 7, even top-tier subscribers paying $20 to $200 a month — like yours truly, now on Fable 5 and loving it — get usage credits covering only part of the plan, with everything beyond on à la carte. Presumably the same will apply to Fable 5’s cousin Mythos 5 as it reaches non-gated customers. It’s the path to the most aggressive capture of frontier-AI economic market share — much like Apple capturing over 70% of the economic value in global smartphones on less than 30% of the unit share. That is the wager from Anthropic’s pole position in the white-hot AI Coding and Enterprise segment — Claude Code, Cowork, and the related array of AI applications — already at a revenue run-rate of over $50 billion. A five-year-old company: an AI startup, but a major tech incumbent, just like its sibling OpenAI.

And this bet is motivated by the preparations for a trillion-plus-dollar mega-AI IPO I’ve talked a lot about. OpenAI — which has raised even more money and has more compute on its roadmap — is inclined to be even more aggressive on price: likely also à la carte for its top 5.6 models as the government takes off the shackles, but priced lower to wean users from Claude Code and Cowork toward Codex and ChatGPT. To widen the lens: this à la carte ambition is breathtaking — it involves billions of dollars that early customers must decide to pay, or experiment with less expensive models from elsewhere — China in particular (as I covered yesterday in ARD #109). Normally a risky move — but they’re balancing tens if not hundreds of billions in AI compute investments against revenues, and the public markets’ enthusiasm for AI infrastructure may not always last.

Sources, in narrative order: Anthropic — Redeploying Fable 5. Digital Applied — Claude Fable 5 Pricing: The July 7 Usage-Credits Switch. WSJ — OpenAI considers drastic price cuts, anticipating a war for users with Anthropic. For longtime readers, in narrative order: ‘Anthropic & OpenAI’s Velvet Rope AI upgrade drift’ in AI-RTZ #1064; ‘Anthropic Mythos earning mythic cybersecurity Pricing’ in AI-RTZ #1106; ‘Anthropic and OpenAI have over half of $2 trillion plus in cloud company backlogs’ in AI-RTZ #1079; and yesterday’s ‘Anthropic & OpenAI in Stop-and-Go US Traffic’ in ARD #109.

(2) Elon’s ‘Not a Phone’ — Direct-to-Satellite AI Handset Ambitions

MP TAKE: The WSJ reports Elon showed prospective investors, ahead of the SpaceX/xAI IPO, a prototype handset that looked like a phone — xAI tech, its own operating system, a Snapdragon processor, a design slimmer than an iPhone. Recall his words, going back to last October: “The idea of making a phone makes me want to die.” Then: “But if we have to make a phone, we will.” And then: “We are not developing a phone.” A lot of word-sounds there — but given Elon’s unparalleled ambitions with all things AI, all those statements could be technically true. Because the device doesn’t technically have to be a phone: it could bypass every cellular carrier on planet Earth entirely — direct-to-satellite, off the thousands of satellites he’s launching — running ‘AI super-app’ services.

This would be a multi-year effort requiring tens if not hundreds of billions of additional dollars, barely in hand. But consider who we’re talking about: someone whose SEC S-1 filing articulates a $30 trillion AI market — equal to the leading economy in the world, the US. Someone who wants AI data centers in space, Terafabs making huge amounts of chips in space, and unimaginable numbers of humanoid robots and robotaxis on Earth. Of course, he can crawl before he walks and runs — small device volumes first as he did with Starlink satellite dishes & subscriptions— very much like what OpenAI is apparently planning with its Jony Ive-inspired acquihire venture and its own rumored AI smartphone. To widen the lens: over the next one to two years we may see two additional devices that look like phones, act like phones, and run their own operating systems — a new beast against Apple’s iPhone and Google’s Android — right in the teeth of the global memory supply constraints and ‘RAMageddon’ I’ve been covering. Daunting challenges, almost Quixotic ambitions. But that’s ‘founder mode’ at planetary scale.

Sources, in narrative order: WSJ — SpaceX showed investors a prototype of Elon Musk’s new AI device. Paul Graham — Founder Mode. For longtime readers, in narrative order: ‘OpenAI’s own OS-driven AI smartphone in 2027’ in AI-RTZ #1078; and ‘Push Comes to Shove on the Global Memory Crisis’ in ARD #108.

