Artificial intelligence is transforming software markets, but it’s still early in. Investors are rewarding the companies that sell the data and compute plumbing behind AI, even as many newly minted application-focused vendors struggle to turn new features into meaningful revenue. The consequence is a growing performance gap at the software level.
Infrastructure plays are now leading the rally
Data and cloud platforms are closest to AI spend, and it shows. Snowflake, the cloud-native data warehouse powering analytics and model training pipelines, is up close to 43% YTD and nearly 96% over twelve months as investors clearly believe data gravity will be the on-ramp to generative AI monetization.
This surge of interest has also helped Oracle as businesses scramble to reserve capacity for AI workloads. After reporting strong growth in its backlog of remaining performance obligations, it also saw a bolstered pipeline of forward revenues, helped along by a multiyear compute deal with OpenAI that The Wall Street Journal disclosed earlier. The takeaway: AI use cases are consumption-heavy and are sticking to platforms that sell them scalable storage, networking, and databases.
Analysts echo the shift. Gartner’s Arun Chandrasekaran notes that adoption of new-age tech like AI starts with data and compute, pushing spend toward infrastructure and platform-as-a-service layers that can ingest, govern, and operationalize massive datasets. In other words, before AI assistants can generate a single sentence, a company needs to construct a secure, high-quality data backbone.
AI is still hard to monetize for app vendors
In contrast, a number of front-end software companies have increased generative features but have not significantly moved the revenue needle. A Morgan Stanley basket tracking the largest SaaS names is lower by more than 6% this year, while bellwethers have underperformed: Salesforce shares are off 28% year to date, and Adobe is down about 21%, despite heavily marketed AI rollouts.
Why the disconnect? First, AI assistants layered onto incumbent workflows are more frequently packaged or discounted, dampening the share of new annual contract value. Second, much of the pricing experimentation is still incremental—often tens of dollars per user per month—and it’s not clear whether that is enough to cover the cost of inference at scale. Lastly, customers want hard ROI and proof points, which has extended sales cycles as CFOs favor measurable productivity gains over hype.
There is also a business model wrinkle. The traditional seat-based licensing model does not align with usage-based AI economics. Vendors who can meter data access, vector search, and inference calls — think data platforms and observability tooling — are structurally better situated to capture growing consumption, while horizontal SaaS suites face tougher decisions around feature bundling and margin protection.
What investors are watching now in AI software
Two metrics loom larger than the rest:
- Capacity signals: queues and remaining performance obligations as proxies for AI infrastructure demand.
- Consumption: data transfer, query volumes, and workload mix to gauge whether AI use cases are sticky and scaling — not just trial splash.
Gross margin is another tell. AI features that rely on expensive inference models can squeeze margins unless vendors optimize their models, cache results, or lean more heavily on retrieval over proprietary data. That means model routing, vector databases, and tight data governance — capabilities that are more natural for infrastructure players than classic application suites.
Moats matter, too. Vendors that control unique data — transaction logs, customer interactions, telemetry — can train domain-specific copilots that are much more expensive to switch away from. Those without differentiated data access run the risk of developing similar experiences on top of the same foundational models, constraining pricing power.
Rotation signals and risks in the AI software trade
The market is quietly turning toward AI plumbing providers. Alongside the hardware leaders, investors are also flocking to cloud databases, data orchestration, and security and observability platforms that sit in the critical path of AI pipelines. Analyst forecasts from large banks call for hundreds of billions to be spent on AI capital in the years ahead, underscoring that infrastructure is where budgets begin.
That’s not to say application vendors are out of the race. Service management, CRM, design, and marketing clouds could potentially reaccelerate if AI copilots can show they reduce ticket volumes, increase conversions, or compress cycle times in measured pilots. As Constellation Research’s R. “Ray” Wang has posited, a new cycle could favor app vendors that ship agentic workflows and measure productivity gains at scale.
Bottom line for winners and losers in AI software
Winners today are the platforms closest to data and compute — Snowflake and Oracle are clear beneficiaries, with adjacent categories like observability and security also well positioned. Laggards are suites whose add-ons have AI sizzle but no easy monetization story, as seen in both broader SaaS underperformance and specific declines at Salesforce and Adobe.
The key to closing this gap is simple but painful:
- Demonstrate lasting ROI.
- Price AI where value is earned.
- Align go-to-market with usage-based economics.
Until then, markets will continue to favor the companies that make AI possible over those just making it visible.