Lines in the Sand being Re-drawn & Entrenched. ARD #98

The theme today: Lines in the Sand being Re-drawn and Dug Into. Strategies are being adjusted on the fly, and in some cases entrenched — by the major AI players and the US government alike. The backdrop: SpaceX, post-IPO, is now nudging close to $3 trillion and pulling on Amazon’s and Microsoft’s valuations, with investors front-running its expected NASDAQ-100 entry in two weeks— leaving two mega-AI IPOs still ahead this year. Three events, each with my Take first — and my Overall Take.

(1) The Anthropic Blip 2.0 Runs Longer Than OpenAI’s Blip 1.0

MP TAKE: I’ve been calling the Anthropic / US-government kerfuffle the ‘Blip 2.0’ — and it’s now running longer than Blip 1.0, the weekend drama three years ago when OpenAI’s non-profit board fired and re-hired Sam Altman and froze the whole industry. That one resolved in days. This one is still being negotiated — Anthropic has flown a team to Washington, but both sides seem to be digging in while negotiating. Honestly, it almost feels like watching the US and Iran negotiate around the Strait of Hormuz — let’s hope it doesn’t stretch to a hundred plus days.

Anthropic is rapidly figuring out how and where to re-draw its lines with the US government — both on the cybersecurity questions around its top Mythos and Fable 5 models, and on its overall relationship with Washington. The hard part is threading the needle between its core safety principles — deep in its founding DNA — and the imperatives of its business ahead of a planned mega-AI IPO this year. Anthropic will probably have to give some accommodation to get the ban lifted. In its corner is the strength of the technology and its unique leadership position, even with rivals running flat-out to catch up — so the leverage, for now, is still Anthropic’s. But the margin of error keeps narrowing. And the broader, unintended consequence — for every US tech and AI company’s ability to sell globally — is the real story. Note that OpenAI has a similar cybersecurity-vetted upgrade coming, and Google Gemini too; it’s not yet clear whether Washington takes a closer look at those as well.

Sources, in narrative order: Axios — Anthropic struggles to speak the Trump administration’s language. Axios — Anthropic export ban sounds alarms for the AI industry. Washington Post — how Anthropic lost the White House’s trust — and then its flagship product. For longtime readers: ‘Anthropic Flailing over Fear of Mythos & Fable 5’ in ARD #97, and Sam Altman re-hired after ‘the Blip’ three years ago in AI-RTZ (’Humpty Dumpty Still on the Wall’).

(2) Microsoft’s Satya Nadella Sides With Enterprise Customers vs the Frontier-Model Companies

MP TAKE: Microsoft is drawing a new line between itself and its former close partner, OpenAI — and putting itself on the side of its enterprise customers. Even though Microsoft still has access to OpenAI’s IP for the next five years, Satya Nadella’s essay, “A frontier without an ecosystem is not stable,” lands the argument plainly. The core line worth highlighting: “What is at stake is not some digital tool or system and its use, but how organizations continue to learn, build IP, differentiate, and thrive in a world where AI models can continuously absorb the expertise of humans and organizations and commoditize it.”

In other words: the frontier models take customer data as those customers use them, make their own models better, and over time hollow out the IP of the companies themselves. Nadella is drawing an explicit analogy to the political fallout of the last two decades — US manufacturing hollowed out to China because US companies leaned on China to make almost everything on the shelf at Costco or in an Apple store. He’s saying that same hollowing-out is in its beginning stages between the frontier-model companies and businesses — and Microsoft is now on the enterprise side of that line. It underlines Microsoft’s shift away from a handful of frontier models toward being a purveyor of a wide array of models — open or closed, large and small — assembled in service of its enterprise customers on Azure. A pragmatic move, given the frontier labs veering toward ‘what the market will bear’ a-la-carte pricing and increasingly retaining customer inference data to train future models.

Sources, in narrative order: Microsoft CEO Satya Nadella — essay “A frontier without an ecosystem is not stable”. Stratechery — ‘The Data Imperative’. Reuters — SpaceX acquires Cursor for $60 billion — fresh competition for Microsoft’s GitHub Copilot alongside Anthropic’s Claude Code and OpenAI’s Codex. For longtime readers: ‘Microsoft slipping in AI coding’ in AI-RTZ #1097, and ‘Microsoft outlines its post-OpenAI path for AI’ in AI-RTZ #898.

(3) OpenAI’s Accelerated AI-Infrastructure Spend vs Anthropic

MP TAKE: OpenAI is stepping on its AI-compute advantage to recapture the lead in AI coding with developers and enterprise customers — taking tactical advantage while Anthropic is tied up with the government. Two items stood out. One, continued rumblings of drastic price cuts vs Anthropic, leaning on its higher current spend and larger compute base to grab share. Two, the FT’s look at how much OpenAI is spending on infrastructure relative to revenue — larger than investors might like.