(3) Apple’s AI-Powered Bottoms-Up Hardware/Software Ambitions Through 2027 and Beyond

MP TAKE: There’s a general perception that Apple is just taking Google’s Gemini and making it available in a wrapper. That’s not what’s going on. Look at what Apple has technically done: there’s no Gemini code. Apple has built a family of Apple models that run on its own architecture — locally on Apple Silicon, and on an Apple version of a cloud with deep built-in trust, safety and privacy technologies (the cloud portion running on Google and Nvidia hardware, but all the technology and architecture is Apple’s own). It’s very similar to what Apple did in the mid-90s when it bought Steve Jobs back with NeXT — I was involved in that episode at Goldman Sachs, where we were working closely with NeXT on their plans in the public and/or private markets. Long story short: NeXT’s Unix-based technology became the fundamental technical basis of today’s Mac OS and other Apple software platforms. Apple built that over the following two-plus decades. This is a 50-year-old company acting like a tech startup — rethinking architectures on AI that will take 10 and 20 years to build. These new devices are the starting points.

And the devices are coming: iPhone 18 on the phone side, laptops with M6 and M7 chips between now and next year — potentially two MacBook releases — all optimized for AI with more memory, more neural compute, and a unified memory architecture friendly to local, unmetered running of models — freeing the user from some of the à la carte pricing I just talked about. To zoom back on the lens: this won’t be an overnight “oh my God, Apple is amazing at AI” moment. It’s one step at a time — but I think we’ll be impressed with Siri AI, and I do think these become the AI devices we’re hoping for. The most pragmatic-seeming AI ambitions may be the most ambitious of them all. They just don’t announce themselves with a roadshow.

Sources, in narrative order: AppleInsider — iPhone 18 Pro leaks: Qualcomm or Apple C2 modem, A20 details, camera upgrades. FT — Tim Cook and EU tech chief Henna Virkkunen held constructive talks on launching Siri AI in the EU and avoiding fines. Bloomberg — Apple readies new M6/M7 touchscreen laptops + redesigned entry iPad Pro for first half 2027. For longtime readers, in narrative order: ‘Apple partnership with Amazon Leo on direct-to-satellite smartphones’ in AI-RTZ #1057; ‘New Bounce for Apple Design in AI Age’ in AI-RTZ #975; and ‘Long-expected Apple Cook-to-Ternus CEO Shift Activated’ in AI-RTZ #1063.

MP OVERALL TAKE

All three companies above are showing unique sets of unbounded AI ambitions — and the ability to strike while the iron is hot is the key thing to focus on in the near term. They have the global resources to execute and see these things to full fruition. They’re acting in ‘Founder Mode’ — the idea Y Combinator founder Paul Graham wrote about eloquently, now a kind of religion in Silicon Valley (the essay is worth reading). And you don’t have to be a founder/CEO like Mark Zuckerberg at Meta or Elon at Tesla. You can be a 50-year-old company like Apple, with a management team that’s been together for decades, utterly instilled in the institutional values — the core North Star — of its founder. Even John Ternus, who takes on the Apple CEO role in September, has been there 25 years: half of Apple’s existence. These are founder-mentality companies.

The broader theme: today’s tech incumbents are all investing in AI like tech startups, like founders — there are just more zeros. They’re investing hundreds of billions of dollars, raising hundreds of billions in debt and equity. And it holds even where professional managers run the companies — the Satya Nadellas, Sundar Pichais and Andy Jassys going up against the founder/CEOs. I always draw the delineation between founder/CEOs and professional managers, but this time it’s important to recognize: they are more similar than different. Worth underlining as we celebrate our founding-fathers-led creation, the United States of America, and wish it a happy 250th birthday this weekend.

Gadget AI — Meta’s AI Smart-Glasses Subscription Ambitions

MP Take: The Verge has another pointed, pithy piece — Zuckerberg wants to experiment with a subscription plan on his AI smart glasses: roughly $20 a month for additional AI features (how the glasses recognize and amplify conversations, presumably more). All of these companies are trying to figure out how to cover the variable costs of services that consume huge amounts of AI inference compute — hence the added charges on devices that are already expensive. A fancy pair of AI smart glasses runs $400-500 — and because they’re glasses, you’ll want two or three versions (sunglasses, colors). So they’re asking regular folk and early adopters to make $1,000-2,000 investments in a world where $300-400 gadgets are becoming $600-700++ between now and next year — and then asking another $120-200+ a year on top. Bold and risky.

