Mythos, Muse, and the Opportunity…

Ben Thompson | April 13, 2026

1. Key Themes

Theme 1: Opportunity Cost — Not Marginal Cost — Is the New Strategic Constraint in AI

The conventional critique of AI economics focuses on rising marginal costs breaking the internet-era model. Thompson argues the real constraint is opportunity cost: compute allocated to one use case cannot serve another, forcing explicit prioritization decisions.

"The cost that Microsoft is contending with here is not marginal cost, but rather opportunity cost: compute spent in one area cannot be used in another area... Microsoft was admitting that they could have made their Azure number if they wanted to, but chose to prioritize their own workloads because those have higher gross margin profiles and higher lifetime value."

"Anthropic isn't facing a marginal cost problem, but an opportunity cost problem: where to allocate its compute."

Theme 2: Aggregation Theory Survives — Demand Control Still Beats Supply Control

Despite compute constraints, Thompson argues the fundamental principle holds: whoever controls end-user demand will ultimately be able to command supply, not the other way around.

"My bet is that owning demand will ultimately trump owning supply, suggesting that the underlying principles of Aggregation Theory lives on... I think that OpenAI will need to win with better products, not just more compute."

"When it comes to AI, distribution and transaction costs are still free — the two preconditions for Aggregators — which means that the winners should be those with the most compelling products. Those products will win the most users, providing the money necessary to source the compute to serve them."

Theme 3: Agentic AI Is Exponentially Accelerating Compute Demand and Enterprise Revenue

The shift from conversational AI to agentic AI is the primary driver of both compute scarcity and the pivot toward enterprise monetization. Agents run LLMs continuously without human interruption, multiplying token consumption dramatically.

"Agents have exponentially increased token demand, as they can use LLMs continuously without a human in the loop. This is a huge driver in sky-rocketing demand for Claude, as well as OpenAI's Codex. Moreover, this use case is so potentially profitable that not only is Anthropic's revenue sky-rocketing, but OpenAI is pivoting its focus to enterprise."

Theme 4: Meta Has a Structural Advantage in the Consumer AI Market

Because Meta has no enterprise cloud business, it faces zero opportunity cost in serving consumers — and already has a monetization engine (advertising) to fund it. This structurally positions Meta as the most natural winner of consumer AI.

"Unlike any of the hyperscalers or the frontier labs, Meta does not have an enterprise or cloud business to worry about. That means that serving the consumer market comes with no opportunity costs. Of course those opportunity costs would be much smaller anyways, given that Meta already has an at-scale advertising business to monetize usage."

"Meta may actually face less competition in winning the consumer space than it might have seemed a few months ago, simply because that is their primary focus."

Theme 5: Distillation by Chinese AI Labs Is an Existential Threat to Frontier Lab Pricing Power

Industrial-scale model distillation by Chinese competitors directly erodes the moats of Western frontier labs — threatening both margins and compute leverage.

"We have identified industrial-scale campaigns by three AI laboratories — DeepSeek, Moonshot, and MiniMax — to illicitly extract Claude's capabilities to improve their own models. These labs generated over 16 million exchanges with Claude through approximately 24,000 fraudulent accounts... Distillation can also be used for illicit purposes: competitors can use it to acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently." (Anthropic blog, via Thompson)

2. Contrarian Perspectives

Contrarian 1: OpenAI's Massive Consumer Base May Be a Liability, Not an Asset

The conventional view is that ChatGPT's scale is OpenAI's crown jewel. Thompson flips this: in a compute-constrained world, a large consumer base is an opportunity cost burden that diverts resources from higher-margin enterprise workloads.

"You can make the argument that one of OpenAI's biggest challenges is the fact it has such a popular consumer product in ChatGPT... until [advertising] materializes, that big consumer base is a big opportunity cost in terms of OpenAI's focus and compute. The temptation to allocate more and more compute to agentic use cases that enterprises will pay for, even at the expense of the consumer business, will be very large."

Supporting evidence: OpenAI is already pivoting toward enterprise, and has communicated to investors that its compute build-out is a key differentiator against Anthropic — suggesting internal prioritization is shifting away from the consumer base.

Contrarian 2: Withholding Frontier Models Is a Smarter Competitive Move Than Releasing Them

The default assumption is that AI labs should release powerful models broadly to maximize adoption. Thompson argues strategic withholding of models like Mythos is rational for multiple reinforcing reasons — compute allocation, pricing power preservation, and blocking distillation by rivals.

