'Then There Were Two': Anthropic's 'The Blip 2.0' & OpenAI's Catch-up Race. ARD #97

The SpaceX IPO is done — SPCX priced and traded up about 20% on Friday, a little over $2.1 trillion. And that leaves two mega-AI IPOs still scheduled this year. So attention shifts to both. Today’s theme: ‘Then There Were Two’ — with increasingly more questions than answers, and more headwinds than tailwinds. Three events, each with my Take first — and my Overall Take.

(1) Anthropic — Flailing Over Fear of Mythos and Fable 5

MP TAKE: I’ve been calling this the ‘Blip 2.0’, Anthropic being ‘fired’ by the US Government vs OpenAI founder/CEO Sam Altman being fired and re-hired over a weekend three years ago, in what is known as ‘The Blip’ — This Blip 2.0, is the second big global blow to the AI Tech Wave, and potentially bigger in its negative impact than Blip 1.0. That first Blip while stopped the momentum of OpenAI, was less threatening to the whole industry— and it got resolved in days. This one is a firing of Anthropic’s latest and best global models, Mythos and the just-released Fable 5 — by the US Government, on Friday night. It explicitly carves out restrictions for non-US parties, echoes of the Administration’s Stephen Miller led anti-immigration campaigns. Fundamentally threatens the plethora of non-US AI Researchers working across the US at AI companies large and small. At great cost and threats to the opportunities of their work. It came on warnings from Anthropic’s earliest equity ($13+ billion) and distribution partner, Amazon. Whose CEO Andy Jassy himself apparently flagged the jailbreak risks to the Treasury Secretary — and it landed just days after founder/CEO Dario Amodei’s own loud AI-safety ‘Wolf Crying’ of AI Risks, and the need for active government oversight. The script writes itself.

It’s all dripping with personality-driven soap opera, feuds and griefs on all sides — especially a US government’s defense department waging a kind of ‘holy, anti-woke war’ with Anthropic. And it runs against the long-term interests of the US and its tech industry: it buys time for China, and it scares long-standing global partners and customers in Canada, Europe and elsewhere away from relying on the US for the latest tech — the way they’ve relied on the US for military defense for decades. That reliance has been a core foundation of US technology leadership and prosperity since the transistor in 1947. Even if this gets resolved and reversed in days, as it likely will, the implications last far longer — because it changes how the world thinks about the trade being asked of it: trust us with your data, your information, the trillions you’ll spend rebuilding your businesses on US models. It’s also a boost for open source AI models and applications, particularly from China. In the meantime it means a whole new set of ‘Risk Factors’ in Anthropic’s and OpenAI’s confidential S-1, and a whole new set of investor questions ahead of its anticipated mega-AI IPOs this year.

Sources, in narrative order: Anthropic — statement on the US directive to suspend access to Fable 5 and Mythos 5. Axios — how Amazon and the US government shut down Anthropic’s Mythos and Fable. Bloomberg — Anthropic says US limits foreign access to Fable 5 / Mythos 5. Axios — “They screwed up” — personality clashes sent Anthropic’s model offline. Axios — Anthropic flies staff to DC to clean up the White House fight. Axios — cyber leaders defend Anthropic’s banned model. Bloomberg — Carney says the Anthropic ban shows the risk of relying on Big AI models. The Information — Amazon’s Jassy raised concerns about the Anthropic model before the Trump crackdown. For longtime readers: ‘Anthropic Flagged to a Screeching Halt by the US Government — Blip 2.0’ in AI-RTZ #1117.

(2) OpenAI — Bigger Spend and Distractions, in a Race to Catch Up

MP TAKE: On the surface, OpenAI is the obvious beneficiary of Anthropic’s troubles — the other of the two that remain. It has its own latest model being vetted for cybersecurity, and Sam Altman is getting more aggressive on price because he has more compute in place, eyeing market share while Anthropic raises its a-la-carte pricing. Given the dynamics of an IPO and relative growth rates, that’s likely the right tactical move — especially with enterprise customers pushing back on rising a-la-carte costs and tokenmaxxing without efficiency and cost controls.

But OpenAI is not exempt from the bigger question the ‘Blip 2.0’ raises — whether the US is a reliable tech partner in AI. It’s not that the world has many choices, but it does have some, particularly in open source, where roughly 50-60% of the world’s inference tokens doing useful work already run on open models — even as 50-60% of the economic value still accrues to the closed-model companies like OpenAI and Anthropic. All of which raises a fresh host of investor questions and new ‘Risk Factors’ in both companies’ mega-AI IPO S-1 filings.

