Andreessen Horowitz is leading a $30 million investment in Prosper AI, the Spanish startup that built an end-to-end AI platform for the patient journey, from booking to insurance reimbursement. A single system that answers calls, schedules directly in EHRs, verifies coverage, and contacts payers when additional information is needed. The promise is bold: cut administrative costs by over 40% and give patients upfront visibility into their financial responsibility.

The platform has already convinced over 40 healthcare organizations, reaching more than 150,000 providers and managing over $1.3 billion in care. These accelerating numbers explain the confidence from a16z, Base10, and Emergence Capital. But behind the growth lies a technical question that few ask until they face European regulations: where does the data live?

Under the hood: agentic AI and workflow orchestration

Prosper AI is not a glorified scheduler. The platform orchestrates workflows that touch disparate systems — EHRs, phone systems, insurance portals — using what the founders call “agentic AI.” In practice, a mix of models (almost certainly Large Language Models combined with deterministic modules) that manage voice interactions, natural language understanding, and multi-step decisions. The complexity lies not in any single task but in integration: bi-directional connections with major EHRs and the ability to hold phone conversations with insurers when APIs fall short.

Technically, this means the platform runs in the cloud, as a centralized service. For U.S. healthcare organizations, where Prosper AI first gained traction, the cloud is often the default. But for markets like Italy and the broader European Union, the picture is different. GDPR mandates that health data, especially data traceable to clinical history, comply with strict residency and control requirements. And the Italian Data Protection Authority’s guidelines on AI in healthcare add further caution around automated decisions.

The sovereignty knot: cloud or on-premise?

Prosper AI’s architectural choice appears cloud-first. For healthcare providers, if the infrastructure is external and managed by a non-EU vendor, the chain of data processing responsibility becomes longer and more complex. Hospitals must deal with Data Processing Agreements, impact assessments, and often the need to prove that data stays within regional borders. For many public hospitals and local health authorities, a self-hosted or on-premise deployment is not a preference — it’s a requirement.

This is the debate where AI adoption in healthcare is moving. AI-RADAR, which tracks LLM and platform deployment decisions, notes that more vertical solution providers — from diagnostics to administrative management — are starting to offer hybrid or edge options to meet control demands. For a platform like Prosper AI, hypothesizing a future on-premise module wouldn’t be a technical downgrade: it would unlock adoption in settings where data cannot leave the perimeter, provided they rethink orchestration between containers, quantized models, and direct integration into local servers.

Crunching the numbers

From a cost perspective, the Total Cost of Ownership of a cloud solution like Prosper AI is low on entry — a subscription, no hardware investment. But once transaction volumes grow beyond tens of thousands of appointments per month, the calculus can shift. For a large hospital trust handling millions of interactions yearly, evaluating an on-premise deployment on its own GPU infrastructure with adequate VRAM could become economically sensible, at least for inference models. In-house skills are needed to maintain the pipeline, but operational control and cost predictability are assets many IT leaders consider non-negotiable.

With the new funding, Prosper AI aims to expand its integration with major EHR platforms and grow across provider groups and health systems. Europe, and Italy in particular, are ripe markets for this kind of automation: administrative spending in healthcare is high everywhere, and streamlining the patient journey directly impacts system sustainability. But the condition to play this game will be architectural flexibility. Those wanting to bring the platform inside hospital walls — for compliance or latency reasons — will need a vendor willing to negotiate the deployment model. It’s not just a technical matter: it’s the real dividing line of the next wave of healthcare AI in Europe.