Through June 30, 2026, the S&P Semiconductors Select Industry Index had gained 144%, versus 20% for the S&P 500 Top 10 Index, according to S&P Dow Jones Indices. The figures show that the AI trade has diverged by layer: the market’s largest gains have gone to the physical infrastructure required to run AI, not to technology stocks as a single group.
NVIDIA reported fiscal second-quarter 2027 revenue of $96.2 billion, up 106% year over year. Data Center revenue rose 117% to $89.0 billion and represented roughly 93% of total sales. That concentration makes the connection between AI infrastructure buildouts and reported revenue unusually direct.
For software investors, the central issue is where AI spending will accrue. It may become a source of new contracted revenue, or it may reduce paid seats and weaken the pricing structure that supported the software-as-a-service model. The software selloff is thus a repricing of where value can be captured in the AI stack, rather than simply a verdict on technology spending.
Chipmakers are capturing AI spending before software monetization is settled
Infrastructure suppliers are benefiting from demand that is visible in orders, system shipments and data-center revenue. NVIDIA’s figures, disclosed in its quarterly results, provide the clearest example. Nearly all of the company’s revenue came from Data Center in the quarter, so the effect of AI buildouts does not need to be inferred from a product roadmap or a promise of future adoption.
That is different from the position of many software companies. A customer can spend on computing capacity before deciding which applications, workflows or agent products will earn a recurring budget. The infrastructure layer is therefore receiving revenue at the construction stage of the AI cycle, while application vendors are still establishing what customers will pay for once those systems are deployed.
The 144% semiconductor-index gain should not be read as evidence that every chip business has the same exposure or that software is absent from the AI trade. It does, however, show how forcefully investors have rewarded the part of the market with the most measurable near-term demand. NVIDIA’s growth has given that preference a financial anchor.
This sequencing can produce an uncomfortable gap in valuations. Chipmakers can show the revenue impact of capital expenditure today. Software companies may need to demonstrate that an AI feature is not merely an added cost or a defensive product response, but a service that expands contract value without undermining the existing subscription base.
AI agents challenge the per-seat SaaS model
The pressure on software shares is rooted in a specific economic concern, not just a broad fear that AI will disrupt everything. Investors worry that AI agents could compress pricing power and reduce demand for traditional per-seat SaaS products, according to Nasdaq Global Indexes. Nasdaq described February 2026 market performance as a software-led selloff tied to those structural concerns.
Per-seat pricing has a straightforward logic when more employees using a service mean more licenses sold. Agents complicate that logic if they automate tasks that previously required a larger number of users to interact with a software product. The market is consequently assessing two risks at once: whether customers can demand lower prices for work handled by automation, and whether they may need fewer conventional licenses.
Neither risk means every software category will be displaced. The effect depends on whether a company sells a narrowly defined application exposed to automation, or controls the data, workflows and customer relationships through which AI is deployed. But the burden of proof has shifted. Revenue growth alone may not settle the question if investors believe that growth rests on a pricing model vulnerable to a change in how work gets done.
That helps explain why the sector’s reaction has been more severe than a routine rotation away from growth stocks. The issue is not only the cost of building AI capabilities. It is whether the technology changes the unit of value from a named user to an automated outcome, with uncertain implications for established subscription economics.
Salesforce and ServiceNow show software is being sorted, not simply displaced
Salesforce’s fiscal second-quarter 2027 results put specific numbers behind the AI debate: revenue was $11.3 billion, up 11% year over year; Agentforce and Data 360 annual recurring revenue reached nearly $3.9 billion, up more than 210%; and the company raised its fiscal 2027 revenue outlook to $46.1 billion to $46.4 billion. The figures were reported in Salesforce’s earnings release.
ServiceNow supplied a separate data point, reporting second-quarter 2026 subscription revenue of $3.877 billion, an increase of 24.5% from a year earlier. ServiceNow AI surpassed $1 billion in annual contract value, according to the company’s results. That measure is not recognized revenue, but it does indicate customer commitments to AI products.
Neither report settles how durable AI economics will be across software, and neither proves that the broader software selloff was mistaken. They do undercut a blanket conclusion that AI is destroying software demand and point toward a more granular market judgment.
The relevant divide is how companies monetize the technology. Platforms able to attach AI to established data, enterprise workflows and contracted relationships have a visible route to capturing spending; businesses viewed as selling more replaceable, seat-based functionality face a tougher valuation debate. “Software” is therefore too broad a category if AI threatens some applications while supporting new contract value at companies with the distribution and operational role to sell it.
The infrastructure winners still depend on hyperscaler returns and physical buildout
The semiconductor rally rests on strong reported demand, but execution and valuation risks remain. NVIDIA said in its SEC filing that Blackwell remained the majority of system shipments and that Vera Rubin began production shipments in fiscal third-quarter 2027. The filing also cautioned that customer demand estimates may prove inaccurate and that shortages of land, power, data-center shells and capital could affect future revenue.
That warning points to a constraint beyond chip supply. AI computing demand must become deployed infrastructure, and the pace of that conversion depends on power, physical space, data-center capacity and financing. A surge in planned capacity can therefore outpace the facilities needed to install it.
The market has begun testing the financial side of the buildout as well. In July, the Philadelphia Semiconductor Sector Index had fallen 4.5% from its recent level as investors questioned elevated chip valuations and whether hyperscalers would earn adequate returns on record AI spending, Axios reported.
Although semiconductor stocks had substantially outperformed through June, that pullback showed the infrastructure trade remains exposed to a second stage of monetization. Chipmakers are converting the buildout into revenue; their largest customers still must demonstrate that the resulting capacity can produce acceptable returns. Software companies face a different burden: showing that AI agents strengthen rather than erode recurring revenue.