Sources, in narrative order: Reuters — Anthropic v. OpenAI: behind the bitter battle for the future of AI. The Information — the math behind Anthropic’s mad revenue growth. WSJ — the AI price war is here, piling pressure on OpenAI and Anthropic. For longtime readers, in narrative order: ‘It’s Business and Personal’ vs Anthropic in AI-RTZ #1048; ‘Meta steps back from AI tokenmaxxing’ in AI-RTZ #1118; and ‘Customers asking how much for AI tokenmaxxing’ in AI-RTZ #1103.

(3) Both Facing Public Behemoths Worth Trillions — Now Including SpaceX/Tesla and Elon Inc.

MP TAKE: Both OpenAI and Anthropic now face not just newly-IPO’d competition from Elon Musk’s $2+ trillion SpaceX (SPCX) — but well-established Mag 7s raising hundreds of billions in debt and equity to compete ‘Frenemies’-style with the core AI companies themselves. It’s not only the IPOs raising money — those three collectively over $200 billion this year, with OpenAI and Anthropic each likely to raise $60-billion-plus if they get out. Google has already raised $80 billion; Nvidia just priced a $20 billion bond; Amazon raised $18 billion; Meta has signaled more to come. The big public balance sheets are accelerating their own raises, countering the IPO supply.

This is a new dynamic for investors to ingest about the two mega-AI IPOs that remain — especially in an environment where the core AI companies can’t always rely on their own equity partners to watch their back. The case in point, of course, is Amazon warning the US government on Anthropic’s Mythos/Fable risks. The competition among the ‘Frenemies’ partners is far sharper than it looks. Note Microsoft CEO Satya Nadella’s warning to its enterprise customers about the very frontier-model companies — Anthropic and OpenAI — it partners with closely and invests in heavily.

Sources, in narrative order: Microsoft CEO Satya Nadella — “A frontier without an ecosystem is not stable”. NYTimes — how Google is starting to win in the AI race. Bloomberg — Nvidia joins the AI debt boom with a $20 billion bond offering. Axios — Apple could win the AI race without running. WSJ — SpaceX, now worth $2.1 trillion, pulls off a Goldilocks debut. The Verge — the world’s first trillionaire and the SpaceX IPO is great for Elon Musk. For longtime readers: ‘The Four Corners & Key Pieces of the 2026 AI Chess Board’ in ARD #93, and the ‘Frenemies’ nature of the AI industry in AI-RTZ #554.

MP OVERALL TAKE

The overall competitive environment and the global government-governance realities are getting sharper and tougher for all the AI companies — especially the two front-and-center leaders lining up next to go public. So far investors are leaning into them all, despite the concerns. But one needs to watch for ebbs and flows in that support, while the underlying AI-model innovation and building over and across the AI Tech Stack continues at unprecedented pace and scale.

All the Blip 2.0 type of drama continues to potentially erode global customer confidence in AI being safe and worth it — especially at the prices increasingly being asked. In the meantime, the competition from local AI models with no metering of inference keeps building — from major players like Apple, Nvidia, AMD and others (see Gadget AI below). Not an immediate headwind for cloud-based frontier models, but it starts to weigh more heavily in 2030-type projections — even against a supply-constrained environment for memory and other components for local AI at scale, in both computers and smartphones. A very unusual time indeed for investors ahead, with no easy parallels in prior tech waves.

Gadget AI — More Local AI Gadgets Scaling Up for Local Inference

Sources, in narrative order: Startupfortune — AMD’s tiny AI PC points to a more local future for model inference — joining Apple and Nvidia/Microsoft, now with unified-memory chips at 128GB and above, competing with Nvidia’s Spark PC and Apple’s upgraded Mac Mini line. For longtime readers, in narrative order: Nvidia and Microsoft local AI PCs in AI-RTZ #1104, and ‘New Focus on Local AI Devices’ in AI-RTZ #387.

MP Take: An important dynamic to watch between now and 2030 versus ‘mainframe’ AI training and inference for compute tokens. Local — with its meter-free promise and implicit privacy — offers big advantages for consumers and businesses ahead. Especially in a multi-model world to come, where a $20-30K local box of compute sitting next to the display, with no meter, becomes an increasingly attractive option for trillion-parameter models.

