The New Compute Economy

The real competition is moving beneath the model

The next phase of the artificial intelligence race may be less about building ever larger models and more about making existing computing power work harder, faster and at lower cost.

Anthropic’s reported interest in acquiring Nvidia-backed Decart AI for around $6 billion is a useful signal of where the industry may be heading. Decart works across AI infrastructure, optimisation and models, including systems designed for real-time video and simulated environments used in robotics.

For Anthropic, the attraction is potentially much broader than acquiring another AI model developer. As the company expands its hardware expertise and prepares for a possible future public listing, infrastructure efficiency is becoming increasingly strategic.

Inference is becoming the real constraint

Training frontier AI models remains enormously expensive, but the economics are beginning to shift towards inference, the cost of actually running those models millions or billions of times for users and businesses.

As adoption increases, the ability to squeeze more performance from existing computing capacity becomes extremely valuable. Faster inference, better optimisation, specialist chips and more efficient data centres can all reduce the cost of delivering AI services at scale.

This creates an important investment theme around the technologies sitting beneath the models themselves. The winners may not only be the companies building the most capable AI systems, but also those able to make those systems dramatically cheaper to operate.

The risk is valuation. Anything capable of reducing AI infrastructure costs is now attracting intense strategic interest, and prices can rise quickly when major technology companies believe an acquisition could provide a meaningful competitive advantage.

Wall Street moves into AI infrastructure

The scale of investment required to support artificial intelligence is also turning AI infrastructure into a major banking opportunity.

Bank of America has committed $250 billion to US infrastructure financing through July 2027, explicitly including AI data centres, computing infrastructure, power generation, energy storage and transport.

Morgan Stanley has announced around $1.5 trillion of technology and infrastructure financing over the coming decade, while JPMorgan has launched a similar $1.5 trillion programme focused on strategically important industries.

These figures show how quickly artificial intelligence is moving from a technology investment story into a mainstream infrastructure and financing market.

A new financing boom

For banks, the opportunity is potentially enormous.

AI infrastructure requires data centres, semiconductors, electricity generation, transmission networks, cooling systems, land, fibre connectivity and increasingly sophisticated financing structures.

That opens significant opportunities across corporate lending, project finance, private credit, capital markets, infrastructure funds and advisory work.

There is also a wider economic effect. The AI boom is beginning to connect technology investment with energy, property, construction and industrial development in ways that resemble previous infrastructure investment cycles.

The danger of building too much

The obvious risk is concentration.

Much of the current investment assumes demand for AI computing will continue growing rapidly for many years. That may prove correct, particularly if artificial intelligence becomes embedded across almost every major industry.

But infrastructure investment often moves in cycles.

If data centre capacity eventually grows faster than the economic demand for computing, banks and investors could find themselves exposed to extremely capital-intensive assets built on assumptions formed during an investment boom.

The AI infrastructure opportunity could therefore become one of the largest financing markets of the next decade, but it may also become one of the clearest tests of whether enthusiasm for artificial intelligence ultimately translates into sustainable economic demand.

Lomond Logic

The most interesting shift in artificial intelligence may no longer be the race to build the smartest model.

It may be the race to make intelligence affordable enough to use everywhere.