Every Company Near the Rack Will Be Bought

In an AI fever, capital first looks for exposure; only later does it look for quality.

When a narrative becomes too strong, the market loses subtlety. At first, it buys the obvious. Then it buys the secondary. Then it buys the peripheral. Then it buys any company that can place "AI", "data center", "GPU", "rack", "liquid cooling", or "server" near its own name. This is not an anomaly. It is typical fever behavior. Capital first seeks exposure. Only later does it seek quality.

In December 2023, the AI boom is no longer secret. Nvidia has become the sun of the system. AMD is trying to be the alternative. But the market begins looking for satellites: Supermicro, Vertiv, Dell, Celestica. Companies that touch the rack begin receiving attention because the thesis has expanded from chip to physical infrastructure. The question stops being "who makes the best accelerator?" and becomes "who can deliver systems, energy, cooling, integration, and capacity quickly?"

Supermicro is the name that best expresses the rush phase. When demand for optimized servers explodes, an agile integrator can look indispensable. The company able to assemble systems with GPUs, adequate thermal architecture, and delivery speed becomes a favorite. But speed is a knife. It increases revenue and also increases execution risk, working capital, internal control, supplier dependence, and margin quality risk.

Vertiv represents critical infrastructure. If racks become denser, cooling and energy become bottlenecks. Dell represents the enterprise channel, the server sold to companies that do not buy like hyperscalers. Celestica represents manufacturing, supply chain, and industrial capacity for electronic infrastructure. Each one touches part of AI's materialization.

Perhaps in 2024 the market buys all these stories with little discrimination. This creates opportunity and danger. The investor can win from narrative expansion, but needs to leave the adolescence of the thesis before adult accounting arrives. Not every company near the rack has the same margin, the same controls, the same competitive advantage, or the same pricing power.

The way to profit is to understand the phases. Phase one: exposure. Phase two: growth. Phase three: quality. Phase four: selection. In the exposure phase, any proximity to the theme works. In the quality phase, the market asks who converts revenue into profit, who has a clean balance sheet, who controls working capital, who has pricing power, and who does not depend on a short demand window.

Nvidia and AMD still lead the imagination. But if their multiples embed too much, capital looks for infrastructure beta. Supermicro can be violent beta. Vertiv can be a more industrial thesis. Dell can be the revaluation of an old company with new exposure. Celestica can be a hidden supplier the market rediscovers. Each offers different convexity and different risks.

The counter-thesis is obvious: this phase tends to produce exaggeration. Supermicro can be priced as if fast growth were permanent quality. Dell can capture revenue, but not necessarily extraordinary margins. Vertiv can become expensive if everyone discovers the same bottleneck. Celestica can be treated as an "AI play" beyond what its fundamentals allow. AMD can disappoint if the alternative takes too long. Nvidia can remain strong and still suffer compression if expectations become absurd.

The mature investor must accept that one can win in a speculative phase without lying to oneself. The problem is not buying a fever. The problem is forgetting it is fever. Whoever enters through flow must exit through flow or demand fundamentals that justify staying. Mixing a narrative trade with a patrimonial marriage is one of the most efficient ways to destroy capital.

In 2024, the market may still be buying capacity. Later it will buy margin. Then cash flow. Then it will punish whoever grew without controls. This script is not prophecy. It is cycle hygiene.

Every bubble begins democratic.

Then it becomes cruel.

Leo Bentier