AI Pilots to Production
Jesse Venticinque@jesse_
−2.4%
since Oct 3
The idea
I think AI models are outpacing enterprises' ability to actually deploy them, so the real value shifts to companies that solve the messy middle: cleaning and connecting enterprise data, orchestrating workflows across legacy systems, watching what models do once live, and the consulting, security and governance work that lets a company sign off on putting this into production. The portfolio spreads across orchestration, security, integration and consulting, data infrastructure, and observability, with no single name dominating and a small cash buffer.
What it owns
- These platforms sit inside the business processes where enterprise AI must actually run and deliver recurring revenue when models move from pilots to production work. · ServiceNowIt already sits in the workflows where enterprise AI must run, and sells AI as a paid tier on top of ticketing, approvals and operations, so its AI bookings are a clean read on whether production deployments are paying., PalantirIt sends engineers in to wire models into an enterprise's real systems and data, and its US commercial revenue growth is the live test of whether companies will pay for that deployment at scale., SAP SEIt provides the cloud-based ERP infrastructure where enterprises modernize legacy systems to support AI; recurring cloud revenue grows as that migration backlog moves forward., PegasystemsIt automates the core business processes where enterprises want AI doing real work; it already holds enterprise distribution and recurring revenue., UiPathIts robots already touch the legacy systems enterprises cannot replace, and it is betting the agentic turn makes that footprint more valuable by extending automation deeper into legacy workflows.25%
- No regulated enterprise moves AI into production until identity, data access and governance are solved, making this layer a gatekeeper that collects recurring fees on that sign-off. · Palo Alto NetworksNo chief information security officer signs off on production AI without a security platform underneath it, and it is consolidating that spend into recurring bills including products for securing AI itself., CrowdStrikeIt sells recurring security modules that enterprises add as their AI footprint expands, deployed on a platform they already operate., OktaEvery AI agent acting inside an enterprise needs an identity and permissions, and it is the independent identity layer positioned to issue them as agentic deployments scale., ZscalerIt controls what every user, app and AI agent is allowed to reach, which is exactly the control a chief information security officer demands before AI touches production systems., VaronisIt governs data access permissions on a mostly recurring SaaS model, which enterprises must clean up before deploying AI models across their files.23%
- Legacy-system modernization is labor booked as revenue quarters in advance, making these firms the earliest visible signal of whether enterprise AI budgets are moving from pilots to production. · AccentureIt books modernization work quarters in advance, and its AI-related bookings are the cleanest single read on whether enterprise AI budgets are shifting from pilots into production at scale., IBMIt combines consulting labor for legacy modernization with software that governs AI inside regulated enterprises, earning backlog-driven revenue on both as pilots convert to production., EPAM SystemsIt is the engineering-heavy integrator that writes the code wiring AI into enterprise systems, and its labor-intensive approach succeeds if enterprise demand for custom AI build-out keeps recovery on track.17%
- Production AI systems depend on clean, queryable data, and these platforms already hold enterprise data under contract. · SnowflakeIt houses enterprise data already under contract and bills for consumption; production AI workloads querying it daily would increase usage-based revenue., MongoDBIt is the database underneath the new AI applications enterprises are shipping, so moving from pilot to production means more Atlas consumption from companies building on their own data., ElasticProduction AI systems need to retrieve the right company data at the moment of query, and it sells that search and retrieval layer to enterprises that already run it in mission-critical places.16%
- Once AI systems run in production, someone must watch costs, errors and behavior, and that monitoring spend scales with deployment rather than pilots. · DatadogIt monitors everything enterprises run including production AI systems, and expanded workloads generate more hosts, logs, and platform spend from existing customers., DynatraceIt serves the large legacy enterprises slowest to absorb AI, and its recent acquisition of Arize puts it directly into the production-AI observability layer as long as enterprises consolidate monitoring spend with it.12%
- Cash8%