Thesis in one line: the market mood is shifting — from celebrating AI hardware to paying for software that actually turns models into repeatable revenue.
I’ve followed the AI trade for years. The current pattern is familiar: a concentrated rally led by a few chip makers produces huge gains, and then the market starts asking where steady, recurring cash will come from. Think late 1990s tech mania, but with silicon instead of portals.
Why this rotation matters now
- Chips addressed a short-term scarcity. Training and inference capacity ballooned, fast. What didn’t keep up was the set of business models that turn all that compute into predictable cash.
- Software captures recurring revenue, builds data flywheels, and locks in enterprise contracts that are stickier than a one-off hardware refresh.
- Cloud and managed AI stacks increasingly bundle hosting, fine-tuning, monitoring and governance. That packaging tends to favor software vendors and operators who run the service.
Three things I’m watching
- Enterprise AI-as-a-service. Companies that wrap models into workflows CFOs will sign monthly checks for — that’s where recurring revenue shows up. Watch both big cloud incumbents and specialist application vendors.
- Orchestration and MLOps. When deployments scale, customers need observability, versioning, cost controls. The plumbing is dull but indispensable — and profitable once it’s essential.
- Vertical AI plays. Healthcare, legal and finance-focused models are already getting paid premiums because of compliance, specialized data, and higher switching costs.
Counterpoints and risks
- Hardware can bounce back. A new chip generation or a sudden datacenter build cycle could re-accelerate chip names. This rotation is not guaranteed or linear.
- Model commoditization. If foundation models become cheap or widely available, software differentiation and margins could compress.
- Macro pressure. Higher rates and tight capex still hit software multiples; product-market fit matters more than hype in this environment.
A concise watchlist
- NVDA (Nvidia): central to compute — a crown jewel, but priced for it. Think of it as exposure to raw compute, not recurring software revenue.
- MSFT (Microsoft): shifting toward a platform-plus-apps model, with AI features tied into cloud subscriptions and enterprise deals.
- AI (C3.ai): a purer look at attempting software monetization in the AI layer — watch revenue and margin trends closely.
- SMCI (Super Micro): a cheaper way to play datacenter cycles if you want hardware exposure without Nvidia multiples.
- PLTR (Palantir): a reminder that data-heavy, government-anchored contracts can yield durable revenue when executed well.
How I’d position capital (editorial)
- Trim some direct chip exposure after the run and redeploy into software names demonstrating clear ARR growth and case studies that actually prove ROI.
- Keep a small tactical stake in select hardware providers as a macro hedge.
- Favor companies with net retention north of 110% and multi-year enterprise contracts — those are the clearest signs of recurring cash.
A final nudge
This is not advice to sell chips and run for the hills. Compute matters — you can’t build AI without it. What’s changing is the market’s willingness to reward predictability and defensibility. If you own AI stocks, ask whether your position generates recurring cash or just depends on another datacenter boom.
Short version: favor firms that turn AI into subscriptions and workflows; use chips selectively for tactical upside.