But Mark is doing what he always does: throwing things against the wall — services, then pricing — seeing what sticks, and Scaling stuff that sticks really fast. To widen the lens: he’s up against smart glasses from Google in the next few weeks, Amazon talking about them, Apple planning them next year, and Snap’s $2,200 AR glasses already out. Meta has the biggest install base so far — over 7 million — though we don’t know how many are sitting in a drawer. That’s Mark’s core challenge: getting people to use these things for hours on end, creating daily habits — and subscription tiers of do-this-and-do-that are tough. That’s why I’ve always viewed them as friction points — for gadgets, and for language models alike, which was the earlier part of today’s discussion.

Sources, in narrative order: The Verge — Meta is adding ridiculous ‘rate limits’ and a soft paywall to its smart glasses. For longtime readers: ‘Meta leans in on AI Smart Glasses’ in AI-RTZ #849.

Questions

Q1 — What is the one AI service MP is fine paying for today?

Truly useful and reliable AI agentic services. I’m experimenting with the higher-end subscription tiers — OpenAI, Anthropic, Google, Perplexity’s Comet product, and a number of others — as they all migrate toward ‘OpenClaw’-style agent services at the $100-200 tiers, and now start adding à la carte on top. The companies are doing what they have to do — charging for the variable cost of these services. I’m fine paying for them as long as I keep seeing the value. But given the unbounded costs of à la carte usage, you’ve got to watch with a keener eye than you would otherwise. That’s my core discipline with products I like and lean into.

Q2 — What is the one AI service MP would NOT be paying for yet?

I’m not sold yet on the à la carte pricing model for frontier models — so I’ll be very careful with the Fable 5 pricing, as an example. If it’s useful, I’ll justify it. But these services are in what I call the honeymoon phase: early adopters enchanted that these AI agents can do what they do — a little bit of ‘the parrot can talk!’ Everyone’s experimenting. We went very rapidly from token maxxing to token budgeting — I’ve written a lot about that. Meta itself swung in two months from leaderboards celebrating engineers using hundreds of thousands of dollars of internal token compute to dashboards telling them to manage within a budget of X thousand per developer. The only time I can recall this kind of environment is the early days of online services in the mid-90s: America Online’s all-you-can-eat subscription against CompuServe’s à la carte model. CompuServe was the consumer-online leader with above average industry margins; AOL took the bulk of the fast developing, new consumer online market within five years and got big enough to buy the then-biggest media company in the world, Time Warner. The rest, as they say, was history. I bring it up because I was there — in those rooms, at the front table — and I’ve seen this tussle between à la carte and all-you-can-eat before. We’re seeing it again, across this range of AI services. With far more zeros.

Wrapping up.

Today’s AI-RTZ #1135 — Meta & ‘Zuck’ aiming for the AI Clouds — Meta planning an AWS-style cloud business off its massive AI infrastructure investments (’ZWS’ — Zuck Web Services, tongue in cheek — like Elon’s ‘EWS’). We’re especially seeing today’s theme with Meta, a founder/CEO-led company: up almost 10% — over $100 billion in market cap — on the rumblings, something I flagged as a possibility back in August 2024, in AI-RTZ #456. The two ‘neocloud’ startups CoreWeave and Nebius collectively lost billions the same day. Investors rewarding tech incumbents acting like tech startups. Recommended as today’s reading post.

Tomorrow — ARD 111 on AI-RTZ 1136. Despite the holiday weekend.

Wish you all the happiest of July 4 Weekends!

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

Full Source Reading —

For the broader context, see the canonical sources for ARD 110 — in today’s narrative order:

Event 1 — Anthropic’s Fable 5 À La Carte Pricing Ambitions

Event 2 — Elon’s ‘Not a Phone’ Direct-to-Satellite AI Handset

Event 3 — Apple’s AI-Powered Bottoms-Up Hardware/Software Ambitions

Gadget AI — Meta’s AI Smart-Glasses Subscriptions

Clips from today

Segment — AI Pricing: The Honeymoon Phase

Watch on YouTube

Frontier AI pricing is in its honeymoon phase — early adopters enchanted that the AI agents can do what they do (’the parrot can talk!’), while à la carte, usage-credit pricing arrives from Anthropic and OpenAI.

