The AI Infrastructure Bubble: Where Does the Money Go?

The recent news that Oracle has spent $28.5 billion on capital expenditures, primarily for its cloud capacity for artificial intelligence (AI), raises a pressing question: what happens to all this money? Beneath the hype surrounding AI and cloud computing lies a complex web of infrastructure costs.

Oracle’s quarterly revenue has risen 30% to $19.3 billion, driven by over $30 billion in new AI contracts. This growth is mirrored in Vertiv Holdings Co., which operates in the layer below Oracle’s cloud boom. Due to grid constraints pushing data-center operators toward self-generation, Vertiv is expanding further into onsite power and cooling.

To put this spending into perspective, consider the average cost of building a large-scale data center: construction costs can range from $10 million to over $50 million per megawatt (MW) of capacity. With Oracle’s recent spending, we’re talking about building multiple data centers with capacities exceeding 100 MW each – an enormous undertaking.

The driving force behind this spending is the AI revolution itself. As companies like Oracle push the boundaries of cloud computing, they create an insatiable demand for processing power and storage. This fuels a frenzy of investment in infrastructure suppliers like Vertiv, which specializes in power and cooling systems.

However, this spending spree comes with risks. The bearish case against Oracle and other high-growth tech companies hinges on their return on capital – can these investments justify today’s construction costs? As the industry becomes increasingly saturated with AI data centers, it’s essential to consider the long-term viability of such projects.

History has shown us that infrastructure spending follows a cyclical pattern: periods of rapid growth give way to periods of consolidation and contraction. Vertiv’s recent acquisition of Utility Innovation Group for $1.45 billion is a telling indicator of this trend. While Vertiv’s bull case relies on continued data center demand, its bearish case centers around valuation and cyclicality – will infrastructure suppliers be able to weather the inevitable slowdown in hyperscaler spending?

In this rush to build AI capacity, we’re seeing a new kind of economic activity emerge: the creation of microgrids and onsite power generation is becoming increasingly important as data centers struggle to access sufficient electricity from the grid. This shift has significant implications for the future of energy consumption and production – will we see a proliferation of small-scale power generation and storage systems, or will traditional utility companies adapt to meet the changing needs of the industry?

As Oracle and Vertiv continue to spend billions on infrastructure, it’s clear that the AI infrastructure bubble is becoming increasingly unsustainable. While this spending may drive short-term growth, it’s essential to consider the long-term implications of such investments – will they justify their costs, or will we see a repeat of past cycles in the tech industry? The stakes are high, and so is the pressure to deliver returns on capital. Will Oracle and its peers be able to keep pace with the demands of this rapidly evolving landscape?