Photo by Jakub Ε»erdzicki on Unsplash
According to market analysis tracked by Google News from TradingView, the divergence between AI hardware and software investment vehicles has reached a level that demands a second look β particularly for investors who may have written off the software category entirely.
The Common Belief: Hardware Is the Only AI Trade That Matters
8.4 percentage points. On a single trading day β June 2, 2026 β chipmakers outran software stocks by that margin, the largest single-session divergence on record since mid-2021, according to Bloomberg. For investors who positioned in semiconductor ETFs last year, that number feels like confirmation: the AI infrastructure buildout is a hardware story, and traditional software-as-a-service (SaaS) companies are bystanders at best, casualties at worst.
The macro numbers reinforce that reading. As of June 17, 2026, hyperscalers β Microsoft, Alphabet, Amazon, Meta, and Oracle β have committed a combined $660β690 billion in capital expenditures for the year, nearly doubling 2025 levels. Roughly 75% of that, or more than $450 billion, is directed at AI infrastructure: data centers, custom chips, and networking fabric. Amazon alone has penciled in $200 billion in 2026 capex, followed by Alphabet at $175β185 billion, Meta at up to $135 billion, and Microsoft at $120 billion-plus. Consensus estimates placed 2026 hyperscaler capex at $611 billion earlier in the year; actual committed figures have since pushed past that ceiling.
Against that backdrop, the VistaShares Artificial Intelligence Supercycle ETF (AIS) had surged 119% through June 3, 2026. BlackRock's newly launched iShares Future AI & Tech ETF (ARTY) went hardware-heavy from inception, weighting TSMC, Marvell, NVIDIA, AMD, Broadcom, and Micron. The ROBO Global robotics ETF suite (ROBO, THNQ, HTEC) grew from $4.1 billion to nearly $7 billion in assets during 2026. The Global X U.S. Infrastructure Development ETF (PAVE) attracted nearly $2 billion in net inflows year-to-date as investors positioned for domestic AI data center construction.
Meanwhile, the iShares Expanded Tech-Software Sector ETF (IGV) sat at -20.62% year-to-date through mid-June 2026. The WisdomTree Cloud Computing Fund (WCLD) clocked -20.28% over the same period. Bloomberg Intelligence gave the underlying fear a name: the "SaaSpocalypse" β a scenario where generative AI tools would mean software companies "need fewer workers, sell fewer licenses, and cease to be going concerns." The conventional trade has been simple: if AI is a new infrastructure cycle, buy the shovels (chips), not what runs on them.
Where It Breaks Down
The SaaSpocalypse thesis has a logic problem, and mid-2026 market data is beginning to expose it.
Not every software ETF is suffering equally. The Roundhill Generative AI & Technology ETF (CHAT) gained 60% year-to-date and 133% over one year through mid-2026 β a direct refutation of the idea that software vehicles are categorically impaired. The divergence inside the software category is more instructive than the category average: funds built around AI-native application layers are separating sharply from those tracking legacy enterprise seat-license models.
Chart: AI-related ETF year-to-date performance through mid-June 2026. CHAT and AIS (AI-native mandates) diverge sharply from legacy software ETFs WCLD and IGV.
The specific trigger for early-2026 software pain is worth naming precisely. The software sector shed approximately 31% in the period following Anthropic's release of open-source enterprise AI plugins, which sparked fears that commoditized AI functionality would compress traditional software margins across the board. But a sector-wide selloff driven by a single disruptive event typically overshoots. That drawdown painted legacy ERP (enterprise resource planning) vendors and AI-forward productivity software companies with the same brush β which is where the analysis starts to break down. IGV and WCLD track software companies with legacy seat-license revenue models. CHAT and AIS are built around companies actively monetizing AI at the application layer. Calling both groups "software ETFs" is like calling a horse and a tractor the same thing because they both work on farms.
