Why Block’s COO is tracking ‘gross profit per employee’—and how AI is on track to double it to $2 million

Block cut ~4,000 jobs and projects gross profit growth of 18% and profits up 54% in 2026; management says this would raise gross profit per employee to about $2.0M (from $1.0M in 2025, $750k in 2024, and $500k in 2019). Management attributes the productivity gains to its internal AI agent 'Goose' (18 months in production), citing a 40% increase in developer productivity per engineer since September and much faster model build times. Leadership frames the layoffs as part of a deliberate two-year transformation to retool workflows, not as pandemic-era bloat.

Analysis

Embedding purpose-built AI agents into end-to-end workflows is intrinsically a productivity multiplier, but the value accrues only to firms that (a) own the right data, (b) can reengineer processes around the agent, and (c) fund the incremental compute and tooling. Expect realized margin and unit-economics improvements to lag initial AI deployments by 6–24 months while companies redesign approvals, controls, and incentives; short-term headcount reductions are a signal of reorganization, not the whole story.

Second-order winners are providers of the underlying stack (GPU/accelerators, hyperscale cloud, and MLOps tooling) and software vendors that can productize AI as embedded features; winners will compound revenue per customer via higher wallet share rather than linear headcount leverage. Conversely, labor-intermediation businesses and owners of redundant office capacity are exposed to structural demand erosion as firms trade recurring payroll expense for fixed tech spend and remote/async workflows.

Key risks that could reverse the trend include higher-than-expected variable compute costs, model governance/regulatory blowback, and failure modes that force manual oversight back into the loop—any of which can wipe out projected unit cost wins inside a single fiscal year. Watch near-term catalysts: measurable developer-velocity metrics, compute-run-rate disclosures, incremental margin by product, and targeted M&A for agent/IP; those indicators precede durable earnings upgrades or re-ratings by 3–9 months.

Key Decisions for Investors

• Long NVDA (or buy 6–9 month call exposure) — rationale: direct play on accelerator demand as enterprises embed agentic workflows. Risk/reward: asymmetric upside from continued enterprise AI capex (30–100%+ potential) with downside capped to premium; monitor quarterly compute guidance for signs of saturation.
• Overweight MSFT and AMZN (12 month horizon) — rationale: capture cloud take-rate and MLOps revenue; add on soft-landing pullbacks. Risk/reward: stable cash flows and cross-sell offset execution risk; hedge with a small short of MAN (Manpower) to reduce macro beta, since staffing revenue is most exposed to structural job repricing.
• Pair trade: long NVDA / short MAN (3–12 months) — rationale: long accelerator demand vs short labor-intermediation. Risk/reward: pair reduces market exposure; expected divergence if AI-driven headcount reductions accelerate, target asymmetric return of 2:1 if NVDA outperforms MAN by 25% within 12 months.
• Short office/centralized-office REITs (e.g., VNO) over 6–12 months — rationale: office demand and leasing spreads face secular pressure as companies optimize for fixed-tech spend. Risk/reward: high carry and clear fundamental catalyst (renewal spreads); principal risk is local market rebounding or office reuse policies reversing, size position accordingly.