The SOXX semiconductor ETF gained roughly 90 percent in the first half of 2026. The IGV enterprise software ETF was down nearly 4 percent. A ninety-four-point performance gap between companies building AI and companies selling it. Then, in the holiday-shortened week before Independence Day, the gap started closing.
On July 1 and 2, the Philadelphia Semiconductor Index fell 6.3 percent and 5.4 percent, a combined loss of roughly 12 percent. ServiceNow, Snowflake, and Palantir each surged between 7 and 10 percent. The Dow Jones Industrial Average closed at a record 52,900. These three things happened simultaneously.
The trigger arrived Thursday morning. The Bureau of Labor Statistics reported 57,000 nonfarm payrolls in June, roughly half the 110,000 consensus estimate. Revisions to April and May cut a combined 74,000 jobs from prior counts. The unemployment rate dipped to 4.2 percent, but only because the labor force participation rate fell to 61.5 percent, its lowest since March 2021. The probability of a Federal Reserve rate hike at the July 29 meeting collapsed to 22 percent. The Dow rallied because the economy weakened.
But the rotation preceded the jobs data. Semiconductor stocks began falling on July 1. The catalyst was not economic. It was a market that had spent six months pricing infrastructure and was now discovering it could price adoption instead. Snowflake had reported Q1 product revenue of $1.33 billion, up 34 percent year over year. Palantir had reported 85 percent revenue growth to $1.633 billion. Enterprise software companies were posting AI-driven results. The semiconductor companies that powered those results had already been paid.
This is the distinction the market repriced in two sessions. SOXX gained 90 percent because investors believed AI infrastructure would be built. Hyperscaler capital expenditure confirmed it: Meta guided $125 to $145 billion, Alphabet raised $80 billion in equity, Microsoft launched a $2.5 billion deployment unit with 6,000 engineers. The infrastructure is being built. That trade is complete.
What remained unanswered was who captures the value once the infrastructure exists. Microsoft spent $2.5 billion creating Frontier Company and embedded 6,000 engineers at customer sites because enterprise AI adoption was not happening on its own. Tesla capped employee AI spending at $200 per week. An unnamed company spent $500 million on Claude in a single month. Uber burned its entire 2026 AI budget by April. The tools exist. The question is whether they earn back what they cost.
Enterprise software companies have an answer. ServiceNow set a $1.5 billion AI annual contract value target for 2026. Palantir's government portfolio includes a $10 billion US Army agreement. Snowflake added 616 net new customers in a single quarter, with 779 now spending more than $1 million annually. These companies do not build AI. They sell the outcome of having used it.
Apple illustrated the divide on the same day. The stock rose nearly 5 percent after Nikkei Asia reported Apple had instructed suppliers to prepare 10 million foldable iPhone Ultra units for autumn, up from an earlier target of 7 to 8 million. Apple drove the Dow to its record while the companies that fabricate Apple's processors fell with the semiconductor index. The market priced Apple as a product company, not a silicon company. The value was in the phone, not the chip inside it.
The venture capital data reinforced the pattern. Global startups raised a record $510 billion in H1 2026. OpenAI and Anthropic alone accounted for $217 billion, or 43 percent of all startup capital deployed worldwide. Twenty-four companies were acquired at $1 billion or more in Q2 alone. Capital is not leaving technology. It is repricing which layer of the technology stack deserves the next dollar.
Ninety percent gains in six months is a thesis fully expressed. The market does not need to reprice AI infrastructure. Nvidia, TSMC, ASML, and SK Hynix will remain essential. They will also shift from the trade to the plumbing. Plumbing does not command a premium. It commands a utility multiple.
The Dow closed at 52,900 on a day when the economy added 57,000 jobs and the semiconductor index lost 12 percent. The job market graduated from overheating to cooling. The AI trade graduated from infrastructure to application. Both repricing events pointed in the same direction: toward the companies that sell outcomes, away from the companies that have already delivered the tools. The graduation is the moment the thesis succeeds and the trade ends.