"There are other reasons for Anthropic to not make Mythos widely available... The key to handling those costs will be to charge more for Claude going forward; that, by extension, means maintaining pricing power, which leads to a second benefit of not releasing Mythos broadly."

"To the extent they can make compute less useful for their potential customers — by stopping open source models from distilling their models — is the extent to which they can acquire that compute for themselves at more favorable rates."

Contrarian 3: Meta Should Open-Source Muse — and Doing So Would Hurt Frontier Labs More Than Meta

The received wisdom is that open-sourcing a frontier model is self-defeating. Thompson argues the opposite: open-sourcing Muse would disproportionately damage Anthropic and OpenAI (by compressing their pricing power and compute leverage) while costing Meta very little, since Meta's value accrues through advertising and consumer distribution, not model licensing.

"This, by the same token, is why Meta should open source Muse, just like they did Llama. The entities that will be the most hurt by widespread availability of a frontier model are other frontier labs, who will see their pricing power reduced and face increased competition for compute. This will make it even harder for them to bear the opportunity cost of pursuing the consumer market, leaving it for Meta."

3. Companies Identified

Microsoft

Description: Enterprise software and cloud giant; Azure is its primary cloud product
Why mentioned: Case study in opportunity cost tradeoffs — deliberately constrained Azure GPU allocation to prioritize first-party products (M365 Copilot, GitHub Copilot), which carry higher gross margins
Quote: "If I had taken the GPUs that just came online in Q1 and Q2 in terms of GPUs and allocated them all to Azure, the KPI would have been over 40." (Amy Hood, Microsoft CFO)

Anthropic

Description: Frontier AI lab; maker of the Claude family of models
Why mentioned: Central case study — releasing Mythos only to select high-paying customers, grappling with compute shortfalls, facing distillation attacks, and pursuing a TPU supply deal with Google as a strategic response
Quote: "Anthropic isn't facing a marginal cost problem, but an opportunity cost problem: where to allocate its compute."

Meta / Meta Superintelligence Labs

Description: Social media and advertising giant; newly launched frontier AI lab
Why mentioned: Positioned as the structurally advantaged player in consumer AI due to its advertising monetization model and absence of enterprise/cloud business creating no opportunity cost tension
Quote: "Muse Spark is the first in the Muse family of models developed by Meta Superintelligence Labs... the first product of a ground-up overhaul of our AI efforts." (Meta blog)

OpenAI

Description: Frontier AI lab; maker of ChatGPT and Codex
Why mentioned: Pivoting toward enterprise agentic use cases; argues its compute infrastructure lead over Anthropic is its key competitive moat, though Thompson questions whether supply advantage beats product/demand advantage
Quote: "OpenAI told investors this week that its early push to dramatically increase computing resources gives it a key advantage over Anthropic PBC at a moment when its longtime rival is gaining ground." (Bloomberg)

DeepSeek, Moonshot, MiniMax

Description: Chinese AI laboratories
Why mentioned: Identified by Anthropic as conducting industrial-scale distillation attacks against Claude, generating 16 million exchanges through ~24,000 fraudulent accounts
Quote: "We have identified industrial-scale campaigns by three AI laboratories — DeepSeek, Moonshot, and MiniMax — to illicitly extract Claude's capabilities to improve their own models." (Anthropic blog)

Google

Description: Search, cloud, and AI company; strategic investor in Anthropic; provider of TPU infrastructure
Why mentioned: Referenced as a parallel to Meta (advertising-based, owns consumer distribution), as Anthropic's key compute partner, and as a supplier that Anthropic is effectively drawing supply away from
Quote: "Anthropic's deal to secure a meaningful portion of TPU supply, which, given the capacity constraints at TSMC, is ultimately an example of taking supply from Google."

Amazon

Description: E-commerce and cloud giant (AWS); strategic investor in both Anthropic and OpenAI
Why mentioned: Example of a hyperscaler managing complex opportunity cost tradeoffs across e-commerce, AWS, and frontier AI investments simultaneously
Quote: "Amazon needs to balance its e-commerce business, AWS, and its strategic investments in both Anthropic and OpenAI."

4. People Identified

Ben Thompson

Description: Author of Stratechery; creator of Aggregation Theory
Why mentioned: Primary author; defends Aggregation Theory against the critique that AI's compute costs have made it obsolete
Quote: "My bet is that owning demand will ultimately trump owning supply, suggesting that the underlying principles of Aggregation Theory lives on."

