Frontier economics faces proof

Capital as the strategy

Martin Casado, General Partner at Andreessen Horowitz, argues that AI has shifted software development from an engineering-constrained problem to a capital-constrained one. He notes that private markets are efficient and consensus-driven, making non-consensus investing a dangerous strategy, and that defensibility is a second or third-order concern in growing markets. m:…4040

Casado also gives a behind-the-scenes look at recent strategic mergers and acquisitions, noting that successful deals are built on deep personal alignment and trust rather than auction-style poker games. Casado also notes that the lifespan of a state-of-the-art model is currently three to six months, and that the primary constraints on AI infrastructure growth are hardware supply, power and materials, not capital. m:…4040

Fatigue in the trade

On Fast Money, the panel concluded that the AI trade is cooling because investors are shifting focus toward free cash flow, while geopolitical shocks such as a closure of the Strait of Hormuz are more significant market drivers than midterm election outcomes. AMD expects data centers to make up 80% of total sales in 2027. m:…3527

Investors are shifting focus toward free cash flow, while the market historically favors divided government, leading to a bond market rally after midterm elections. That combination, spending continuing but scrutiny rising, is the pattern of a build-out entering a more demanding phase. m:…3527

Builders diversify

Developers are shifting away from single-model setups toward multi-model portfolios and self-hosted AI infrastructure. OpenRouter customers spending $100 or more a month now use a median of 6 model families, double the figure a year ago, while Revolut's self-hosting bill is one-tenth of what it spent on Anthropic and OpenAI, and it measured a 70% improvement in lead time after moving to self-hosted GLM 5.1. m:…1407

The three items describe the same turn from different angles. Investors deploy capital at industrial scale but demand free cash flow, markets stay committed to infrastructure spending while asking harder questions, and builders respond by diversifying providers and bringing models in-house to control cost. The frontier economics story is no longer about whether AI matters; it is about who captures the value, and on what evidence. m:…4040 m:…3527 m:…1407

What the turn means

Casado's warning that non-consensus investing is dangerous reads differently once the market starts asking for free cash flow. In a growth narrative, many companies can raise on promise; in a proof narrative, the questions get harder even as spending continues. The Fast Money panel's conclusion, that the AI trade is cooling as investors shift focus to free cash flow, describes exactly that turn: the build-out continues, but the funding bar rises. m:…4040 m:…3527

Builders are responding the way buyers do when a market matures: they diversify suppliers and bring work in-house. OpenRouter customers spending $100 or more a month now use a median of 6 model families, and Revolut's self-hosting bill is one-tenth of its previous Anthropic and OpenAI spend, both pointing to buyers treating models as substitutable inputs rather than unique partners. That is a healthy development for buyers and a harder one for any single provider counting on lock-in. The frontier economy is growing up, and growing up means being measured. m:…1407

For founders, the implication is direct. Casado notes companies will always pay a premium for the smartest model to maintain a competitive edge, which means cost discipline and model choice are strategic decisions, not procurement details. m:…4040