Fall is officially underway, heralding a season of change — and what a difference three months makes. The past quarter brought a dramatic Federal Reserve policy pivot and notable shifts in market sentiment and ETF trends. But within equities, the most striking reversal occurred inside the AI trade itself.
Key Takeaways
- Since June 30, software (IGV) rallied 17%, while semiconductors (SOXX) fell 22%, wiping out $2 trillion in October market cap.
- After semi ETFs gathered $23 billion in 1H, IGV gathered $2 billion in fresh monthly inflows as software AI revenue materialized.
- Rising debt costs hit builders — pushing CoreWeave yields past 15% — while software offered lower valuation multiples and asset-light balance sheets.
For the first half of 2026, the AI trade ran along a single track. Investors aggressively bought the infrastructure “builders” — chipmakers, memory producers, and power suppliers — while shunning software companies. Wall Street feared autonomous AI agents would undermine traditional subscription-based business models, threatening the software-as-a-service (SaaS) economy.
Fueled by relentless demand for compute capacity, semiconductor ETFs absorbed over $23 billion in net new capital during the first half alone, propelling the iShares Semiconductor ETF (SOXX) to more than double in value. Meanwhile, software strategies faced deep capital redemptions as the iShares Expanded Tech-Software Sector ETF (IGV) lost nearly a fifth of its value as SaaS multiples compressed into the low 20s.
Dramatic Reversal in AI Leadership
Since midyear, the AI trade has turned on its head, with leadership shifting from infrastructure builders to enterprise software companies putting AI to work.
The turning point crystalized in September amid growing questions about the pace and economics of AI infrastructure expansion. Since June 30, IGV has rallied 17%, while SOXX has fallen nearly 22% — tumbling 25% from its 52-week high. On September 14, after Microsoft, OpenAI, and Anthropic called for an “emergency brake” on frontier AI development, software posted its best day relative to chips in 25 years.
Semiconductor funds suffered sharp pullbacks as valuation froth unwound, wiping out nearly $2 trillion (roughly 17%) in market capitalization in October alone. SOXX logged its longest losing streak since 2008, while neocloud provider CoreWeave plunged 25% over five trading sessions, falling 50% from its summer highs.
Semiconductor funds suffered sharp pullbacks as valuation froth unwound, wiping out nearly $2 trillion (roughly 17%) in market capitalization in October alone. Memory exposure was hit particularly hard, with the Roundhill Memory ETF (DRAM) sinking into bear market territory and suffering its largest single-day outflow last week to the tune of $481 million. Overall, SOXX logged its longest losing streak since 2008, while neocloud provider CoreWeave plunged 25% over five trading sessions, falling 50% from its summer highs.
What’s Driving the Rotation?
Several core forces drove this sharp rotation from infrastructure builders and software users:
-
Valuations and Crowding: Semi ETFs were priced for perfection after an explosive first-half rally, leaving indices vulnerable to rapid profit-taking. Software, beaten down after the first quarter’s “SaaSpocalypse,” looked comparatively cheap and presented a compelling mean-reversion setup. Crucially, market breadth expanded into midcap and equal-weighted software strategies. Amid the dip buying, the SPDR S&P Software & Services ETF (XSW) — which caps individual holdings at ~1% — captured strong bids as advisors sought broad software exposure without megacap concentration risk.
-
Software Earnings Resilience: Software firms began reporting AI revenue rather than AI losses, as annual recurring revenue (ARR) picked up from copilots and agentic workflows. That fundamental visibility triggered an institutional pivot: After heavy early-year redemptions, IGV gathered over $2 billion in fresh monthly inflows, signaling that institutional bottom-fishers were stepping in to back software’s earnings floor.
-
Soaring Borrowing Costs: Rising rates and widening credit spreads hit highly leveraged AI infrastructure builders hardest. CoreWeave bond yields breaching 15% sparked concerns over data center unit economics, while asset-light SaaS balance sheets offered insulation.
-
Today’s Shortages, Tomorrow’s Glut? While advanced GPU capacity remains tight today, aggressive long-term capex spend across semiconductor foundries raised fears of an eventual capacity glut, threatening scarcity-driven profit margins.
-
Cybersecurity as AI Safety Net: Rapid AI adoption spurred increasingly complex cyber threats, turning security software into a nondiscretionary tailwind. While broader SaaS models faced scrutiny, IT budgets prioritized zero-trust architectures and automated defense — providing the steady growth that anchored software’s midyear rebound. Focused security allocations via funds like the Amplify Cybersecurity ETF (HACK) helped anchor the broader software bounce back as investors sought software names with bulletproof end-demand.
How Real Are the Risks?
Much of the recent selling reflects apprehension rather than deteriorating fundamentals. AI computing demand remains robust, chip shortages persist, and hyperscalers continue to commit extraordinary sums of money to infrastructure — with major tech giants projected to spend upward of $700 billion on AI development this year alone.
Yet the investment debate is evolving. The question is no longer simply how much AI capacity can be built, but how reliably that capacity can generate revenues, profits, and returns on invested capital.
Software companies currently offer greater earnings visibility, while infrastructure operators face increasing scrutiny over financing costs, utilization rates, and the timing of investment paybacks.
Next Phase: From AI Buildout to Payoff
The AI gold rush isn’t over, but Wall Street is starting to ask a tougher question: Who’s actually making money? After rewarding companies for building ever-greater computing capacity, investors are increasingly scrutinizing whether those massive investments can generate sustainable returns. Every major tech cycle follows a familiar path: first build the pipes, then monetize what flows through them.
This doesn’t mean the infrastructure build-out is over. Demand for advanced chips and computing power remains robust, while concerns over future oversupply and mounting debt burdens are risks rather than foregone conclusions. But after an exceptionally concentrated rally, the market’s reassessment of valuations and returns is a welcome breather.
Investment Playbook: Balance the AI Stack
bottleneck bets
The AI trade isn’t fading — it’s simply maturing. The next winners may be determined less by who builds the most computing capacity and more by who can translate that capacity into durable profits.
Originally posted on ETF Trends
For more news, information, and analysis, visit VettaFi | ETF Trends.
A message from Advisor Perspectives and VettaFi: Discover something new! Click here to register for our upcoming webcasts.
More Innovative ETFs Topics >
