Google Commits $750 Million to AI Agent Startups — and Opens a DeepMind Campus in Seoul

Google announced a $750 million fund for AI agent startups on April 27, and it’s structured differently from the typical corporate venture program. Rather than writing equity checks and stepping back, the package bundles cloud credits, direct engineering support, and Marketplace distribution — three things a startup building on top of Google’s models actually needs.

What the $750M fund provides

Cloud credits give portfolio startups the compute budget to train and serve agent models at scale without burning through cash runway on infrastructure. That matters: the difference between a startup that can run continuous agent evals and one that can’t is often the compute bill, not the engineering talent.

Engineering support connects founders directly with Google teams working on the same problems — model serving, tool use, memory systems, multi-agent orchestration. This is harder to replicate with money alone. Getting early access to Google engineers who’ve thought deeply about agentic failure modes is worth more than the infrastructure credits in many cases.

Marketplace distribution is the most strategically interesting piece. Qualified startups get direct access to Google’s enterprise customer base without building an independent sales motion. For an agentic AI product — where enterprise buyers are still trying to understand what they’re evaluating — distribution through a trusted channel shortens the sales cycle substantially.

The $750M figure represents committed cloud credits and program investment, not equity. Google isn’t acquiring ownership in portfolio companies through this vehicle, which means startups retain flexibility to raise from other investors. Full eligibility criteria haven’t been published.

DeepMind’s Seoul campus

On the same day, Google DeepMind confirmed it is establishing a research campus in Seoul through a partnership with the South Korean government. The campus follows direct conversations between Demis Hassabis and South Korean leadership, positioning Korea alongside Japan and Singapore as a priority hub for DeepMind’s East Asian footprint.

South Korea has moved aggressively to attract AI investment since 2025, offering infrastructure incentives and streamlined regulatory frameworks for foreign AI companies. The Seoul campus will focus on connecting with local research talent and the country’s dense hardware ecosystem — Samsung and SK Hynix give any AI research operation in Seoul a natural advantage when it comes to memory and compute access.

The competitive context

Both announcements are direct competitive moves. The agent fund is a response to Microsoft’s Copilot Studio ecosystem, AWS Bedrock AgentCore (which AWS expanded this same week with a new CLI and 14-region coverage), and the growing number of agent-focused platforms that have emerged in 2025-2026.

The Seoul campus extends DeepMind’s reach into a market where competitors are also investing. Meta’s AI Research lab has presence in Seoul, and Microsoft has committed multi-billion dollar infrastructure investments in South Korea over the past 18 months.

Google’s $32 billion acquisition of Wiz closed earlier in April. The company appears to be deploying its credibility simultaneously in cloud security and AI infrastructure, betting that both will be defining platforms for enterprise technology over the next three to five years. The agent fund is the clearest signal yet that Google views the infrastructure layer under agentic AI as a strategic priority — not just a product line.

For startups building agents today: the Google fund is worth evaluating seriously, particularly for those already building on Google Cloud or using Gemini models. The Marketplace distribution alone, if delivered on, could compress a typical enterprise sales cycle from 12 months to 3.