When the Gale Changes Course

Schumpeter’s Gale, otherwise known as ‘creative destruction,’ describes the relentless force of technological innovation. Each gust sweeps what came before: Netflix replaced Blockbuster, Uber displaced taxis, and Amazon transformed retail. This progress has always blown the wind in one direction, toppling the weakest branches whilst allowing the stronger ones to take their place. Today, artificial intelligence has changed the direction of that wind.

Unlike previous waves of innovation, AI does not merely replace companies; it is increasingly replacing tasks, workflows, and in some cases, human judgment itself. The forest is no longer losing a few aging branches, but entire trees are beginning to sway.

Building AI infrastructure requires large amounts of capital, with data centres, GPUs, networking infrastructure and power becoming the new industrial bases of the digital economy. Thus, the very companies driving Schumpeter’s Gale are also becoming some of the largest borrowers in the market. They are both the forces of destruction, and the very companies whose expansion capital markets are increasingly financing.

The question then, is no longer whether Schumpeter’s Gale still blows, it is what happens when the firms driving the creative destruction themselves also become the largest consumers of credit. Historically, technology revolutions do not fail because the tech itself disappoints, rather the financing outruns the fundamentals. The telecom boom of the 1990s demonstrated that innovation can endure even as the capital structure supporting it collapses, but will AI be any different?

Setting the scene

The late 1990s telecom boom offers an insightful comparison. As the internet was becoming widely adopted, many telecom operators rushed to build infrastructure to connect the economy. They were expanding capex to invest in fibre, networks and capture market share before their rivals could, heavily relying on leverage to keep up with competition, only for peak telecom capex to hit $213 billion in 2000.

However, growth failed to materialise as quickly as anticipated, balance sheets could not sustain the leverage, credit ratings deteriorated, and spreads widened. Within a few years, 85% to 95% of the fibre sat dark and unused.

Why this matters: the leverage has moved

Today’s AI investment cycle rhymes with that story, it’s a period of extraordinary infrastructure spending built on expectations of future demand, however, the balance sheet dynamics are wholly different. Unlike the telecom operators of the 1990’s, the largest AI investors – Microsoft, Amazon, Alphabet and Meta – are funding much of their capex through diversified cash flows, and so their balance sheets remain IG, backed by the wider business, capable of absorbing elevated spending.

The financial risk associated has not disappeared, in fact, it has migrated far beyond the forest’s tallest trees. Whilst the hyperscalers possess the balance sheets to withstand years of investment, neocloud providers, data centre operators and specialised infrastructure companies, need to rely on more specialised financing structures, such as private credit facilities and asset-backed lending. Nebius, for example, has raised financing against GPU-backed lending and contracted revenue streams, whilst CoreWeave has used asset-backed lending and delayed draw term loan facilities to fund its expansion.

This distinction matters because the next AI downturn, if it occurs, may not uproot the large parts of the forest the way telecom did. As established, hyperscalers have the balance sheet capacity to withstand slowdowns. The question is whether the companies built around them can survive if AI demand fails to justify the infrastructure being financed.

Unlike telecom, where the companies building the future were also the companies carrying the leverage, AI's financial vulnerability sits amongst the specialised firms expanding capacity around the hyperscalers. A failure among neocloud providers may represent a natural reallocation of capital or it may expose a deeper mismatch between the infrastructure built and the demand required to support it.

The gale has not disappeared; it has changed direction. AI’s strongest trees can withstand a storm; the question is whether the roots beneath the rest of the forest were built to survive the weather.