Anthropic Has No Moat news update news update

Anthropic filed an amended S-1 on November 11, 2026, pricing its initial public offering at $70 to $78 a share across 825 million shares, targeting a Nasdaq listing under the ticker ANT as early as the week of November 16. trevorlasn.com's report lays out the bear case: at that price range, The Motley Fool estimates the company's total value at roughly $1.27 trillion to $1.41 trillion.

This piece is for developers and investors trying to understand what, if anything, keeps Anthropic's customers from switching to a cheaper rival once performance is close enough. It covers the company's revenue trajectory, the five classic types of competitive moat and how Anthropic measures against each, the speed at which customers have already begun to move, and the narrowing gap between frontier and open-weight models.

01Revenue growth masks a structurally weak lock-in

Anthropic's business is real and growing at an extraordinary pace. Digital Applied's breakdown of the S-1 puts revenue at $386 million in 2024, $4.6 billion in 2025, and $33.2 billion across the first nine months of 2026 — thirteen times the same period a year earlier. No single customer exceeded 10% of revenue in those nine months, and the top five together accounted for roughly 25%, a sign the base has broadened.

The composition of that revenue, however, works against lock-in. Seventy-nine percent of the nine-month figure was pay-as-you-go, billed by the token the way electricity is billed by the kilowatt-hour.

There is no contract to wait out, no migration project to budget, and no proprietary data stored on Anthropic's side. Morningstar, which values Anthropic at $1.7 trillion, declines to count switching costs as a competitive moat for exactly that reason.

02Five classic moats, none of them Anthropic's

Investors typically credit Warren Buffett with popularising the idea of a moat — the reason a customer stays when a rival offers the same thing for less. Five kinds are usually counted: high switching costs (Apple's iPhone, where photos, messages and paid apps are tied to the ecosystem), monopoly supply (ASML, the only maker of EUV chip-printing machines at over $350 million apiece), network effects (Visa, where shops accept it because shoppers carry it), lowest-cost production (Costco), and prohibitive replication cost (a railroad).

Anthropic's S-1 describes its strategy as one that "prioritises model capability as a principal source of differentiation," which the author compares to a gas station: you drive past a cheaper one because the fuel is better today, but nothing about your car stops you filling up elsewhere tomorrow. A company using Claude sends it text and receives text back, paying per token, with its own software, instructions, and data sitting on its own computers — nothing material transfers to Anthropic's side.

The one partial moat is replication cost. Some $895 billion in chip contracts keeps small players out, but that barrier does not apply to Google, OpenAI, Meta, or SpaceXAI.

03Claude Code itself can run on rival models

Anthropic's own documentation demonstrates how small a migration is in practice. Its published example shows Claude being called through the OpenAI software kit by changing two lines: the base URL to https://api.anthropic.com/v1/ and the model name to claude-opus-5-5.

Reverse those two values and the same program routes to a different company entirely. Anthropic describes this compatibility layer as "primarily intended to test and compare model capabilities."

Claude Code, the coding assistant the author uses daily, is not tied to Claude either. Three rival labs — Z.ai, Moonshot, and DeepSeek — have cloned Anthropic's server protocol.

A single settings file in ~/.claude/settings.json redirects Claude Code to Z.ai's API, where a Chinese model called GLM handles inference. MCP, the standard that lets a model use external tools, was created by Anthropic and transferred to the Linux Foundation's Agentic AI Foundation in December 2025, so the entire tool ecosystem built around Claude is now portable.

04Large customers have already begun switching

The migration from Claude is not theoretical. At AI Engineer NYC on October 13, David Abitbol of the fintech Revolut said the company now runs GLM 5.3 itself on 56 Nvidia B200 chips as the default model for roughly 2,000 engineers, at a blended cost near $0.30 per million tokens. Revolut still uses Claude Code and OpenAI's Codex for problems GLM cannot handle, but the default has moved.

SpaceX closed its $60 billion acquisition of Cursor on August 14. The S-1 notes two customers each represented 12% of revenue in 2025; VentureBeat's reporting, via PYMNTS, identified those customers as Cursor and GitHub Copilot, together accounting for about $1.2 billion of Anthropic's $5 billion in 2025 revenue. Meta separately cut its Claude Code user base from roughly 60,000 to 30,000, moving most to internal tools.

SpaceX closed its $60 billion acquisition of Cursor on August 14. The S-1 notes two customers each represented 12% of revenue in 2025; VentureBeat's reporting, via PYMNTS, identified those customers as Cursor and GitHub Copilot, together accounting for about $1.2 billion of Anthropic's $5 billion in 2025 revenue. Meta separately cut its Claude Code user base from roughly 60,000 to 30,000, moving most to internal tools.

05Open-weight models closing a four-month gap

Anthropic's frontier models are currently at the top of the benchmarks. On Vals AI's Terminal-Bench 4.0, a coding benchmark, Claude models held the top three positions as of November 9.

The concern is the trajectory. Open-weight models — whose files anyone can download and run on their own hardware — sell at prices close to bare chip cost and trail frontier models by roughly four months on coding benchmarks, according to Epoch AI's measurements from January through the period covered.

Rival models score about two points behind Claude on coding benchmarks, and the gap has been shrinking from both sides: open-weight models improve, and frontier prices fall. Nothing in Anthropic's cost structure insulates it, because it rents its chips from the same cloud providers that sell competing models. A customer choosing between Claude and a self-hosted open-weight model at $0.30 per million tokens is weighing two points of benchmark performance against a substantial and recurring cost difference — and several large engineering organisations have already decided that trade-off.