The chip industry just told the White House to keep its hands off memory pricing.

The Letter

SEMI, the trade association representing chipmakers and equipment suppliers, sent a letter to senior Trump administration officials warning that any government attempt to influence memory prices or production capacity would deepen the historic supply squeeze rather than fix it. The AI buildout is the driver of that squeeze, and SEMI’s ask is narrow: keep letting companies sign long-term supply agreements with customers, and extend the tax breaks already aimed at growing US output. In other words, more market mechanism, not less.

The US small modular reactor industry crossed a psychological threshold this year. No SMR has entered commercial operation domestically, but 2026 is shaping up as the year the first construction permits and criticality milestones actually land — and the trigger isn’t climate policy, it’s AI data-center power demand.

The Federal Push

The Department of Energy’s Gen III+ SMR Pathway to Deployment Program has become the primary vehicle for de-risking early builds. DOE issued a $900 million solicitation in March 2025, followed by $800 million in Tier 1 awards in December 2025 to the Tennessee Valley Authority and Holtec Government Services. In May 2026, DOE added a further $94 million across eight companies under Tier 2, targeting licensing, supply chain, and site preparation gaps that have historically stalled domestic nuclear projects.

The June employment report landed well short of expectations. The Bureau of Labor Statistics reported nonfarm payrolls rose by just 57,000 in June, against a Dow Jones consensus of 115,000 and a range of Wall Street estimates that ran as high as 180,000 from more bullish shops like RSM. The unemployment rate fell to 4.2 percent from 4.3 percent, a move that would ordinarily read as a positive but instead is consistent with people leaving the labor force rather than a genuine acceleration in hiring.

Every memory forecast published in the last twelve months has been wrong in the same direction: too low. IDC, TrendForce, and Bank of America have each revised DRAM and NAND demand estimates upward multiple times since early 2025, and the pattern is not noise — it is a structural failure of forecasting methodology colliding with a demand curve that refuses to plateau.

The Forecasts Keep Missing In One Direction

The current numbers are already staggering. IDC now expects 2026 DRAM supply growth of only 16% year-on-year, with NAND supply growth at just 17%, both well below the 20-30% historical norms that defined the post-2018 memory market. HBM demand alone is projected to grow 70% year-over-year in 2026, with HBM consuming 23% of total DRAM wafer output, up from 19% the year before. Bank of America forecasts DRAM revenue surging 51% year-over-year and NAND 45%, with ASPs rising 33% and 26% respectively.

Palantir closed July 1 at $125.73, up 7.8% on the day and adding roughly $21.7 billion in market value on unusually heavy volume of 57 million shares. The move capped a nine-day run of catalysts, but the headline driver was Alex Karp’s appearance on CNBC’s “Squawk Box,” where the CEO combined a genuine product announcement with a public airing of grievances against the rest of the AI industry.

The Announcement Underneath The Noise

The interview was nominally about Palantir’s expanded partnership with Nvidia: an “intelligent engine” that runs Nvidia’s Nemotron open models inside sovereign, secured environments for government agencies and critical infrastructure operators. The pitch is data and model-weight control — customers keep ownership of their compute and their data stack rather than routing it through a third-party API.

Every AI capex chart published this year gets laid over the same overlay: 1999-2000. Capex-to-sales ratios, VC concentration, IPO froth, Fed warnings — the comparison is everywhere, and it is not baseless. But it is incomplete in a way that matters, because it measures this cycle using the only historical yardstick available and assumes the yardstick still applies.

The Comparison Investors Keep Reaching For

The parallels are real and worth taking seriously. Big Tech’s combined AI capex is expected to reach $650-700 billion in 2026, and one Wells Fargo analysis noted this drove 42% of Q1 GDP growth while representing 2.4% of total US GDP — a figure some analysts expect to surpass dot-com-era peaks by Q4 2026. MIT research found that 95% of organizations report zero measurable return on their GenAI investments so far, and the ratio of infrastructure spending to actual AI software revenue runs close to twenty-to-one. Morgan Stanley has pointed out that capex-to-sales ratios are on track to exceed the 32% peak hit at the top of the dot-com bubble.

Private employers added just 98,000 jobs in June, according to the ADP National Employment Report released Wednesday, missing the Dow Jones consensus of 110,000 and down sharply from May’s unrevised 122,000. The miss lands one day ahead of the Bureau of Labor Statistics’ more closely watched nonfarm payrolls count for June, due Thursday.

The headline number undersells the composition problem

Nearly all of June’s job growth came from services, and within services, one sector did the heavy lifting: education and health services alone accounted for 48,000 of the 98,000 total, roughly half. Trade, transportation and utilities added 15,000, financial activities 14,000, and other services 8,000. Goods-producing industries barely registered — manufacturing added 5,000, construction just 2,000 — while natural resources and mining was the only sector to shed jobs, down 5,000. Leisure and hospitality, often read as a proxy for consumer demand, added only 2,000 positions.

Samsung and SK Hynix unveiled a combined roughly $1.3 trillion (2,000 trillion won) decade-long investment plan for new fabs, AI data centers, and chip cluster development. Both stocks fell anyway — Samsung down over 5%, SK Hynix down over 3% on the announcement day, following an even sharper 9%+ plunge earlier in the week. The knee-jerk read: investors think the spending is reckless, a repeat of the 2018-2019 memory bust, or proof the AI trade is cracking.

There is a respectable bull case that benefits eventually grow faster than capital expenditure, and it does not depend on optimism. It depends on arithmetic. The trouble is that the same arithmetic reduces the entire thesis to a single contested number, and that number is currently being attacked by the man who shorted the last bubble.

The Mechanical Case Is Stronger Than the Hype Case

Strip away the cheerleading and the bull argument is structural. Capex is a step function: compute, data centers, substations, and grid interconnects are sunk before a dollar of inference revenue arrives. Monetization is an S-curve that begins only once capacity is utilized. The two curves are offset in time by construction, which means peak capex intensity is not the steady state of the benefit-to-cost ratio. It is the trough. The ratio improves mechanically the moment spending shifts from building capacity to running it, even if technological progress stopped entirely.

On June 2, Jensen Huang turned to Matt Murphy on a Computex stage in Taipei and called Marvell the next trillion-dollar company. The stock rose 32.52% that day, its largest single-session gain on record, adding roughly fifty billion dollars in market value before the close. Every desk on the Street ran the clip. Almost none of them ran the number underneath it.

The number was disclosed six days earlier, on the May 27 fiscal first-quarter call, and it carried no theater at all. Marvell raised its forward guide by roughly five billion dollars and lifted interconnect growth from 50% to over 70% year over year. The trillion-dollar line moved the tape. The guide raise moved the thesis. Those are not the same event, and conflating them is how investors end up paying for sentiment while telling themselves they bought fundamentals.