Anthropic just launched tools to show companies what their AI spending is buying

Anthropic reported an operating profit in Q2 and 900% net revenue retention, with spending by the same customers growing ninefold from Q3 2025 to Q3 2026. Its new enterprise reporting tools and pooled budgets aim to help customers measure AI-tool ROI and manage token costs. Separately, Jellyfish found agentic coding-tool spending had risen 10x since the start of the year among its surveyed customers, reaching about 2% of engineering costs; Anthropic’s adoption has so far remained resilient despite competition from OpenAI’s Astra.

Analysis

The strategic value of the new reporting is less about adding features than controlling the enterprise buying cycle: making token use legible can help customers defend budgets, while workflow recommendations and accumulated process data raise the cost of switching. That creates a useful flywheel for ANT—better visibility can support renewal even if it also prompts customers to use fewer tokens. The key uncertainty is whether efficiency gains expand deployment enough to offset lower spend per task.

Near term, the reported cohort expansion and operating profitability could reinforce a premium narrative, but neither establishes durable unit economics. Verify the cohort definition and concentration, whether expansion reflects recurring production usage versus pilots or a small number of large accounts, and whether reported profitability is repeatable. The Jellyfish sample is evidence of usage, not proof of broad Fortune 500 ROI.

Over 1–3 months, watch for customer case studies with independently measurable productivity outcomes, enterprise renewal/expansion data, and any evidence that reporting reduces token intensity. A faster-than-expected shift from experimentation to budget optimization would be a mixed signal: stronger retention, but potentially slower consumption growth. Over 6–18 months, workflow integration and embedded customer support may matter more than model benchmarks; OpenAI can pressure pricing or improve its enterprise deployment motion, but model quality alone may not dislodge installed workflows. Contrarian risk: investors may treat extraordinary spend expansion as a durable growth rate when it may instead trigger procurement controls and competition.

Key Decisions for Investors

• Treat ANT as a conditional long on pullbacks, not a momentum chase: the thesis is enterprise workflow lock-in, not extrapolation of the reported cohort growth. Reassess after the next disclosed financials or credible enterprise usage data; do not size up without cohort concentration and recurring-usage clarity.
• Set an alert for evidence that customers’ token cost per accepted code change is falling while renewal or deployment breadth rises. That would support the efficiency-to-retention flywheel; falling token intensity alongside stalled seat/team expansion would falsify the growth case.
• Monitor OpenAI’s enterprise adoption and pricing, and ANT’s renewal/expansion disclosures. A material slowdown in customer expansion, discounting to retain accounts, or a rival displacing standardized workflows would weaken the relative-long thesis; avoid assigning a valuation premium solely from survey usage share.
• Treat the 900% expansion figure and single-quarter operating profit as verification items: establish the cohort period, customer concentration, revenue-recognition basis, and whether profitability persists before translating them into forward estimates.