Two Screens
Two Screens
Equities & Energy · Monday, 11 May 2026 · No Attribution
Screen one: Micron up 124% year-to-date. SK Hynix at a record in Seoul this morning, +5% before lunch. The Philadelphia Semiconductor Index has gained 55% in a single quarter. The Roundhill Memory ETF — DRAM, and yes, that ticker is doing exactly what you think it's doing — soared 30% last week alone. Sandisk up sixfold on the year. Micron's market cap just cleared $700 billion. The S&P 500 broke 7,300 on Friday, a record. Six consecutive weeks of gains.
Screen two: Brent crude surged this morning after Trump called Iran's latest peace proposal "totally unacceptable." The Strait of Hormuz has been functionally closed since late February. The IEA says the conflict is pulling 14 million barrels per day from global supply. US distillate inventories sit 11% below their five-year average. Midwest refinery outages compounding the shortfall. University of Michigan consumer sentiment deteriorating further in May. Inflation expectations still above their long-run averages. April CPI drops tomorrow and the consensus is calling for 3.7% year-over-year. Cleveland Fed's model had it at 3.6%. It doesn't matter which one is right — neither is good.
Both of these screens are showing you the same economy.
The thing that should unsettle you, the thing worth sitting with before the open today, is not that these two realities are in conflict. It's that the market has decided — coolly, systematically, with full information — that they aren't.
Tech earnings beat estimates at an 83% clip this season. That's against a long-run average of 67%. The hyperscalers — Alphabet, Amazon, Microsoft, Meta — collectively guided to $725 billion in capital expenditure for 2026. That's a 77% increase on last year's already-unhinged record of $410 billion. Amazon is spending $200 billion. Alphabet bumped its guidance to $180–190 billion and flagged that 2027 will be "significantly higher." Google Cloud grew 63% in Q1. The cloud contract backlog at Alphabet alone sits at $460 billion — roughly double where it was six months ago.
At least two of these companies explicitly blamed memory chip prices for pushing their budgets above prior guidance. DRAM contract prices rose roughly 95% quarter-over-quarter in Q1. Q2 is projected to add another 58–63% on top of that. All NAND output for 2026 is already committed. The companies spending $725 billion on AI infrastructure are doing so into a memory market where prices are doubling every two quarters and supply is maxed.
This is not a sector story. This is a capital allocation story at a scale the modern economy has never run before — and it's being stress-tested by an energy shock simultaneously.
The market's working theory seems to be that these two things — the AI infrastructure build and the oil price spiral — exist in separate compartments. Energy is a geopolitical problem; AI is a structural opportunity. One will resolve. The other will compound. Buy the semiconductors, hedge the crude, collect the spread.
It's a tidy theory. It might even be right. But there's a version of events where it isn't, and that version deserves airtime because nobody seems particularly interested in it right now.
The version where it isn't right goes something like this: the energy shock is not resolving. Trump called today's Iranian proposal "totally unacceptable." The diplomatic framework that produced the April ceasefire has been grinding backward ever since. The structural thesis underpinning the AI trade — that productivity gains will arrive fast enough to justify $725 billion per year in infrastructure spending — has not been tested in an environment of prolonged energy-cost inflation, tariff-driven input cost escalation, and a central bank whose incoming chair has pre-committed to cuts while inheriting 3.7% CPI.
Because here's what the memory price surge is also telling you: the component costs feeding AI infrastructure are themselves energy-intensive. Semiconductor fabs are among the most power-hungry facilities on Earth. Every DRAM wafer that goes into an Nvidia HBM stack passes through a supply chain that prices power. When energy is $120 a barrel and climbing, the deflation story embedded in every AI bull thesis takes longer to arrive. The productivity gains are still coming — no reasonable person argues otherwise — but the timeline has a cost structure attached to it that current valuations may not fully reflect.
Nintendo fell 10% in Tokyo this morning on warnings about higher chip prices. That's the canary nobody wants to look at. A company entirely outside the hyperscaler orbit, making consumer hardware, getting squeezed by the same memory repricing that's sending Micron to record highs. The spoils of a shortage economy are not evenly distributed.
Cloudflare dropped 24% on Friday — cutting 20% of its workforce and guiding Q2 revenue slightly below consensus. In the same session, the S&P hit an all-time high. The divergence inside tech is widening into a canyon: infrastructure winners, application losers; the picks-and-shovels crowd in the stratosphere, the people trying to build things with the picks and shovels getting priced out.
The question isn't whether AI is real. It is. The question isn't whether the hyperscalers will earn back their capex. They probably will. The question is whether the 55% quarterly gain in the Philadelphia Semiconductor Index is pricing a 2027 world that requires this week's geopolitical and monetary environment to resolve in exactly the right sequence — ceasefire by the Trump-Xi summit on May 14–15, Warsh confirmed and benign, April CPI soft enough to not spook the bond market — and what happens to that pricing if the sequence breaks.
The market has essentially written itself a script and is now rooting for reality to follow it.
Screen one is up. Screen two is also up — oil this morning, after the news.
Both are real. The question is which one is leading.