MP Take: The industry went from token maxxing to token budgeting in two months — Meta swung from engineer leaderboards to spend-cap dashboards. The only precedent I know first-hand, from the front table: AOL’s all-you-can-eat against CompuServe’s à la carte in the mid-90s. AOL took the market — and bought Time Warner. That tussle is replaying across AI services right now.

Clip 1 — Apple’s AI Ambitions: Not Just Google’s Gemini

Watch on YouTube Shorts

The perception is Apple just wraps Google’s Gemini in its AI. Technically, that’s not what’s going on — there’s no Gemini code inside.

MP Take: Apple built its own family of AI models on its own architecture — local on Apple Silicon, with deep trust, safety and privacy technologies built in. Very much like the NeXT deal that became the Unix roots of today’s Mac OS. A 50-year-old company acting like a tech startup, building AI architectures that take 10-20 years. The most pragmatic-seeming AI ambitions may be the most ambitious of them all.

Clip 2 — Meta’s Zuck Web Services: A New AWS?

Watch on YouTube Shorts

Meta may rent out excess AI compute, AWS-style — call it ‘ZWS,’ Zuck Web Services, like Elon’s ‘EWS’ at SpaceX/xAI, which signed $70B in compute deals with Anthropic and Google days before its mega-AI IPO.

MP Take: I wrote about Meta’s AWS-size opportunity two years ago — now it’s closer to reality. Investors flashed green: Meta jumped almost 10%, over $100 billion in market cap in a day, while neoclouds CoreWeave and Nebius fell double digits. Markets are rewarding tech incumbents acting like tech startups.

Clip 3 — AI Pricing: From Token Maxing to Budgeting

Watch on YouTube Shorts

The AI industry swung from token maxxing to token budgeting in about two months — Meta went from leaderboards celebrating engineers burning hundreds of thousands of dollars in internal AI compute to dashboards capping spend per developer.

MP Take: That swing is the tell on à la carte AI pricing. Businesses everywhere are asking the same question as frontier labs move from all-you-can-eat subscriptions to usage credits: what is this actually worth, day in and day out? A question mark on pricing that ties directly to the ambitions of Anthropic and OpenAI.

Clip 4 — AI Services: Value vs. Cost

Watch on YouTube Shorts

What AI is worth paying for today? The $100-200 agentic tiers from Anthropic, OpenAI, Google and Perplexity are the closest examples — and now à la carte usage pricing is arriving on top.

MP Take: I’m fine paying for these services as long as I see the value — but given the unbounded costs of à la carte usage, you’ve got to keep a keener eye than usual. We’re still in the honeymoon phase: early adopters basking in ‘the parrot can talk!’ wonder. The fine-grained reporting, monitoring and control of how AI agents burn tokens doesn’t exist yet — that discipline is the missing piece.

Clip 5 — Tech Giants in ‘Founder Mode’ for AI

Watch on YouTube Shorts

Today’s tech incumbents are running in ‘founder mode’ — YC founder Paul Graham’s idea that’s now practically a religion in Silicon Valley — with the ferocity of the startups of prior tech waves.

MP Take: Anthropic, Elon’s SpaceX/xAI, and Apple are all showing unbounded AI ambitions with global resources to execute. And you don’t have to be a founder/CEO — Apple’s 50-year-old team is utterly instilled in its founder’s North Star. Today’s incumbents invest like founders, just with more zeros. Fitting, as the founder-led USA turns 250 this weekend.

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 #1135.

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 #110.

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 — Anthropic’s Fable 5 À La Carte Pricing Ambitions:

Take 2 — Elon’s ‘Not a Phone’ Direct-to-Satellite AI Handset:

Take 3 — Apple’s AI-Powered Bottoms-Up Hardware/Software Ambitions:

Overall Take — Founder Mode:

Gadget AI — Meta’s AI Smart-Glasses Subscriptions:

Q1 + Q2 — MP on paying for AI services:

  • (MP’s own analyst view — AOL/CompuServe mid-90s history first-hand)

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

Cadeler Powers To An All-Time High
Cadeler Powers To An All-Time High
Riley Rosebee Trends with No Friends
Good evening. Welcome to Trends with No Friends. FOR THOSE JUST JOINING US, HERE’S AN OVERVIEW Markets marched higher. Every major index closed positive....
2 years ago
3 MIN READ
BROWSE ALL ARTICLES