The Bubble Rewards Capacity and Punishes Bad Margin

In AI's second phase, the market stops buying mere proximity and starts demanding cash.

Every productive bubble is born from a truth. That is the part that confuses. Bad bubbles are easier to avoid when they are born from pure lies. But dangerous bubbles are born from real fundamentals: the internet was real, fiber was real, smartphones were real, cloud was real, AI is real. The problem is not reality. The problem is the price the market pays for anything that touches reality.

In March 2024, AI has already become the dominant narrative. Nvidia and AMD concentrate attention. Supermicro, Vertiv, Arista, and Broadcom enter as ways to play the chain. The first phase rewards capacity. Whoever delivers GPU, server, network, energy, or cooling receives a premium. The second phase, however, will be less gentle. The market will begin asking: which revenue has quality? Which margin is sustainable? Which balance sheet can endure? Which growth depends on working capital? Which company has internal control? Which supplier has real power and which one merely assembled quickly?

Supermicro is the symbol of the risk. It can grow a lot and still demand hard analysis. Fast growth in hardware can consume cash, pressure inventory, increase receivables, depend on suppliers, and create operational complexity. The market loves accelerated revenue until it discovers that revenue needs to be financed. The difference between growth and swelling appears in the balance sheet before it appears in the headline.

Vertiv is different: critical infrastructure with real demand, but still subject to capex cycle, execution, and valuation. Arista has higher quality, but needs to justify price and dependence on large customers. Broadcom has diversification, silicon, software, and capital discipline, but it is not pure exposure. Nvidia can be excellent and still vulnerable to superhuman expectations. AMD can be treated as an alternative before proving enough capture.

Perhaps by the end of 2024 the market begins punishing companies that rose only through proximity. This will not destroy the AI thesis. On the contrary, it is a sign of maturity. Serious narratives need to expel parasites. The first market buys theme. The second buys fundamentals. The third buys cash flow. The investor should try not to be holding the wrong stock during the passage from one phase to another.

The way to profit is to create filters. First: gross margin. Who has pricing power? Second: working capital. Who needs to finance too much growth? Third: customer concentration. Who depends on a few giant buyers? Fourth: backlog and conversion. Does order become revenue? Does revenue become cash? Fifth: technical advantage. Who is replaceable? Sixth: governance. Whoever grows without controls can become a trap. Seventh: valuation. A good thesis at an absurd price becomes a bad investment.

The counter-thesis is that euphoria can last longer than the skeptic can endure. Shorting a real bubble is an activity for men who like being right and broke. Companies can keep rising despite clear risks. Nvidia can keep delivering large numbers. Supermicro can survive the doubts for longer. Vertiv can keep revising upward. Arista can maintain quality. Broadcom can be revalued as broader AI infrastructure. AMD can receive credit for roadmap. The market owes no obedience to the prudent analyst.

But prudence is not pessimism. Prudence is knowing that every narrative eventually meets accounting. The question is not whether AI is real. It is who transforms AI into free cash flow after enthusiasm pays for the first round of capacity.

The investor who wants to survive needs to love the truth more than the thesis. When the numbers change, the thesis must change. When margin does not appear, the thesis must shrink. When growth requires too much financing, the thesis must be discounted. When governance fails, the low price may still be expensive.

Every bubble first rewards capacity.

Then it asks who knows how to manage it.

Leo Bentier