Market moves feel gradual — until they don't. Over the past quarter a quiet rotation has taken shape: the capital that chased Nvidia through the AI hardware frenzy is starting to flow into AI software and cloud AI services. That does not mean the GPU leader is finished. Far from it. But where investors now find the most compelling growth and margin stories has shifted.
Why now
- Supply is easing. New chips from rivals and expanded foundry capacity have softened the shortages that once justified sky-high hardware multiples. When scarcity fades, the obvious upside narrows.
- Clouds are making accelerators an on-demand utility. If you can spin up inference and training capacity from a major provider, fewer companies feel pushed to buy racks of cards. Budgets migrate.
- Recurring revenue is suddenly more comforting. Subscriptions and managed services from AI software firms look steadier than one-off GPU cycles to a lot of investors.
Signals worth watching
- Hardware valuations have compressed after big runs, while certain software and platform stocks are outperforming on signs that enterprises are actually adopting and paying.
- More M&A noise aimed at niche AI software vendors with sticky contracts — buyers are looking for margin leverage and predictable cash flow.
- Cloud providers bundling inference, deployment toolchains, and orchestration, which lowers the friction for third-party software to be adopted and monetized.
Concrete examples
- Firms that wrap models, pipelines, and compliance for regulated sectors are drawing real interest. They solve integration and uptime, not just compute — customers pay for those guarantees.
- Companies that stockpiled GPUs in 2023–24 are now redirecting spend toward model governance, labeling, and monitoring once their infrastructure is in place. Makes sense; the problem shifts from compute to control and quality.
Counterpoints and risks
- GPUs still define the cutting edge. For top-tier training and pushing model performance, hardware remains essential, and companies that control the full stack will keep commanding premiums.
- Widespread open-source model adoption could put pressure on software pricing over time. What looks like sticky revenue now might face tougher competition later.
What this implies for investors
- If you own hardware-heavy names, ask whether future growth depends on continued scarcity or on driving enterprise adoption that resembles subscription economics.
- For software bets, the focus should be customers, retention, and gross margins. Rapid ARR growth without retention is noise, not a business.
Short-term ideas to consider
- Tilt toward platforms that actually connect models to mission-critical workflows rather than pure-play experimentation toolkits.
- Track cloud partnership announcements and contract renewals — they’re early clues that budgets are moving from capex to opex.
This is not a zero-sum handoff so much as different parts of the stack moving through maturity at different speeds. Hardware bought the marquee returns; software and services are now asking for the recurring dollars that keep the engine running.
Pedro Marini
