Enterprise AI spending is surging to nearly $700 billion in 2026. But look at where the money is going, and the picture gets complicated. Hardware and infrastructure are capturing most of the investment while enterprise software stocks have cratered more than 20% this year.
The market is voting with its wallet: physical compute infrastructure matters more than the software layers built on top. For now.
The Numbers Tell the Story
The iShares Expanded Tech-Software Sector ETF has tumbled over 20% in early 2026. Median SaaS multiples compressed from over 10x revenue to roughly 5x. Software sector earnings expectations turned negative for the first time since 2022.
Meanwhile, hardware plays are thriving. Micron Technology is reaching all-time highs. The four hyperscalers — Amazon, Google, Microsoft, and Meta — are spending close to $700 billion on infrastructure this year, with roughly 75% ($450 billion) going directly to AI infrastructure: servers, GPUs, and data centers.
Oracle’s March 10 earnings showed 14% revenue growth but sent the stock lower. Palantir continues struggling despite being positioned as an AI leader. The market is skeptical that software companies can capture AI value proportional to their investments.
Why Hardware Is Winning
Three dynamics are driving the bifurcation:
Physical scarcity: You can’t spin up GPUs on demand the same way you can deploy software. NVIDIA faces a 40% production cut from memory shortages. AMD and Intel have warned customers of six-month lead times for CPUs. The companies controlling physical compute have pricing power that software companies don’t.
Capital intensity: Training and running AI models requires massive upfront infrastructure investment. Companies are building that capacity before they figure out which software will run on it. Hardware spending happens first; software spending comes later.
Uncertainty about software winners: No one knows which AI software companies will dominate. OpenAI has 78% of enterprise CIOs using their models, but Anthropic and Google are gaining share rapidly. Enterprises are hedging their software bets while committing to infrastructure they know they’ll need regardless of which models win.
The Seat-Based Pricing Problem
Legacy software pricing is colliding with AI efficiency gains. Enterprise software companies built their businesses charging per user seat — more employees meant more revenue.
AI agents threaten that model. If AI handles 80% of customer interactions without human involvement, you need fewer customer service seats. If AI agents process invoices automatically, you need fewer finance users. The same efficiency gains that make AI valuable to enterprises make seat-based software licenses less valuable.
Software companies are scrambling to shift toward usage-based or outcome-based pricing, but the transition is messy. Investors aren’t sure who will manage it successfully.
Consolidation Accelerates
The bifurcation is driving M&A. Venture capitalists surveyed by TechCrunch predicted 2026 would be the year enterprises consolidate AI spending to fewer vendors. That’s happening.
IBM closed its $11 billion Confluent acquisition. Zendesk bought Forethought. Salesforce, Oracle, and Microsoft are all acquiring AI capabilities.
The pattern is clear: established infrastructure players are absorbing AI software companies rather than letting them compete independently. M&A deal volume in mid-market enterprise software is expected to increase 30-40% year over year, with total deal value potentially reaching $600 billion.
For AI software startups, the exit path is increasingly acquisition rather than IPO. The window for independent survival is narrowing.
Who Wins, Who Loses
Winners:
- Hyperscalers controlling physical infrastructure
- Memory and chip manufacturers (Micron, NVIDIA, AMD)
- Enterprise platforms assembling end-to-end AI stacks through acquisition
- AI-native companies with usage-based pricing models
Losers:
- Traditional SaaS companies with per-seat pricing
- Mid-market enterprise software lacking AI integration
- Independent AI startups unable to reach scale before consolidation
- Software investors expecting 2021-era multiples
What Happens Next
The bifurcation won’t last forever. Eventually the infrastructure buildout will slow and attention will shift to software that delivers ROI on all that compute.
But that transition could take years. In the meantime, enterprises are making massive infrastructure commitments while treating software vendors as replaceable.
For software companies, survival means proving AI drives measurable business outcomes — not just efficiency gains, but revenue growth and competitive advantage. The market is no longer impressed by AI features; it wants AI profits.
The companies that can show outcome-based value capture will command premiums. The rest will face compression, consolidation, or irrelevance.