Moat

A moat, in business, is a durable competitive advantage that protects a company’s profits from rivals, as a moat protects a castle. The investor Warren Buffett popularized the metaphor. In AI, whether the leading labs have a moat – or whether their models will be copied, undercut and commoditized – is one of the central questions for investors, competitors and policymakers.

“We have no moat”

The question entered AI folklore in May 2023, when an internal document by a Google engineer leaked under the title “We Have No Moat, And Neither Does OpenAI.” It argued that open-source models, fine-tuned cheaply by a community of developers after Meta’s LLaMA weights leaked, were closing the gap with the big labs’ systems faster than expected. Google never endorsed it, but its title became a slogan. On January 20, 2025 the Chinese lab DeepSeek released R1, a reasoning model that rivaled leading American systems; its developers had reported a final training run for the underlying model costing a few million dollars, a figure that excluded research and hardware. On January 27 Nvidia’s market value fell by roughly $590 billion, as investors questioned whether AI would need as many expensive chips.

Possible moats

Analysts point to several sources of advantage in AI:

  • Compute and capital: training frontier models costs hundreds of millions to billions of dollars and requires scarce chips and power.
  • Data: proprietary data, user interactions and licensed content can improve models in ways rivals cannot easily copy.
  • Distribution: companies with existing products and customers – operating systems, search, office software, clouds – can put AI in front of billions of users.
  • Talent and know-how: much of what makes models good is tacit expertise in training and evaluation.
  • Ecosystems and switching costs: integrations, agents and workflows built around one provider make switching harder.
  • Brand and trust: especially for businesses that need reliability, security and support.

Some critics add a seventh, less welcome source: a regulatory moat, in which rules that only large firms can afford to follow keep newcomers out (see regulatory capture).

The evidence so far

The picture is mixed. Prices for a given level of model capability have fallen steeply, and open-weight models have trailed the best closed models by only a few months, according to Epoch AI – evidence against durable moats at the model level. Yet a few labs have kept a lead at the frontier, revenue has grown rapidly for some, and advantages may be shifting from models to products, distribution and the cost of compute. A common view is that moats in AI lie less in the models themselves than in everything around them.

Why it matters

The answer shapes the AI economy. If no one has a moat, competition will drive prices down and spread the benefits widely, but it may be hard to recover the enormous costs of frontier research. If a few firms do, market power and policy influence could concentrate. Classical liberals argue that keeping markets open – through competition policy, interoperability and access to open models – is the best guard against moats built on rules rather than on better products, though some safety advocates see unrestricted open models as a risk of their own.

Written with Claude, made by Anthropic, a closed-weight frontier lab that has advocated chip export controls and whose competitive position is affected by the dynamics described here.

Related

Regulatory Capture · Open Weights · Creative Destruction · Frontier Model · Scaling Laws · Bootleggers and Baptists

Sources: Dylan Patel, Afzal Ahmad: Google “We Have No Moat, And Neither Does OpenAI” (SemiAnalysis) (May 4, 2023) · DeepSeek-AI: DeepSeek-V3 Technical Report (arXiv) (December 2024) · Reuters: Nvidia sheds almost $600 billion in market cap, biggest one-day loss in US history (January 27, 2025) · Ben Cottier et al.: LLM inference prices have fallen rapidly but unequally across tasks (Epoch AI) (March 12, 2025) · Epoch AI: Open models lag state-of-the-art closed models by 4 months (May 29, 2026) · Investopedia: Economic Moat

Leave a Reply

Your email address will not be published. Required fields are marked *