AI-focused ETF inflows overall jumped to $19 billion in 2025 from $4.2 billion in 2024 β a 352% increase β suggesting the market is rapidly expanding its definition of what qualifies as an AI investment vehicle. That capital doesn't all flow into chips. Some of it is already staging for the next rotation.
The Application Layer Bet β and Who's Making It
JPMorgan's equity research team moved to name specific "AI software underdogs" in May 2026, identifying companies the desk believes are materially mispriced. ZoomInfo (GTM) carries a price target implying 214% upside; HubSpot (HUBS) 115%; Figma (FIG) 91%. The common thread, per JPMorgan's analysts: these companies are "integrating AI copilots, automated customer engagement tools and workflow intelligence directly into existing software ecosystems" β rather than simply issuing AI feature press releases and hoping the market notices.
That distinction matters for anyone evaluating workflow automation tools or CRM (customer relationship management) platforms for their team. A sales intelligence platform that has rebuilt its lead-scoring engine around AI is a different product than one that added a chatbot to its sidebar and called it an AI transformation. The former changes the unit economics of a sales team. The latter is a demo. The demo is not the product β and JPMorgan's list reflects a bet that investors have not yet separated the two cohorts inside the software sector.
Industry analysts cited in the TradingView coverage describe 2026 as "a transition year where the focus is shifting from pure infrastructure to AI-enabled revenue models, with software and services firms beginning to prove their worth by delivering tangible productivity gains." That framing tracks with the underlying economics: hyperscalers spent the last two years building the compute foundation; enterprise software companies now access that infrastructure via API (a way for two apps to talk to each other) at falling marginal costs. The question is execution β specifically, which companies have rebuilt their revenue model around AI outputs rather than around AI announcements.
As noted in Smart AI Trends' coverage of the Anthropic export control story, regulatory and licensing disruptions in the AI stack can trigger rapid sector rotations β which is precisely what software ETF holders experienced in early 2026. The lesson is not that software is permanently impaired; it's that externally triggered selloffs create dislocations that active analysis can exploit.
The global AI market is projected to reach $4.8 trillion by 2033 from $189 billion in 2023. That growth does not terminate at the GPU (graphics processing unit) layer. Compute infrastructure enables applications; applications generate revenue; revenue justifies the next round of capex. The $660β690 billion hyperscalers are spending in 2026 eventually needs to show up in enterprise software productivity metrics, not just in chip shipment numbers.
A Better Frame: Three Ways to Actually Use This
The job-to-be-done here is not simply picking the right ticker. It is understanding which part of the AI value chain is currently underpriced relative to its likely contribution to enterprise revenue over the next 18β36 months. Three practical frames for working through it:
IGV's -20.62% YTD tells you how legacy enterprise software positions performed under a specific AI disruption event β not how AI-native software positions are performing. Before drawing conclusions from any software ETF's recent return, read its actual top-10 holdings. CHAT at +60% YTD and IGV at -20.62% are both labeled "software ETFs." They are categorically different bets. The mandate determines the exposure; the trailing return describes what happened to that specific mandate under specific 2026 conditions. Treating them as interchangeable is the primary analytical error driving premature sector dismissals.
As of June 17, 2026, hyperscalers have committed $660β690 billion in aggregate 2026 capex, with roughly $450 billion-plus earmarked for AI infrastructure. That compute eventually needs to generate enterprise revenue β and the route to revenue runs through application software and business tools, not raw chip throughput. The lag between infrastructure investment and software monetization has historically been 18β36 months. The early-2026 software selloff may be pricing in permanent disruption when the actual dynamic is a temporary monetization gap. These are very different investment theses with very different time horizons.
Any software company can issue an AI roadmap. The filter that matters is whether AI has changed the company's core revenue metric β net revenue retention, seat expansion rate, or pricing power. ZoomInfo, HubSpot, and Figma made JPMorgan's underdogs list precisely because analysts identified specific AI-driven changes in product economics, not just feature announcements. Apply that same scrutiny to any ETF's top 10 before treating a compressed price-to-earnings ratio (the multiple investors pay per dollar of company earnings) as an automatic buying signal. Low price alone is not a thesis. A disrupted business model trading at a low price is a value trap.