Questions

Q1 — What is MP most excited about with Siri AI in the Fall?

The ability for Siri to leverage my local information across Apple — and eventually third-party apps — to provide AI use cases not possible with Anthropic, OpenAI and other frontier models today. Even in developer beta number one (not for regular people — I run it on backup devices), Siri AI is already a lot more useful than people expected, because it indexes, in my words trains on, local data across Apple Mail, Notes, Photos, Messages and more. And then being able to build deeply relevant inference narratives that are unlike any other AI model and agents, large or small. That’s what’s startling developers testing the early developer builds. Expect final releases this Fall, just a few months away.

Q2 — What is the biggest near-term hurdle?

Two of them. One — the time to index (what I call ‘train), local user data privately. In early beta it’s measured in days most of the time; hopefully shorter soon, with more personal cloud compute by the Fall release. Two — the current lack of interoperability across third-party apps. Apple Siri currently indexes Apple Mail but not Gmail, Apple Photos, not Google Photos, etc. And there’s no MCP third party app and data interoperability support yet. A lot of issues still to resolve — but once it’s done, the utility is greater than the cost of the wait. And Apple is in a relatively unique position to do this vs peers. Even Google.

Full Source Reading —

For the broader context, see the canonical sources for ARD 97 — in today’s narrative order:

Event 1 — Anthropic: Mythos and Fable

• Anthropic — Statement on the US directive to suspend access to Fable 5 / Mythos 5

• Axios — How Amazon and the US government shut down Anthropic’s Mythos and Fable

• Bloomberg — Anthropic says US limits foreign access to Fable 5 / Mythos 5

• Axios — “They Screwed Up” — personality clashes sent Anthropic’s model offline

• Axios — Anthropic flies staff to DC to clean up the White House fight

• Axios — Cyber leaders defend Anthropic’s banned model

• Bloomberg — Carney says the Anthropic ban shows the risk of relying on Big AI models

• The Information — Amazon’s Jassy raised concerns about the Anthropic model before the Trump crackdown

• AI-RTZ #1117 — Anthropic Flagged to a Screeching Halt by the US Government, Blip 2.0

Event 2 — OpenAI: Race to Catch Up

• Reuters — Anthropic v. OpenAI: behind the bitter battle for the future of AI

• The Information — The math behind Anthropic’s mad revenue growth

• WSJ — The AI price war is here, piling pressure on OpenAI and Anthropic

• AI-RTZ #1048 — It’s Business and Personal vs Anthropic

• AI-RTZ #1103 — Customers asking “how much” for AI tokenmaxxing

Event 3 — Public Behemoths Worth Trillions

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

• NYTimes — How Google is starting to win in the AI race

• Bloomberg — Nvidia joins the AI debt boom with a $20 billion bond offering

• Axios — Apple could win the AI race without running

• WSJ — SpaceX, now worth $2.1 trillion, pulls off a Goldilocks debut

• The Verge — The world’s first trillionaire

• The Verge — The SpaceX IPO is great for Elon Musk

• ARD #93 — The Four Corners & Key Pieces of the 2026 AI Chess Board

• AI-RTZ #554 — Frenemies nature of the AI industry

Gadget AI — Local AI Gadgets

• Startupfortune — AMD’s tiny AI PC points to a more local future for model inference

• AI-RTZ #1104 — Nvidia and Microsoft local AI PCs

• AI-RTZ #387 — New Focus on Local AI Devices

Shorts Clips from today

Clip 1 — Siri AI: More Useful Than Expected

Watch on YouTube Shorts

Even in developer beta number one, Siri AI is a lot more useful than people expected — maybe two years late, but it does real work. The core reason: it indexes, in MP’s words trains on, local data across Apple Mail, Notes and Messages, so the model knows a lot more about you and gets far more relevant to your queries.

MP Take: What MP is most excited about is Siri leveraging local information across Apple, and eventually third-party apps, to do things the frontier models from Anthropic and OpenAI simply can’t do today. Local plus personal is a different kind of useful.

Clip 2 — AI Training: Why Local Still Takes Time

Watch on YouTube Shorts

Training has always been the slowest part of large language models. Training on your local data is no different in kind: with ample cloud compute it can take hours, but in MP’s own developer beta the local indexing has run into days.

