Key Takeaways
- Cloudflare’s Workers AI processed over 1 trillion inference requests in Q2 2026.
- Edge inference reduces latency by up to 80% vs centralized cloud for real-time applications.
- Canadian developers are among the top global users of the Workers AI free tier.
- Cloudflare is competing directly with AWS Lambda and Google Cloud Run for serverless workloads.
Cloudflare’s Workers AI platform is emerging as a credible infrastructure layer for AI inference at the edge. Running on Cloudflare’s 310-city network, Workers AI allows developers to deploy custom models and AI agents with sub-10ms latency to 99% of the world’s internet users a capability that centralized cloud providers cannot match on latency alone.
The platform processed 200 billion AI inference requests in Q2, up from 80 billion in Q1, suggesting an inflection in adoption. Cloudflare charges $0.11 per million tokens for its most capable hosted model, undercutting OpenAI’s API pricing by approximately 40%, while delivering lower latency for edge-deployed use cases.
Wall Street’s current Cloudflare valuation of approximately $35 billion assigns almost no value to Workers AI the market is still pricing the business primarily on its security and CDN revenue. If Workers AI reaches $500M ARR by 2028, a conservative outcome given current growth rates, it could add $5-7 billion to the market capitalization at typical SaaS multiples.
The risk is concentration: Cloudflare’s AI customers are predominantly developers and startups, not the enterprise contracts that drive durable revenue. The path to enterprise AI infrastructure requires winning Fortune 500 security buyers a motion Cloudflare is still learning to execute.
Edge AI Economics: Why Latency Matters for Enterprise
Cloudflare’s competitive advantage in AI inference is geographic distribution. With more than 300 data centre locations worldwide including points of presence in Toronto, Montreal, and Vancouver Workers AI can serve inference requests with sub-20ms round-trip latency for the majority of Canadian users. That compares to 80–150ms typical for centralized inference calls routed through U.S.-based hyperscaler endpoints.
The business model is usage-based, charged per million inference tokens, which aligns well with enterprise procurement preferences for variable AI workloads. Cloudflare is also positioning Workers AI as a privacy-first alternative: customer data does not leave the local PoP, which matters enormously for Canadian enterprises subject to PIPEDA and Quebec’s Law 25 data residency requirements.
| Metric | Q2 2026 | Q2 2025 | YoY Change |
|---|---|---|---|
| Total Revenue | $665M | $529M | +26% |
| Workers AI Revenue (est.) | ~$55M | ~$12M | +358% |
| Developer Network (registered) | 4.2M | 3.0M | +40% |
| PoP Locations | 310+ | 285+ | +9% |
| Adj. Operating Margin | 14.2% | 8.4% | +5.8 pts |
| Net Revenue Retention | 114% | 116% | -2 pts |