Anthropic
Deconstructed

How a 3-hour AI-assisted dissection reveals the real business behind the hype — and a repeatable framework for analyzing any company.

📈 Valuation Rocket

Early 2025$60B

Sep 2025$183B

Feb 2026$380B

May 2026~$900B (≈ $6.5T)

From $60B to $900B in ~16 months — the fastest value creation in AI infrastructure.

🔍 Core Insight

Not a Chat Subscription Company

Anthropic is not primarily a Claude subscription business ($20/$100/$200 plans). It is an AI infrastructure company selling to developers, enterprises, and cloud platforms. The real revenue engines:

API (pay-per-use) Claude Code Enterprise deployments Cloud partnerships

Claude Code alone reached a $2.5B+ run rate by Feb 2026, doubling in months.

❓ The 3-Hour Framework

Three questions in order — never swap them:

Hour 1 · Who is this company?
Build a coordinate system. Don't ask for a description — ask for an analogy.

Hour 2 · Why does it survive?
Find the real moats. Ask like a skeptical investor, not a fan.

Hour 3 · How could it die?
Play the short side. Three biggest risks that could break the company.

⏳ Hour 1 · Who Is It?

            The analogy that clicked: “Anthropic in the AI era is like Intel in the PC era.” Not the same business — but the same role: building infrastructure for developers and enterprises, not just a chatbot.
        

Revenue breakdown (real story):

#1 · API

Developers & enterprises

Pay-per-use inference

#2 · Claude Code

$2.5B+ run rate

AI coding for programmers

#3 · Subscriptions

Web & app

$20 / $100 / $200 plans

The biggest counter-intuitive finding: retail users are not the majority of revenue.

⏳ Hour 2 · Why Does It Survive?

Four real moats (asked like a skeptical investor):

1. Developer lock-in

Programmers = high frequency, high willingness to pay, influence enterprise tooling. Claude is a wedge into B2B.

2. Enterprise trust

AI safety is not a slogan — it's the brand narrative. Founders left OpenAI because safety mattered. Enterprises fear data leaks, compliance risks, hallucinations. Anthropic sells: “Strong enough, but also stable.”

3. Cloud ecosystem

$100B+ commitment on AWS over 10 years (announced Apr 2026). Deep partnerships with AWS & Google — compute = survival.

4. Growth velocity

Annualized revenue run rate hit $30B by Apr 2026 — 8x faster than internal forecasts.

⏳ Hour 3 · How Could It Die?

Three risks from a short seller's perspective:

Risk #1 · Compute bottleneck

Biggest problem: not demand, but compute capacity. More users → more inference cost. The $100B AWS deal is not a luxury — it's existential.

Risk #2 · Revenue quality

How much is recurring vs. one-time? Cloud-resold revenue: gross or net? Scrutiny before a potential IPO will be intense.

Risk #3 · Competition

OpenAI, Google, open-source models — no one stops. Claude Code is strong today, but users are already shifting to Codex. No moat is permanent in AI.

🎯 Final Judgment

Short term · Compute capacity.
Mid term · Enterprise willingness to pay.
Long term · Can Claude expand from programmers to all knowledge workers, creatives, designers?

One-liner: Anthropic is an AI infrastructure company that turned safety into enterprise trust, programmers into super-users, and compute into a strategic asset. Very strong, but not risk-free. The balancing act is between compute, capital, and customer growth.

🧠 The 5-Step Method

Deconstruct any company in 3 hours

Ask for an analogy — not a description. “What is this company like?” Build a coordinate system fast.
Map the business — who pays for what? Which revenue line is biggest? Which grows fastest? Often the real answer is counter-intuitive.
Find the lifeblood — don't ask “what are your strengths?” Ask “what would be hardest for a competitor to kill?”
Play the opponent — have AI argue as a short seller. What are the three best attacks? This prevents hype intoxication.
Form your own judgment — in one paragraph. Synthesis > aggregation. Your view will evolve, but you must have a view.

            Why this matters: In the past, researching a company meant reading reports, news, interviews, and building spreadsheets. AI doesn't think for you — but it unfolds the landscape fast, pushing you from data collection to judgment faster.
        

“I'm not saying Anthropic will win. I'm saying you can go from name to your own independent judgment in hours.”

AnthropicDeconstructed

Anthropic
Deconstructed

Anthropic Deep Dive