Merck Bet $1 Billion on Google Cloud. It Also Did Not Stop Running AWS.

Merck announced a $1 billion deal with Google Cloud this week, according to a company press release. The pharmaceutical giant also runs Anthropic models on Amazon Web Services for some AI workloads — querying databases using plain English. Both facts are in the public record. That combination is what makes the announcement worth a second read.

What Merck has already shipped makes the scale of the commitment concrete. The company has used generative AI to prepare clinical study reports — the documents that accompany every new drug application — cutting the time to produce a first draft from 180 hours to 80, and halving error rates, Constellation Research reported. Dave Williams, Merck's chief information and digital officer, told Reuters the deal is not a pilot: Merck is already submitting these AI-assisted documents to regulators in live markets, with plans to scale globally.

The disclosure that makes the announcement worth a second read is what Constellation Research surfaced separately: Merck runs text-to-SQL workloads using Anthropic models via AWS Bedrock, alongside an internal AI platform and Celonis process mining. This is not a company that picked a side. This is a company running a deliberate multi-cloud stack and buying Google Cloud's Gemini Enterprise as an additional layer, not a replacement.

Gemini Enterprise itself is worth understanding plainly. Google's documentation describes it as an intranet search, AI assistant, and agentic platform — meaning AI that can execute multi-step tasks autonomously rather than just answering a single query — with prebuilt connectors for Confluence, Jira, SharePoint, and ServiceNow. It is an existing product, not new technology unveiled at this event. The press release language about "industry-first agentic ecosystem" is marketing copy. The product description is infrastructure.

The forward-deployed engineer model — Google engineers embedded inside Merck teams — is the delivery mechanism being tested. Williams expects the partnership to run at least a decade. Whether that horizon materializes depends on whether the embedding model produces different outcomes than standard professional services, a question nobody can answer yet.

Merck's internal AI platform reached 80 percent of its workforce as of January 2026, Constellation Research said. The company moved its Celonis center of excellence — process mining tools that identify where efficiency gains hide — into the Chief AI Office. These are the unglamorous plumbing details that make an agentic deployment actually work.

The question the announcement does not answer is accountability. When a Gemini agent automates a step in a clinical study report or flags a manufacturing anomaly, who owns the output? Regulators reviewing a drug approval do not care whether a human or an AI produced the dossier section. They care whether it is accurate, complete, and auditable. Merck is already submitting these documents in live markets. That is the strongest evidence the technology is ready for regulated use — and the clearest signal of what is at stake if it fails.

What changes first at Merck is not the technology. The technology exists and is already in use. What changes is the scope: from workflow assistance to full-company orchestration, from API calls to decade-long partnerships with embedded engineers, from proof-of-concept to regulatory submissions in global markets. If it works, every major enterprise watching has a template. If it fractures, Merck's disclosed multi-cloud architecture — the Anthropic and AWS workloads it never stopped running — becomes the story instead.

Williams is right about one thing: this is not a pilot. Merck has already decided it is real. The $1 billion is the measure of how much it is wagering on that bet being correct. The AWS Bedrock stack still running next to it is the measure of how much it is not wagering everything.