The Forecasting Research Institute’s panel of AI experts and superforecasters expects semiconductor stocks to lose steam through 2028 while software stocks outrun their historical growth pattern, according to the group’s newest survey wave. The split rests on one question the panel itself flagged as unresolved: whether artificial intelligence ends up selling more software or making large swaths of it obsolete. That framing matters because it treats the AI trade as a bet on where value settles in the technology stack, not a single uniform boom.
The forecasts come from Wave 11 of the Longitudinal Expert AI Panel, known as LEAP, run between July 13 and August 11, 2026. Respondents included subject-matter experts, professional superforecasters, and members of the public, each asked to predict investment and stock trajectories rather than assert them as outcomes.
On chips, the median superforecaster expects the VanEck Semiconductor ETF (SMH) to reach 1.36 times its July 2026 value by the end of 2028; the median expert predicts 1.5 times. A straight-line extrapolation of the fund’s run since January 2024 would put it at roughly 3.37 times that value. Both groups assign the trend-line outcome less than a 10 percent chance (experts: 8 percent, superforecasters: 7 percent), and both consider an outright decline more probable than staying at trend (17 percent and 27 percent, respectively).
Software gets the opposite treatment. The panel expects the iShares Expanded Tech-Software Sector ETF (IGV) to sit at 1.2 to 1.3 times where it stood in mid-2026 once 2028 closes, ahead of a trend-extrapolated 1.14 times. Experts give a 63 percent chance the fund beats its own trend line; superforecasters give it 58 percent. Written rationales split along a familiar fault line: forecasters bearish on software argued AI could hollow out incumbent vendors, while bulls argued those same vendors are positioned to convert the threat into new revenue.
The panel also forecasts continued growth in physical AI infrastructure spending despite regulatory friction, including New York’s statewide moratorium on new hyperscale data centers passed in July. From a 2025 baseline of $40.8 billion, experts predict data center structure investment will climb 76 percent by the end of 2028; superforecasters predict 70 percent. Forecasts for the broader IT equipment category, a $514.8 billion market in 2025, run higher still: roughly 48 percent growth by 2028 among experts, with both groups projecting the category will pass $1 trillion by 2035. Investment in electrical and communication infrastructure, tied more loosely to the AI buildout, is expected to grow far slower, around 15 percent by 2028.
The Institute flagged a separate problem with its own methodology. Superforecasters put the median combined annualized revenue run rate for OpenAI and Anthropic at $300 billion by 2030, assuming at least one remains independent, versus $72 billion at the time of the survey. But the group noted that both expert and public median forecasts landed below the run rate already reported by the time the survey closed, suggesting some respondents anchored to stale figures rather than the most current numbers available to them. The Institute is now running a separate accuracy study on how elicitation design affects forecast quality.
None of this is a prediction about what AI will do to the economy; it is a prediction about what a structured panel of forecasters currently believes, with the panel’s own confidence levels attached. The semiconductor call in particular is a bet against continuation of the exact pattern that has defined AI investing since 2024, and the panel puts real odds, not certainty, on that reversal actually holding. Investors and operators weighing AI infrastructure exposure over the next two years should treat the software-over-semiconductors call as a hypothesis with a stated failure rate, not a settled market view.
Forecasting Research Institute, Wave 11 LEAP survey, published August 27, 2026.
