Amperity is the AI-Powered Customer Data Cloud serving Alaska Airlines, Wyndham, Virgin Atlantic, Brooks Running, BECU, and dozens more — with 200+ pre-built pipelines, lakehouse-native zero-copy data sharing, identity resolution at scale, and AmpAI's MCP-server architecture. Cloudflare's developer platform is the natural runtime for the agent inference plane, the identity-resolution pipeline, the per-brand tenancy, and the edge-resident activation layer between Amperity and every downstream channel.
You've already made the hard repositioning move: from "customer data platform" to "agent-powered Customer Data Cloud." AmpAI's MCP server architecture is the proof. The next infrastructure layer — where AmpAI's agents actually run, where the model calls go through, where per-brand tenancy is enforced, where 200+ pipelines activate to downstream channels — is exactly what Cloudflare's developer platform was built to be.
"What customers are most likely to churn next quarter?" → an MCP call → an AmpAI agent → an Anthropic model → lakehouse retrieval → a structured response. Each step has its own latency budget, its own cost line, its own audit requirement. AI Gateway in front of the model layer and Workers for Platforms behind the MCP boundary are the cheapest, most observable way to ship that pipeline.
Identity resolution, lakehouse-native sharing, AmpAI agents, the 200+ pipeline activation layer, the generative AI self-service tier — each one has a different shape, and each one maps cleanly to a different Cloudflare primitive as its natural runtime. Nothing here requires ripping out the AWS estate or the Vercel marketing site.
Every AmpAI query is an Anthropic call (verified on your apex TXT). AI Gateway sits in front: semantic cache hits return without ever touching the model, per-customer attribution is automatic, budget caps prevent runaway spend, BYO keys for enterprise customers who require them.
MCP is exactly the shape Workers was built for: lightweight, edge-resident, per-customer-bound endpoints that accept structured requests, route to upstream services, and stream structured responses. Sub-millisecond cold starts, 330+ POPs, native streaming.
Deterministic + fuzzy + probabilistic matching across online + offline PII data is at heart a similarity-search problem. Vectorize indexes every identity vector and returns "is this the same customer?" in single-digit milliseconds against billions of records.
Wyndham's tenant should never be able to touch Alaska's tenant. Brooks Running's pipeline configuration should never leak into Virgin Atlantic's audit logs. Workers for Platforms makes those boundaries infrastructure, not config.
200+ pre-built pipelines to every downstream channel = a lot of small, fan-out workloads. Queues for the async event flow, Workflows for the durable multi-step pipelines with retries and audit trail. No Kafka cluster, no Airflow, no per-pipeline Lambda config.
Audience segments, campaign payloads, scored profiles — pushed to Salesforce, Braze, HubSpot, Iterable, Klaviyo, ESPs. Each push includes a payload. R2 (zero egress) holds those payloads at the edge for re-delivery, debugging, and audit.
Lakehouse-native zero-copy is the Amperity differentiator. Hyperdrive gives Workers persistent, pooled connections to Databricks SQL warehouses + Snowflake compute — turning every Worker into a lakehouse client without re-auth overhead.
You ship SOC 2, HIPAA, GDPR, CCPA, APP. You already run Detectify scans. Cloudflare's Zero Trust closes the access loop — identity-aware access to Amperity's internal consoles, customer admin UIs, and the AmpAI experiment dashboards without VPN sprawl.
Today AmpAI inference observability lives in Anthropic dashboards. Activation observability lives in CloudWatch. Per-brand attribution lives in spreadsheets. Cloudflare gives you one dashboard for inference + activation + access + audit — broken out per customer brand from request one.
Two costs dominate launching an agent-powered CDP: Anthropic inference spend (every AmpAI query is a call), and lakehouse egress (every cross-tenant activation pulls data). AI Gateway turns the first into an attributable dashboard. R2 turns the second into a flat line.
Hotels care about loyalty conversion. Retail cares about audience activation. Airlines care about identity resolution across booking + loyalty + crew. Financial Services care about HIPAA-adjacent compliance and member-level activation. Each brand inside each industry has its own data residency, its own AmpAI budget, its own audit requirements.
Every row is sourced from public DNS records, the amperity.com apex TXT, and HTTP response headers. The Cloudflare column is additive — nothing requires ripping out Vercel, AWS, Databricks, Snowflake, or any partner investment.
AmpAI's MCP architecture is the architectural inflection. The minute Amperity exposes MCP endpoints to third-party agents, the inference economics, observability, and per-brand attribution stop being internal problems — they become product surface. AI Gateway is the cheapest, fastest way to instrument that surface before it ships at scale.
Anthropic is already a vendor. The TXT record on amperity.com confirms it. AI Gateway in front of Anthropic is the lowest-friction observability + cost-control upgrade available — no model migration, no prompt rewrite, just a header change. Per-customer-brand attribution from request one.
Amplify 2026 just happened. The conference was the public moment Amperity committed to the "agent-powered Customer Data Cloud" positioning. The next 18 months are when the architectural decisions behind that positioning get locked in. Picking the runtime now is materially cheaper than re-platforming when AmpAI ships to every customer.
The interesting conversation is which of these primitives is closest to your current sprint: AI Gateway behind Anthropic, Workers as the MCP runtime, Vectorize for identity resolution, or Workers for Platforms behind the per-brand tenancy. I'd rather hear what's actually on your roadmap than guess.