MP Take: This is the biggest near-term hurdle for Siri AI — that, and the current lack of third-party interoperability. Siri indexes Apple Mail but not Gmail, and there’s no MCP support yet. By the Fall release this should accelerate with more personal cloud compute. The utility, once done, is greater than the cost of the wait.

Clip 3 — Blip 2.0: Global Tech Reliance at Risk

Watch on YouTube Shorts

Over the weekend the US government told Anthropic to pull Mythos and Fable 5 from the market for non-US parties — days after founder/CEO Dario Amodei’s loud AI-safety warnings, and after earliest partner Amazon flagged the risks. MP calls it Blip 2.0, echoing the 2023 Altman firing that briefly froze the whole industry.

MP Take: The deeper damage is to trust. Partners in Canada and Europe are already saying they may need to rethink relying on the US for the latest AI — the way they’ve relied on it for defense for decades. Even if resolved in days, that question lingers, and buys time for China.

Clip 4 — The Blip 2.0 AI Trust Crisis

Watch on YouTube Shorts

The Anthropic episode is dripping with personality-driven soap opera on all sides — a US government feud, Amazon flagging its own partner’s models, a team flown to DC to clean it up. All of it adds a fresh set of Risk Factors to Anthropic’s confidential S-1, and a new set of investor questions ahead of a mega-AI IPO this year.

MP Take: This is what’s very different this time versus prior tech waves — how much of it is emotional and personality-driven rather than technical. All the Blip 2.0 drama erodes global customer confidence in AI being safe and worth it, especially at the prices increasingly being asked.

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 #1118.

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 #97.

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: Flailing Over Fear of Mythos and Fable:

• Anthropic — Statement on US directive to suspend Fable 5 / Mythos 5

• Axios — How Amazon and the US government shut down Anthropic’s Mythos and Fable

• Bloomberg — Anthropic says US limits foreign access to Fable 5 / Mythos 5

• Axios — “They screwed up”: personality clashes sent Anthropic’s model offline

• Axios — Anthropic flies staff to DC to clean up the White House fight

• Axios — Cyber leaders defend Anthropic’s banned model

• Bloomberg — Carney says the Anthropic ban shows the risk of relying on Big AI models

• The Information — Amazon’s Jassy raised concerns about the Anthropic model before the Trump crackdown

• AI-RTZ #1117 — Anthropic Flagged to a Screeching Halt by the US Government, Blip 2.0

• AI Tech Wave / Building Value Over Time evergreen

Take 2 — OpenAI: Bigger Spend & Distractions vs Anthropic:

• Reuters — Anthropic v. OpenAI: behind the bitter battle for the future of AI

• The Information — The math behind Anthropic’s mad revenue growth

• WSJ — The AI price war is here, piling pressure on OpenAI and Anthropic

• AI-RTZ #1048 — It’s Business and Personal vs Anthropic

• AI-RTZ #1118 — Meta steps back from AI tokenmaxxing

• AI-RTZ #1103 — Customers asking “how much” for AI tokenmaxxing

Take 3 — Both Facing Public Behemoths Worth Trillions:

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

• NYTimes — How Google is starting to win in the AI race

• Bloomberg — Nvidia joins the AI debt boom with a $20 billion bond offering

• Axios — Apple could win the AI race without running

• WSJ — SpaceX, now worth $2.1 trillion, pulls off a Goldilocks debut

• The Verge — The world’s first trillionaire

• The Verge — The SpaceX IPO is great for Elon Musk

• ARD #93 — The Four Corners & Key Pieces of the 2026 AI Chess Board

• AI-RTZ #554 — Frenemies nature of the AI industry

Gadget AI — More Local AI Gadgets Scaling Up for Local Inference:

• Startupfortune — AMD’s tiny AI PC points to a more local future for model inference

• AI-RTZ #1104 — Nvidia and Microsoft local AI PCs

• AI-RTZ #387 — New Focus on Local AI Devices

Overall Take — AI Tech Stack:

• AI Tech Stack / Building Value Over Time evergreen

Companion text:

• AI-RTZ #1118 — Meta steps back from AI tokenmaxxing

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

Today’s AI-RTZ #1118 — Meta steps back from AI tokenmaxxing — Meta is pulling back from the AI tokenmaxxing leaderboard race it kicked off a few months ago, now curbing the spend through internal cost-budget dashboards. A more responsible turn worth reading in full.

Tomorrow — ARD 98 on AI-RTZ 1119. Hopefully with some resolution on the Anthropic situation with the government.

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.