OpenAI made a quiet announcement that deserves far more attention than it received. The company published its first B2B Signals report , a data product tracking how frontier enterprises actually use AI versus everyone else. The results suggest something that no consulting report or survey has been able to prove empirically until now: the AI productivity gap between leading and lagging firms is not just widening. It is compounding. And once a productivity gap starts compounding, historical precedent suggests it does not stop on its own.
What Actually Happened
OpenAI launched B2B Signals, a recurring publication built on privacy-preserving, aggregated data from enterprise use of OpenAI products. The inaugural report reveals that frontier firms , defined as those at the 95th percentile of AI usage intensity , now use 3.5 times as much intelligence per worker as typical firms. That figure was 2x just twelve months ago. The gap did not just widen , it grew by 75% in a single year. The analysis tracks what OpenAI calls intelligence intensity: a composite of usage volume, task complexity, model tier, and agentic delegation depth across enterprise customer accounts.
The most striking finding is in the attribution. Only 36% of the frontier advantage comes from sending more messages. The remaining 64% comes from richer, more complex AI use , longer, more sophisticated prompts, agentic workflows, multi-step task delegation, and systematic use of advanced tools. Frontier firms are not using AI more often than average firms. They are using it in fundamentally different ways. And the single most differentiating tool in the report is Codex: frontier firms send 16 times as many Codex messages per worker as typical firms. Codex itself has grown 5x since January 2026 and now has 3 million weekly active users, with OpenAI's APIs processing more than 15 billion tokens per minute globally.
Why This Matters More Than People Think
The productivity research debate of the past two years has centered on one frustrating observation: AI adoption has been high, but economy-wide productivity gains have been hard to measure. McKinsey, BCG, and Goldman Sachs have all published surveys showing high AI adoption rates alongside modest aggregate productivity gains. B2B Signals offers a structural explanation for why: the gains are not distributed across all AI users. They are concentrated in a thin stratum of firms that have crossed some threshold of AI integration depth. When the top 5% of enterprise AI users capture most of the productivity advantage through depth of use rather than frequency of use, the aggregate statistics will look flat , right up until the deep-use practices diffuse broadly enough to show up in macro data.
The implication for enterprise strategy is significant and uncomfortable. Being an average AI user , giving employees ChatGPT access, running a few pilots, adding AI assistance to one workflow , does not put a company in the frontier category. What puts a company in the frontier category is systematic delegation: moving from "AI helps workers with tasks" to "workers manage teams of AI agents that do tasks." That is a fundamentally different organizational model, not just a different software configuration. And according to OpenAI's data, the companies that have made that structural shift are already 3.5x more intelligence-intensive per worker and pulling further away every month.
The Competitive Landscape
The Professional, Scientific, and Technical Services sector ranks first in both Codex adoption and API intensity across OpenAI's enterprise data , meaning software development firms, legal technology companies, consulting practices, and financial services are the current early winners. This aligns with what the industry already knew empirically: coding agents like Codex have the highest measurable ROI because code either works or it does not, making value creation verifiable. The harder and more important question is what happens when agentic AI achieves similar output verifiability in other domains: legal contract review, financial modeling, clinical documentation, and supply chain planning. Those domains represent the next wave of frontier firm formation.
OpenAI is deliberately positioning B2B Signals as a competitive intelligence product, not purely a transparency initiative. Publishing this data creates urgency for lagging enterprise customers: if your peer group is already 3.5x more AI-intensive and the gap grew 75% in the past twelve months, the cost of not upgrading your AI strategy is now quantifiable, not hypothetical. This is a retention and upsell play disguised as research. It also signals OpenAI's long-term enterprise revenue strategy: the next phase of growth is not new customer acquisition , it is depth expansion among existing customers who are currently average users of a product they could be using far more intensively.
