Morningstar began coverage of artificial intelligence labs ahead of Anthropic’s expected IPO, asking whether Anthropic and OpenAI can build durable competitive advantages and earn profits. The investment research firm’s first installment in a three-part series argues that leading labs are developing economic moats, though those advantages are not fully formed. Its concern follows years of questions about whether providers can profit when large language models can perform similar tasks, such as writing emails, summarizing earnings and generating code.

The question first arose in 2024 and 2025 around public cloud companies Amazon, Microsoft and Alphabet, which each invested billions in partnerships with Anthropic and OpenAI. In 2026, with investors watching for possible AI lab IPOs, the focus shifted to whether labs face commodification or can build lasting businesses. If they cannot earn profits, Morningstar says, the concern extends to spending on chips, power and data centers, which assumes AI can produce attractive returns.

Morningstar sees two potential sources of advantage: a data-and-product network effect, in which usage data helps improve products for users, and a structural cost advantage from rising training costs and discounts on gigawatt-scale computing. It describes a six-stage flywheel, with early adoption of coding products as an example, and expects similar dynamics across other uses and user groups as adoption broadens. Anthropic and OpenAI have also used algorithmic advances, custom silicon and multiyear computing deals to reduce the cost of serving their models, according to the firm.

Morningstar’s central economic argument is that demand for inference—using existing models to generate code, text and answers—is growing exponentially, faster than the costs of training and research. Training and research costs are front-loaded and tied to each model generation, while inference costs rise with ongoing use. Labs spent most resources on training and research while building models and tools; inference growth accelerated in 2025-26 as those investments made the systems useful for deployment. The firm says longer-term margins depend more on inference than training, and estimates pure inference could have margins near 80%.

Anthropic reached operating profitability in the first quarter of 2026, reporting 5% operating margins on $11 billion in quarterly sales, as its computing mix shifted toward inference. Morningstar says this happened faster than bearish forecasts that assumed the labs would remain unprofitable indefinitely. Its moat ratings measure whether a company can earn returns on invested capital above its estimated cost of capital: a narrow moat means excess returns are more likely than not for at least 10 years; a wide moat means they are expected for 10 years and are more likely than not to last at least 20 years. Morningstar lists five moat sources: intangible assets, switching costs, network effects, cost advantages and efficient scale. The article notes that its author or authors own shares in one or more of the securities discussed.

Morningstar began coverage of artificial intelligence labs ahead of Anthropic’s expected IPO, asking whether Anthropic and OpenAI can build durable competitive advantages and earn profits. The investment research firm’s first installment in a three-part series argues that leading labs are developing economic moats, though those advantages are not fully formed. Its concern follows years of questions about whether providers can profit when large language models can perform similar tasks, such as writing emails, summarizing earnings and generating code.

The question first arose in 2024 and 2025 around public cloud companies Amazon, Microsoft and Alphabet, which each invested billions in partnerships with Anthropic and OpenAI. In 2026, with investors watching for possible AI lab IPOs, the focus shifted to whether labs face commodification or can build lasting businesses. If they cannot earn profits, Morningstar says, the concern extends to spending on chips, power and data centers, which assumes AI can produce attractive returns.

Morningstar sees two potential sources of advantage: a data-and-product network effect, in which usage data helps improve products for users, and a structural cost advantage from rising training costs and discounts on gigawatt-scale computing. It describes a six-stage flywheel, with early adoption of coding products as an example, and expects similar dynamics across other uses and user groups as adoption broadens. Anthropic and OpenAI have also used algorithmic advances, custom silicon and multiyear computing deals to reduce the cost of serving their models, according to the firm.

Morningstar’s central economic argument is that demand for inference—using existing models to generate code, text and answers—is growing exponentially, faster than the costs of training and research. Training and research costs are front-loaded and tied to each model generation, while inference costs rise with ongoing use. Labs spent most resources on training and research while building models and tools; inference growth accelerated in 2025-26 as those investments made the systems useful for deployment. The firm says longer-term margins depend more on inference than training, and estimates pure inference could have margins near 80%.

Anthropic reached operating profitability in the first quarter of 2026, reporting 5% operating margins on $11 billion in quarterly sales, as its computing mix shifted toward inference. Morningstar says this happened faster than bearish forecasts that assumed the labs would remain unprofitable indefinitely. Its moat ratings measure whether a company can earn returns on invested capital above its estimated cost of capital: a narrow moat means excess returns are more likely than not for at least 10 years; a wide moat means they are expected for 10 years and are more likely than not to last at least 20 years. Morningstar lists five moat sources: intangible assets, switching costs, network effects, cost advantages and efficient scale. The article notes that its author or authors own shares in one or more of the securities discussed.