But in an environment where investors are leaning into AI infrastructure spend ahead of demand — and you can see that tolerance in SpaceX’s post-IPO run — OpenAI’s posture is essentially: the market seems fine with it, so throw in the kitchen sink and keep stepping on the gas, both for the business and relative to Anthropic. Highlighting the spend now is thus a marketing strategy as much as anything — and continued prep of investor expectations ahead of the next IPO steps, especially given SpaceX’s success last week. OpenAI, too, will have to figure out where it stands on the model-access lines the US government is now drawing.

Sources, in narrative order: WSJ — OpenAI considers drastic price cuts vs Anthropic et al.. FT — OpenAI spending hits $34 billion in 2025 ahead of planned IPO. For longtime readers: ‘OpenAI IPO prospects, assessed 6 months early’ in AI-RTZ #1021.

MP OVERALL TAKE

The AI Tech Wave — now in its fourth year since ChatGPT in November 2022 — is facing unusual headwinds and dynamics we haven’t seen this early in prior tech waves. Two more mega-AI IPOs are still lined up this year, assuming things get back on track.

What’s particularly striking is Microsoft recasting the frontier models as ‘China’ in the globalization wave — with manufacturing hollowing out in the US cast as the expected end result for enterprise AI customers, if they keep handing their data to the frontier-model companies in exchange for short-term AI benefits. That’s a distinctly different way of framing the enterprise AI opportunity than any of the cloud vendors have used. And the horizontal frontier models are increasingly building themselves as an application layer on top of vertical industries — finance, healthcare, education and others— extracting data through their customers, getting better, and in many cases able to replace the vertical solutions. Watch whether the other ‘Frenemies’ — Amazon AWS, the largest cloud, and Google Cloud — carry the same line forward. Expect the re-drawing of partnership-and-rivalry lines to accelerate, with unexpected twists and turns. It’s barely gotten started.

Gadget AI — The Rising Importance of ‘Markdown’ (MD) in the AI-Agent, Machine-to-Machine Phase

MP Take: In a world where machines increasingly need to talk to machines much faster than humans talk to humans, a two-decade-old file standard called Markdown (MD) — built for the blogging industry by tech guru John Gruber — has quietly become the core framework for how your prompts get translated, codified and reasoned over by the AI. Think of it as the PDF for machines — a structure machines use to talk to other machines, but one a human can still read because it looks like English. It’s also the HTML of the AI era: the way HTML was the standard for building web pages when the commercial web took off in 1994-95, MD files are the building structure for AI agents to take your instructions and act on them. Right now, if you’re touching Claude Code, Codex or Cowork, you’ll see it everywhere — and it’s worth understanding what it is and why it’s cool for both humans and computers.

Sources, in narrative order: The Vergecast — history & discussion of Markdown. Google Gemini AI Mode — on Markdown. Medium — “Why AI Prefers Markdown”. For longtime readers: ‘Not AI, But AIs’ in AI-RTZ #347.

Questions

Q1 — What is MP’s positive experience with Markdown files?

The biggest plus is that MD files are both human- and machine-readable — so you can quickly follow the ‘reasoning’ logic of what an AI is doing for you (Claude Cowork, ChatGPT, Gemini and others taking actions on your behalf) without having to be a Python expert to read the code itself. Every major model uses MD files, and I’ve found working with them noticeably easier than reading through other instruction sets. It’s reasonably accessible to people who aren’t full-time AI coders. Without it, there’d be far fewer signals to understand why the AI agents are doing what they do.

Q2 — What is the downside of the Markdown file architecture?

It’s still partly in computer format and syntax — a geeky, transitional thing. So there’s a cognitive load in parsing the machine part of the language, even though it’s far more readable than HTML or other machine-to-machine formats. My view: this is an intermediate phase. Over the next year or so it should disappear into the foundation and re-surface in more human-readable formats. If you’re a mainstream AI user wondering what all the markdown mumbo-jumbo is about — in a few months, hopefully, you won’t have to.

Full Source Reading —

For the broader context, see the canonical sources for ARD 98 — in today’s narrative order:

Event 1 — Anthropic Blip 2.0 Runs Longer

• Axios — Anthropic struggles to speak the Trump administration’s language

• Axios — Anthropic export ban sounds alarms for the AI industry

• Washington Post — How Anthropic lost the White House’s trust — and then its flagship product

• ARD #97 — Anthropic Flailing over Fear of Mythos & Fable 5

• AI-RTZ — OpenAI’s Sam Altman re-hired after ‘the Blip’ (Humpty Dumpty Still on the Wall)

Event 2 — Nadella Sides With Enterprise Customers

• Microsoft CEO Satya Nadella — “A frontier without an ecosystem is not stable”

• Stratechery — ‘The Data Imperative’

• Reuters — SpaceX acquires Cursor (Anysphere) for $60 billion

• AI-RTZ #1097 — Microsoft slipping in AI coding vs frontier models

• AI-RTZ #898 — Microsoft outlines its post-OpenAI path for AI

Event 3 — OpenAI’s Accelerated Spend

• WSJ — OpenAI considers drastic price cuts vs Anthropic

• FT — OpenAI spending hits $34 billion in 2025 ahead of planned IPO

• AI-RTZ #1021 — OpenAI IPO prospects assessed 6 months early

Gadget AI — Markdown (MD)

• The Vergecast — Markdown history & discussion

• Google Gemini AI Mode — on Markdown

• Medium — “Why AI Prefers Markdown”

• AI-RTZ #347 — Not AI, But AIs

Shorts Clips from today

Clip 1 — Markdown Files: Human & Machine Readable

Watch on YouTube Shorts

The biggest plus of Markdown (MD) files is that they’re both human- and machine-readable. That lets you follow the ‘reasoning’ logic of what an AI agent — Claude Cowork, ChatGPT, Gemini — is doing on your behalf, without needing to be a Python expert to read the code. Every major model uses MD files.

