AI-security isn’t a “whole stack” spend—identity, endpoint, and agent governance each claim a different budget line

When investors price “cyber beats,” the underlying question is not whether companies buy security—it’s which budget line the buyer is actually moving.

In AI-era security, that budget routing tends to split into three practical lanes:

• Identity + account takeover control (who can access AI and enterprise systems)
• Endpoint and cloud-native threat prevention (where detections turn into real-time block actions)
• Security platform governance for AI systems and agent runtime (how enterprises keep AI outputs, code, and agents from becoming the next threat channel)

This article maps the value chain using what large vendors themselves disclosed in filings—then connects that to what those disclosures imply for the next 1–3 quarters of demand and the next 1–3 years of platform lock-in.

To ground the AI-security value chain in verifiable disclosures, the core facts used here come from:

• Okta reporting forward contract metrics and ongoing scale in identity security.
• CrowdStrike describing an AI-native platform at the center of its “Security Cloud,” alongside acquisitions that expand identity security and runtime security.
• Palo Alto Networks laying out revenue mix (product vs. subscription) and explicitly describing AI security platform modules that support AI ecosystem protection.
• Microsoft providing governance and risk-control context for AI and cybersecurity, which matters because enterprise buyers treat hyperscalers as platform constraints for how security products integrate.
• Zscaler supporting the idea that zero-trust network access is part of the same buying motion, even if it’s not the highest-growth line by itself.

Identity is where most AI-security failures start: compromised credentials grant access to AI apps, internal tools, and data workflows.

Okta’s latest SEC reporting shows a scale-and-contract profile consistent with a renewal-heavy identity spend:

• Okta reported current remaining performance obligations of $2,585 million and total remaining performance obligations of $4,858 million (as of July 31, 2026).
• Okta also reported dollar-based net retention of 107% (trailing 12 months) and 5,255 customers with ACV above $100,000 (as of July 31, 2026).

Those metrics matter for the “who gets paid” question because identity vendors are typically embedded in access lifecycle workflows. When enterprises adopt AI tooling, they often standardize authentication and authorization at the same time—locking Okta into the identity layer budget rather than letting it be replaceable commodity IAM.

CrowdStrike’s model is built to convert AI assistance into action at scale: it emphasizes an AI-native “Security Cloud” that can automatically prevent threats.

In its SEC reporting, CrowdStrike described its AI-native Falcon platform and the way it uses cloud-scale intelligence to enrich and correlate signals—explicitly framing its platform as operating across a massive event volume.

The supply-chain angle matters: endpoint and cloud threat prevention vendors sit downstream of identity because they are the enforcement layer once an attacker has authenticated. But they also sit upstream of agent security because agents operate on endpoints and in runtime environments—so successful agent governance often depends on having a strong detection/prevention base.

Agent governance is not just policy—it’s operational security for how AI models, code, and agents behave in production.

Palo Alto Networks is explicit about tying AI security to platform modules. In its SEC filing, it described an “AI security platform” and listed modules such as AI model security, AI posture management, AI red teaming, AI runtime security, and AI agent security.

Two additional “who gets paid” signals show up in the same filing:

• It discloses a revenue mix where subscription and support is the dominant share (subscription/support is typically where continuous updates, new AI modules, and ongoing enforcement live).
• It reports a large remaining performance obligations number used as a demand proxy.

Taken together, it supports a thesis that AI security budgets are being structured like platform subscriptions—not one-off tools. That structure usually favors incumbents with installed bases and an integrated path from telemetry to response.

Microsoft matters in an AI-security value chain even when it doesn’t “win” the security control directly. Enterprise security teams treat hyperscaler governance and security development lifecycle requirements as constraints that determine which third-party security products are easiest to deploy and integrate.

In its SEC cybersecurity disclosures, Microsoft describes a governance framework (including how the board and senior management review cybersecurity risk), risk management processes for third parties, and how it uses security products and AI models for defense against threats.

This isn’t a vendor pitch; it’s a buying reality. When Microsoft’s platform changes the defaults (secure development lifecycle, vulnerability monitoring cadence, and third-party risk requirements), security tool selection tends to follow what integrates cleanly into that governance surface.

Zero trust network access and SASE architectures determine how fast authenticated sessions reach internal systems—and how consistently policy can be applied to traffic.

Even if the biggest AI-security margin comes from detection/prevention and governance modules, network control layers still control the deployment friction. That’s why identity and endpoint vendors increasingly need consistent network policy enforcement to deliver outcomes the AI promises.

In other words, SASE/zero trust is less often the “hero” in AI-security earnings, but it can be the “gate” for whether AI-security tooling is deployed broadly or stalled by rollout complexity.

That framework is what lets you turn “the cyber rally priced the beats” into a practical question: which layer’s disclosures show forward demand that can plausibly carry into the next quarters?

Short-term (days to quarters):

• Buyers react fastest to layers that reduce incident probability immediately—identity takeover controls and endpoint/cloud prevention.
• Filings that disclose forward-demand visibility (like remaining performance obligations) tend to confirm that reaction.

Long-term (1–3 years):

• The AI-security stack shifts toward runtime governance for models and agents.
• Vendors that explicitly enumerate AI runtime and agent-security modules (and monetize them through subscription structures) tend to compound via platform lock-in.

The investable conclusion from the filings used here is that the AI-security market is not simply “more tools.” It is re-bundling.

• Identity vendors like Okta monetize the access layer with renewal-heavy contract pipelines.
• Endpoint/cloud enforcement vendors like CrowdStrike monetize the action layer where AI converts signals into blocks.
• Platform security vendors like Palo Alto Networks monetize the governance layer for AI models, runtime, and agent security via subscription and large forward contract backlogs.

If you’re mapping “who gets paid,” follow the modules that match the operational attack surface: access (identity), execution (endpoint/runtime), and decision-making (agent/model governance).