'We have maybe 20 months' to rebuild for AI agents, Meta's infrastructure VP tells VB Transform 2026

Meta VP Barak Yagour says “agentic” demand is rapidly stressing enterprise infrastructure, with agentic queries up 30x in a half and 63% of new Meta dashboards (within 3 months of launch) built via agentic data apps. He also flags governance needs as autonomy rises (“Autonomy without governance is nothing but chaos”), while infrastructure changes target capacity, identity, and velocity (e.g., real-time streaming replacing 24-hour batch ETL). Meta is aiming toward 500M queries/sec and petabyte-per-second training-data reads, reflecting a step-change in how recommendation systems move from correlation to reasoning.

Analysis

This reads more like a spending architecture shift than a headline-driven catalyst. The first-order beneficiary is anyone whose pricing scales with machine consumption rather than human seats: usage-based data infra, query engines, observability, and policy enforcement. The first-order loser is the old “human-in-the-loop” software stack, where seat growth and batch workflows become the wrong unit of account as agents multiply load faster than headcount.

The more interesting second-order effect is that this is not just a compute story; it is an identity and governance story. Every autonomous workflow expands the attack surface, so budget should leak toward authentication, audit, masking, and real-time authorization before it leaks toward speculative agent platforms. That favors names like CRWD and PANW more cleanly than pure AI-app vendors, while SNOW and DDOG should benefit if enterprises are forced into real-time pipelines and telemetry-heavy control planes.

For META, the trade is subtler: better internal data and recommendation tooling can lift ad ROI over 6-18 months, but near term it likely raises infra intensity faster than revenue recognition. Consensus may be missing that hyperscalers can internalize much of the agent stack, which caps upside for third-party software even as it validates the category. The thesis weakens if agent adoption stalls, if regulators tighten data access, or if Meta’s next two quarters show capex/inference costs rising without a matching engagement or monetization inflection.

Key Decisions for Investors

• Long CRWD vs short XLK for 3-6 months: play the rising identity/governance spend tied to agent proliferation; target a 10-15% spread if security budgets re-rate, stop if security billings decelerate or enterprise spending rolls over.
• Buy SNOW on pullbacks for a 1-3 month trade: real-time data and schema-aware storage should see incremental demand as enterprises move from batch ETL to streaming control planes; trim if next quarter consumption growth fails to reaccelerate.
• Initiate a smaller long DDOG position over 2-4 months: agentic traffic increases observability load and makes telemetry less discretionary; invalidation is a guide-down in cloud spend or evidence that customers are consolidating tooling instead of expanding it.
• Keep META as a core hold but only add on weakness, not strength: the upside is 6-18 months out from ad-product and recommendation improvements, while the near-term risk is capex intensity compressing free cash flow; use a pullback-triggered call spread if you want convexity into earnings.
• Do not chase generic 'agent platform' software here; wait for proof of enterprise monetization. If you need exposure, prefer infrastructure and security over app-layer AI names until billings data confirms spend migration.