Agentic AI And The Next Layer Of The Trade: Beyond Nvidia

Jensen Huang called it “the agentic AI inflection point” on NVIDIA’s most recent earnings call, and for once the hyperbole may be justifiable. The first chapter of the AI investment cycle — the infrastructure buildout of GPU clusters, data centers, and networking fabric that drove NVIDIA’s stock up several hundred percent and established the semiconductor complex as one of the decade’s defining trades — is not over. But a second chapter is clearly beginning, and the investment logic in that next chapter is meaningfully different from the one that made infrastructure the obvious place to be in 2023 and 2024.

Agentic AI refers to AI systems capable of autonomous multi-step task execution — systems that can plan, reason, take actions in software environments, and adapt to new information without requiring a human to direct each step. These are not chatbots. They are software agents that can browse the internet, execute code, interface with APIs, manage workflows, and complete complex tasks that previously required skilled human labor. Morgan Stanley’s Jeff McMillan, speaking at a 2026 internal research session, described the Silicon Valley landscape: “seemingly overnight, every firm has become an agentic one.” He was quick to add that much of this is aspirational — but the direction is clear, and the investment implications are beginning to show up in revenue numbers.

Why the Application Layer Is Now the Story

The infrastructure layer of the AI stack has been extraordinarily well-compensated. NVIDIA has captured approximately 90% of AI accelerator spend — roughly $180 billion annually at current run rates. Data center operators, power infrastructure providers, and networking companies have all seen demand surge. This will continue: PineBridge and MetLife have described data center equipment growth as “essentially locked in for the next four to five years” with annual growth near 25%. The structural demand for compute is not in question.

What is now in question — and what the market is increasingly focused on — is whether the application layer can generate the revenue needed to justify the infrastructure investment. Sequoia’s $600 billion revenue gap analysis captures this anxiety in a single number. The answer to that gap, if it comes, will come primarily from enterprise software: AI-native applications, AI-augmented workflows, and agentic systems that demonstrably reduce labor costs or increase revenue for the businesses deploying them. The companies building and distributing these applications are the logical next beneficiaries of the AI investment cycle.

Where the Revenue Is Starting to Show Up

The evidence of enterprise monetization is early but real, and it is concentrated in a few specific categories. Cybersecurity is arguably the most mature AI monetization story outside of the hyperscalers themselves. AI-powered threat detection, automated incident response, and identity security systems have moved from pilot to production at major enterprises at a pace that is showing up in revenue figures. CrowdStrike, Palo Alto Networks, and SentinelOne have all reported acceleration in AI-driven product adoption, with annual recurring revenue growth that reflects customers willing to pay for measurable security improvements.

Workflow automation and enterprise software more broadly are the second wave. Microsoft’s Copilot products — embedded AI capabilities in Office, Teams, and Azure — crossed a $37 billion annual revenue run rate in Q1 2026, up 123% year-over-year. That growth is coming from enterprises paying premium prices for productivity tools that demonstrably improve output. ServiceNow, Salesforce, and several other enterprise software platforms have reported similar dynamics: customers are deploying AI workflows, measuring the output improvement, and expanding their deployments based on measurable results. This is different from the AI pilot phase of 2023-2024, when proof-of-concept spending dominated.

The Agentic Opportunity Specifically

Agentic AI is the next level of complexity above co-pilot or assistant AI, and its economic value proposition is proportionally larger. A co-pilot that helps a knowledge worker write better emails or summarize documents improves individual productivity at the margin. An agent that manages a complex multi-step workflow — analyzing a customer account, identifying an issue, generating a resolution plan, drafting the communications, executing the recommended actions, and logging the outcome — replaces or dramatically augments an entire category of knowledge work.

The verticals where agentic AI is creating the most immediate economic value include financial services (compliance monitoring, trade reconciliation, customer onboarding), healthcare (clinical documentation, prior authorization, revenue cycle management), and software development (autonomous code generation, testing, and deployment pipelines). In each of these verticals, the labor cost savings are significant enough that enterprises are willing to pay meaningful software subscription fees for provably capable agents — a dynamic that is completely different from the $30-40 billion in corporate AI spending that MIT’s Project NANDA found produced zero measurable P&L impact in 2025. Agentic AI, when it works, produces P&L impact that is easy to measure and hard to argue with.

Investment Implications

The investment thesis for the agentic AI layer is more selective than the infrastructure thesis was. Nvidia was a near-perfect vehicle for the infrastructure buildout because it captured 90% of the relevant spending with a dominant product and a near-impenetrable competitive moat. The application layer is more fragmented, with dozens of companies competing for enterprise deployment at different points in the stack.

The highest-conviction positions in this layer tend to have three characteristics: distribution moats (existing enterprise relationships through which AI products can be deployed at scale), vertical specificity (deep domain expertise in the workflow being automated, making the product harder to replicate with a general-purpose agent), and demonstrated revenue momentum (not pilot programs — actual ARR growth from paying enterprise customers). Companies that check all three boxes are not necessarily the largest or most familiar names. Some of the most compelling agentic AI investment opportunities are in mid-cap enterprise software, vertical SaaS, and cybersecurity — companies that are not yet commanding Nvidia-scale valuations but are showing the early revenue characteristics that indicate they have found genuine product-market fit in the agentic layer.

The infrastructure trade will continue to reward patient holders. But the risk-adjusted opportunity in the next 18-24 months increasingly sits in the companies converting that infrastructure into revenue. Finding them before the market fully prices them is the challenge — and the opportunity — that defines the second chapter of the AI investment cycle.