Frequently Asked Questions
What are the best software ETFs for AI investment right now?
As of June 17, 2026, the Roundhill Generative AI & Technology ETF (CHAT) has shown the strongest performance among software-oriented AI ETFs, up 60% year-to-date and 133% over one year through mid-2026. Traditional software ETFs like IGV (-20.62% YTD) and WCLD (-20.28% YTD) have underperformed significantly. The key variable is mandate: ETFs built around AI-native application companies are behaving very differently from those tracking legacy seat-license enterprise software models. There is no single "best" software ETF for AI exposure without first defining which part of the software stack you are trying to own.
Why are software ETFs like IGV underperforming semiconductor ETFs so dramatically in 2026?
Two forces converged. First, hyperscalers committed $660β690 billion in 2026 capex concentrated in compute infrastructure, directly benefiting chip and hardware ETFs. Second, Anthropic's release of open-source enterprise AI plugins in early 2026 triggered a sector-wide drawdown of approximately 31% in software stocks β Bloomberg Intelligence dubbed the resulting fear the "SaaSpocalypse" β as markets priced in the possibility that seat-based SaaS licensing models could face structural compression. The single-day divergence of 8.4 percentage points between chipmakers and software stocks on June 2, 2026 was the largest such gap since mid-2021. Whether the selloff correctly prices a permanent structural shift or an overreaction to a transitional disruption event is the central debate inside the software ETF category right now.
Should I invest in software or semiconductor ETFs for AI exposure going forward?
The binary framing may be the wrong one. Semiconductor and hardware ETFs have captured the infrastructure phase of the AI buildout effectively. The next leg β if industry analysts are correct that 2026 is a transition year toward AI-enabled revenue models β may favor application-layer software companies that have demonstrably embedded AI into their core workflow automation and productivity software products. JPMorgan's May 2026 research identified ZoomInfo (GTM), HubSpot (HUBS), and Figma (FIG) with upside targets of 214%, 115%, and 91% respectively. A staged allocation approach β maintaining hardware exposure while selectively adding AI-native software ETFs with clearly defined mandates β is how institutional investors appear to be positioning for the next cycle.
How much are hyperscalers actually spending on AI infrastructure in 2026, and does it matter for software stocks?
As of June 17, 2026, the five major hyperscalers have committed a combined $660β690 billion in capital expenditures for the year β roughly double 2025 levels β with approximately 75% directed at AI infrastructure. Amazon leads with $200 billion in planned capex, followed by Alphabet at $175β185 billion, Meta at up to $135 billion, and Microsoft at $120 billion-plus. This matters for software stocks in a lagged way: the compute infrastructure being built in 2026 creates the API access and reduced inference costs that will allow software companies to rebuild their products around AI outputs in 2027 and beyond. The $690 billion spent on infrastructure is not a competitor to software value β it is a prerequisite for the next generation of productivity software products to exist.
Bottom line: In my analysis, the early-2026 software ETF crash is a genuine signal wrapped inside a significant market overreaction. The hardware-first narrative is correct for the infrastructure phase β but infrastructure phases create application phases, and $660β690 billion in AI compute spending eventually needs software to convert it into enterprise revenue. I'd argue the productive question for investors is not "hardware or software?" but rather "which software companies have actually rebuilt their revenue model around AI, and which are just demo-ware waiting to be disrupted?" The 8.4 percentage-point single-day divergence Bloomberg reported on June 2, 2026 is a data point about where we are in the cycle. It is not a permanent verdict on an entire asset category.
Disclaimer: This article is editorial commentary for informational and educational purposes only and does not constitute financial or investment advice. ETF performance figures, analyst price targets, and capital expenditure projections are time-sensitive and subject to rapid change. Always verify current holdings, prospectus details, and risk factors through official fund documents and regulatory filings before making any investment decisions. Research based on publicly available sources current as of June 17, 2026.