Hidden Insight: The Productivity Gap Has a Compounding Mechanism
Here is what makes the B2B Signals data structurally alarming for lagging firms: the 3.5x advantage compounds through a mechanism that is genuinely difficult to interrupt once established. Firms using AI more deeply generate better institutional knowledge of which workflows respond to AI delegation, develop more refined prompt libraries and agent orchestration patterns, and attract employees who are more AI-fluent , who in turn generate even higher-quality usage patterns. This is not just a capability gap. It is an organizational learning gap embedded in hiring, culture, and process. Organizational learning gaps close slowly and unevenly.
The historical parallel worth examining is enterprise software adoption in the 1990s and 2000s. Early ERP and CRM adopters , the frontier firms of that era , compounded their advantage through institutional knowledge of how to extract real value from complex, expensive software. Late adopters could purchase the identical software license but could not replicate the organizational knowledge that made the software yield ROI. The productivity gap between early and late ERP adopters took the better part of fifteen years to fully close. If the AI productivity gap follows a similar diffusion curve, the companies that are 3.5x ahead today will not be 3.5x ahead in five years , but they will likely be further ahead in absolute terms, because their organizations will have developed agentic workflow competencies that average users have not yet begun to build.
The deepest insight in B2B Signals is implicit rather than stated. The report measures usage intensity, not business outcomes. But the correlation between Codex adoption , the most outcome-verifiable agentic tool , and frontier firm status tells an important story. Codex usage is measurable precisely because code outputs can be tested, deployed, and evaluated objectively. If frontier firms are 16x more intensive in their use of the most output-verifiable AI tool, and they have reached 3.5x overall intelligence intensity, the inference is strong: they are generating compounding productive advantages that will eventually manifest as lower software development costs, faster product iteration cycles, and ultimately, higher margins and market share in their respective sectors.
What to Watch Next
Watch OpenAI's enterprise pricing announcements over the next 90 days. B2B Signals creates the empirical foundation for tiered enterprise contracts priced on intelligence intensity rather than seats , a shift that would dramatically increase average revenue per user from top-decile enterprise accounts. If OpenAI introduces usage-based pricing tiers aligned with intelligence intensity metrics, the B2B Signals report will look in retrospect like the intellectual groundwork for a significant monetization architecture change. Watch for Anthropic, Google, and Microsoft to publish analogous enterprise usage data. If competitors go quiet after the B2B Signals publication, it likely signals either that their data tells a less favorable story or that they have not yet recognized the strategic value of publishing enterprise intelligence data as a sales and retention tool.
Watch the labor market for leading indicators that the frontier firm advantage is beginning to show up in hiring patterns. If the most AI-intensive firms , identified through public job postings, employee LinkedIn profiles, and earnings calls , begin attracting disproportionate talent at premium compensation, it will be the first external sign that the intelligence intensity gap is translating into business performance. The 30-to-90-day window is too short for the compounding mechanism to show up in earnings. But talent flows are faster, and they are the canary in this particular coal mine.
The AI productivity gap is not widening because frontier firms use AI more , it is widening because they have crossed a threshold into systematic delegation, and that threshold represents organizational capability that a software purchase alone cannot replicate.
Key Takeaways
- 3.5x intelligence per worker at frontier firms , The gap versus typical firms grew 75% in one year, up from 2x in 2025, according to OpenAI's first B2B Signals enterprise data report.
- 64% of the gap comes from depth, not volume , Most of the frontier advantage is driven by richer, more complex agentic workflows rather than simply sending more AI messages.
- Codex 16x gap and 5x growth , Frontier firms send 16x as many Codex messages per worker as typical firms; Codex hit 3 million weekly active users after growing 5x since January 2026.
Questions Worth Asking
- If 64% of the AI productivity gap comes from depth of use rather than frequency, what specific organizational changes , hiring, process redesign, incentive structures , are actually required to join the frontier cohort, and are most companies willing to make them?
- If the AI productivity gap follows the ERP adoption curve and takes 10-15 years to fully close, which industries will see the most structural market share concentration toward firms that crossed the agentic threshold earliest?
- Is OpenAI publishing B2B Signals primarily as a transparency initiative, or is this the intellectual groundwork for a shift to intelligence-intensity-based enterprise pricing that would dramatically increase ARPU from top-decile customers?