Google's Gemini Spark Runs 24/7 for AI Ultra Users
Key insights
- Gemini Spark runs on dedicated cloud VMs that persist between sessions, executing automations inside Gmail and Calendar without user presence.
- Google AI Ultra costs $100 per month, making Spark the first persistent consumer agent tied to a premium subscription tier.
- Spark directly competes with OpenAI Codex and Anthropic Claude workflows in the emerging persistent-agent market segment.
Why this matters
Persistent agents on dedicated VMs signal that consumer AI infrastructure costs are shifting from per-query pricing to always-on subscriptions, forcing every major provider to recalculate unit economics at scale. Google's deep integration with Gmail and Calendar gives it a data access advantage that OpenAI and Anthropic cannot replicate without equivalent platform ownership, creating a structurally asymmetric competitive position. The two-week gap between Google I/O announcement and production rollout compresses the response window for rivals, making the next 60-90 days the critical period for OpenAI and Anthropic to announce comparable persistent-agent capabilities.
Summary
Google began shipping Gemini Spark this week to US AI Ultra subscribers ($100/month), the first consumer-scale deployment of a continuously running personal agent by a major tech company.
Spark runs on dedicated cloud VMs that persist between sessions, executing tasks inside Gmail, Calendar, and third-party apps without user presence.
Essentially: (Google, OpenAI, Anthropic) are now competing directly on persistent agent infrastructure.
- Announced at Google I/O on May 19, Spark entered US production within two weeks.
- Google absorbs persistent VM compute costs per subscriber, not just per-query inference costs.
- App integrations suggest an ambient platform play extending well beyond a chatbot upgrade.
Owning a user's background compute is now the central contest in the consumer AI platform war.
Potential risks and opportunities
Risks
- If a Spark automation triggers an unintended Gmail send or Calendar change while a user is absent, Google faces reputational and potential legal exposure before agent-initiated action consent frameworks are legally codified.
- OpenAI and Anthropic could announce competing persistent-agent subscription tiers within 60-90 days, compressing Google's first-mover window before Spark's subscriber base reaches meaningful scale.
- Enterprise Workspace admins who permit AI Ultra adoption may create compliance gaps if Spark accesses regulated employee communications without auditable, consent-based agent action logs.
Opportunities
- Identity and authorization vendors (Okta, Auth0) gain leverage as third-party app developers need agent-specific OAuth scopes and permission frameworks to integrate safely with Spark.
- Workflow automation platforms (Zapier, Make) have a narrow window to reposition as Spark-compatible orchestration layers before Google closes the integration surface to direct competitors.
- Cloud cost monitoring vendors (Datadog, Cloudability) can market persistent-agent compute visibility tools to enterprises whose employees adopt AI Ultra and generate unbudgeted always-on VM spend.
What we don't know yet
- Whether Google has disclosed AI Ultra subscriber counts, making it impossible to assess how many users are stress-testing Spark's persistent VM economics at launch.
- Which specific third-party apps are authorized beyond Gmail and Calendar at launch, and what the agent OAuth permission model looks like for those integrations.
- How Google assigns liability and generates audit trails when Spark executes actions in a user's account while the user is offline, given existing GDPR and CCPA obligations.
Originally reported by technobaboy.com
Read the original article →Original headline: Gemini Spark, Google's 24/7 Persistent AI Agent, Goes Live for US AI Ultra Subscribers