Oil prices have broken through $100, and semiconductors have plummeted—but will AI really stop because of this?

$NVIDIA(NVDA.US) $Broadcom(AVGO.US) $Oracle(ORCL.US)

Conclusion first: Brent crude oil broke through $100/barrel on March 12th, the semiconductor ETF (SMH) fell 6.4% in a week, and the market fear index slid into "extreme fear." Many analyses believe that the surge in oil prices will drive up the power costs of AI data centers, leading tech giants to cut investments. However, a careful calculation reveals that the proportion of power costs in the total expenditure of AI infrastructure is far lower than most people imagine—the oil price shock is more like an emotional panic, not a fundamental turning point.

Why is the market so afraid?

Iran's new supreme leader claimed that the Strait of Hormuz will "remain closed." This strait handles about 20% of the world's daily oil supply. Oil prices rose 9% in a single day, and Goldman Sachs warned that if the blockade persists, oil prices could hit $150-$200. The three major U.S. stock indices fell for three consecutive days, hitting new lows for the year.

The semiconductor sector fell particularly sharply because the market's logic is: AI data centers are especially power-hungry (AI data centers consume 3-5 times more power than regular data centers) → oil prices rise → electricity costs rise → it's too expensive for tech companies to build data centers → reduced investment.

Where is the flaw in this logic? Electricity costs aren't that important.

The biggest expense for data centers is not electricity, but hardware—GPU chips, servers, storage equipment, etc., accounting for over 35% of the total cost. Power accounts for 15%-25% of a data center's operational costs, but if we include construction costs (i.e., the Total Cost of Ownership, TCO), the power proportion drops to single-digit percentages.

An analogy: You buy a $5 million house with a monthly electricity bill of $3,000. If the bill increases by 20% to $3,600, would you decide not to buy the house because of that? No. The situation with AI data centers is similar—electricity costs are indeed rising, but the impact is negligible compared to construction investments that often run into tens of billions of dollars.

A more crucial point: The money has already been spent.

The combined capital expenditure plans of the five major cloud computing giants (Amazon, Google, Microsoft, Meta, Oracle) this year total about $65-$75 billion, and they were already locked in during the February earnings season. Amazon $200 billion, Google $175-$185 billion, Meta $115-$135 billion—these numbers are annual plans approved by the board of directors, not monthly budgets that can be adjusted at any time. GPUs have been ordered, data centers are already under construction, and long-term power purchase agreements (PPAs, fixed-price contracts with power companies) have been signed.

Oracle even holds $553 billion in undelivered orders, locked in until 2028. Broadcom's CEO explicitly stated in the earnings call: All component supplies supporting revenue for the next two years have been fully locked.

Therefore, oil prices rising from $70 to $100 will not make these companies change their already-signed investment plans worth hundreds of billions of dollars.

What should we really worry about?

The real risk of the oil price shock to the AI industry chain is not the increase in electricity costs, but two indirect things:

First: Interest rates. The market no longer expects the Fed to cut rates this year. The yield on the 10-year U.S. Treasury note (which can be understood as the benchmark cost of borrowing) has risen to 4.26%. Tech giants need to borrow over $400 billion this year to build data centers; the higher the interest rate, the greater the borrowing cost.

Second: The risk of economic recession. Wall Street's well-known analyst Yardeni raised the probability of a U.S. recession to 35%. If a real recession occurs, AI investment by small and medium-sized enterprises will shrink—but the arms race among hyperscale customers is unlikely to stop, because the logic of "not building equals losing" still holds.

Specifically for investment, what's the view?

If you agree with the judgment that "the oil price is a short-term shock, not a structural turning point," then the current panic selling in the semiconductor sector provides a temporary window. The latest EIA (U.S. Energy Information Administration) forecast: Brent average price in Q2 around $91, falling back to $70 in Q4, and dropping to $64 in 2027—provided the strait gradually resumes navigation.

A few noteworthy targets:

Oracle ORCL: Up 9% today (earnings beat expectations), cloud infrastructure revenue up 84% year-over-year. Current price-to-sales ratio (PS) only 1.8x, compared to about 4x for peers like AWS.

Broadcom AVGO: AI chip Q2 guidance exceeds expectations by 15%, FY2027E P/E ratio only 16.6x.

NVIDIA NVDA: GTC conference opens next week (March 16-19), Neocloud track confirmed against the trend.

If you are worried about an extreme scenario like "strait blockade lasting over 3 months, oil prices persistently above $120," then you indeed need to wait for clearer signals before making a judgment—in that case, all risk assets would be under pressure, not just semiconductors.

What to watch next?

Signal 1: Whether Meta and Google's Q1 earnings calls in April-May release signals of capex reduction. This is the only hard evidence that could disprove the "AI capex unchanged" judgment.

Signal 2: Whether Brent can stabilize between $95-$105 instead of continuing to surge. The EIA's baseline assumption is that the strait resumes navigation within Q2.

Signal 3: The CPO technology roadmap and customer cooperation progress released by NVIDIA at the GTC conference (March 16th), which will directly affect short-term sentiment in the AI interconnect sector.

Do you think the impact of oil prices breaking $100 on the AI industry chain is a short-term panic or a long-term turning point?