MP Take: Working with MD files is noticeably easier than reading through other instruction sets, and it’s reasonably accessible to people who aren’t full-time AI coders. Without it, there’d be far fewer signals to understand why the AI agents do what they do.

Clip 2 — AI’s Markdown: The PDF for Machines

Watch on YouTube Shorts

In a world where machines increasingly need to talk to machines faster than humans talk to humans, the two-decade-old Markdown standard — built for blogging by tech guru John Gruber — has become the core framework for how your prompts get codified and reasoned over. Think of it as the PDF for machines, and the HTML of the AI era.

MP Take: Right now, if you’re touching Claude Code, Codex or Cowork, you’ll see MD files everywhere. It’s a critical, under-discussed enabling mechanism — worth understanding what it is and why it’s cool for both humans and computers.

Clip 3 — AI Hollowing Out Business IP

Watch on YouTube Shorts

Satya Nadella’s argument: frontier models take customer data as those customers use them, make their own models better, and over time hollow out the IP of the companies themselves. He draws an explicit analogy to US manufacturing hollowed out to China over the last two decades — and says the same is in its beginning stages between the frontier-model companies and businesses.

MP Take: It’s a distinctly different way of framing the enterprise AI opportunity than any cloud vendor has used. The horizontal models are building an application layer on top of vertical industries — finance, healthcare, education — and in many cases can replace the vertical solutions.

Clip 4 — Microsoft’s AI Shift: Enterprise vs Frontier

Watch on YouTube Shorts

Microsoft is drawing a new line between itself and former close partner OpenAI, putting itself on the side of its enterprise customers — even while it retains OpenAI IP access for five more years. Nadella: “A frontier without an ecosystem is not stable.” The shift is toward a wide array of models, open or closed, assembled for enterprise missions on Azure.

MP Take: A pragmatic move, given the frontier labs veering toward ‘what the market will bear’ a-la-carte pricing and increasingly retaining customer inference data to train future models. A notable repositioning post Microsoft’s iconic OpenAI partnership.

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 #1119.

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 #98.

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 — The Anthropic Blip 2.0 Runs Longer Than OpenAI’s Blip 1.0:

• Axios — Anthropic struggles to speak the Trump administration’s language

• Axios — Anthropic export ban sounds alarms for the AI industry

• Washington Post — How Anthropic lost the White House’s trust — and then its flagship product

• ARD #97 — Anthropic Flailing over Fear of Mythos & Fable 5

• AI-RTZ — Sam Altman re-hired after ‘the Blip’ (Humpty Dumpty Still on the Wall)

Take 2 — Microsoft’s Satya Nadella Sides With Enterprise Customers:

• Microsoft CEO Satya Nadella — “A frontier without an ecosystem is not stable”

• Stratechery — ‘The Data Imperative’

• Reuters — SpaceX acquires Cursor (Anysphere) for $60 billion

• AI-RTZ #1097 — Microsoft slipping in AI coding vs frontier models

• AI-RTZ #898 — Microsoft outlines its post-OpenAI path for AI

Take 3 — OpenAI’s Accelerated AI-Infrastructure Spend vs Anthropic:

• WSJ — OpenAI considers drastic price cuts vs Anthropic

• FT — OpenAI spending hits $34 billion in 2025 ahead of planned IPO

• AI-RTZ #1021 — OpenAI IPO prospects assessed 6 months early

Gadget AI — The Rising Importance of ‘Markdown’ (MD):

• The Vergecast — Markdown history & discussion

• Google Gemini AI Mode — on Markdown

• Medium — “Why AI Prefers Markdown”

• AI-RTZ #347 — Not AI, But AIs

Q1 + Q2 — MP’s experience with Markdown files (upside + downside):

• (no external sources — MP’s own developer-beta experience)

Companion text:

• AI-RTZ #1119 — The coming personal-AI training wave

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

Today’s AI-RTZ #1119 — The coming personal-AI training wave — Apple and Google are leading the charge on training AI on our personal data, with Meta now mucking around on the public side too. An interesting set of dynamics worth being aware of — recommended as today’s reading post.

Tomorrow — ARD 99 on AI-RTZ 1120. Hopefully with more progress on the Anthropic / US-government Blip 2.0.

Thanks for joining us today, 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.)

Subscribe to AI: Reset to Zero for daily AI Ramblings + Sunday Bigger Picture posts.