Doug O'Laughlin

Description: Author of Fabricated Knowledge newsletter; semiconductor and AI analyst
Why mentioned: His January 2025 thesis — that reasoning models and their marginal costs signal the death of Aggregation Theory — is the central argument Thompson engages and refutes throughout the article
Quote: "The era of Aggregation Theory is behind us, and AI is again making technology expensive... One of our fundamental assumptions about this period is unraveling." (O'Laughlin)

Amy Hood

Description: CFO of Microsoft
Why mentioned: Provided the key on-record confirmation that Microsoft is deliberately making opportunity cost tradeoffs with GPU allocation, choosing first-party products over Azure revenue
Quote: "The first thing we're doing is solving for the increased usage in sales and the accelerating pace of M365 Copilot as well as GitHub Copilot, our first-party apps."

Satya Nadella

Description: CEO of Microsoft
Why mentioned: Confirmed on the earnings call that first-party workloads carry higher gross margin profiles and higher lifetime value — the rationale for prioritizing them over Azure
Quote: "Those have higher gross margin profiles and higher lifetime value." (Thompson paraphrasing Nadella)

Mark Zuckerberg

Description: CEO of Meta
Why mentioned: Credited with making the correct strategic call to undertake a "ground-up overhaul" of Meta's AI efforts, validated by the subsequent explosion in agentic AI demand
Quote: "The last nine months of AI have made clear that CEO Mark Zuckerberg made the right call to embark on that 'ground-up overhaul of [Meta's] AI efforts'."

5. Operating Insights

Insight 1: Price for Compute Scarcity — Move Customers to Usage-Based Models Now

Subscription plans that don't charge per usage create disproportionate compute burden during scarcity. Anthropic's Mythos withholding is partly driven by the fact that flat-rate subscribers would consume Mythos compute without commensurate revenue. For any AI-native product company, the operational implication is clear: migrate toward consumption-based pricing before compute constraints force painful quality degradations.

"Making Mythos more widely available — particularly to subscription plans that don't pay per usage — would make the situation much worse."

Insight 2: Model Access Strategy Should Be Treated as a Competitive Moat Decision, Not Just a Safety Decision

Whether to release, restrict, or open-source a model is not merely a safety or PR question — it is a capital allocation and competitive strategy decision with direct implications for pricing power, compute access, and competitive positioning against both open-source and proprietary rivals.

"The key to handling those costs will be to charge more for Claude going forward; that, by extension, means maintaining pricing power, which leads to a second benefit of not releasing Mythos broadly... To the extent they can make compute less useful for their potential customers — by stopping open source models from distilling their models — is the extent to which they can acquire that compute for themselves at more favorable rates."

Insight 3: Enterprise Agentic Use Cases Should Be Top Priority for Revenue Allocation

For AI companies or software companies building on AI, agentic workloads that run continuously without human oversight are the highest-value use case to prioritize — they drive token consumption, enterprise willingness to pay, and sustained revenue that can fund further compute acquisition.

"This use case is so potentially profitable that not only is Anthropic's revenue sky-rocketing, but OpenAI is pivoting its focus to enterprise."

6. Overlooked Insights

Insight 1: TSMC Capacity Constraints Create a Zero-Sum Compute Market Among Hyperscalers

Thompson makes a passing but significant point: Anthropic securing TPU supply from Google is not merely a bilateral deal — it is physically drawing compute away from Google's own workloads given TSMC's production constraints. This implies that aggressive compute deals by any frontier lab are directly competitive moves against the hyperscalers hosting them — a dynamic with major implications for the long-term relationship between labs and their cloud backers.

"Anthropic's deal to secure a meaningful portion of TPU supply, which, given the capacity constraints at TSMC, is ultimately an example of taking supply from Google."

Insight 2: AI's Security Threat Is a "Wolf That Will Come" — Regardless of Which Model Gets There First

Thompson briefly but importantly separates the question of whether Mythos specifically is a security threat from the structural certainty that some AI model will be. This reframes cybersecurity investment not as a response to a specific model release but as a permanent, escalating infrastructure requirement.

"It's actually not important whether or not Mythos represents a major security threat: if this model doesn't, a future model will; to that end, I do support leveraging Mythos to proactively find and fix bugs before bad actors can find and exploit them."