Making room in a data center
Google DeepMind reported that an AlphaEvolve-discovered scheduling heuristic had been in production for more than a year and continuously recovered an average 0.7% of Google's worldwide compute resources.
- Maturity
- Deployed, stage 4 of 4
- Support
- 1 source · institution
- Evidence detail
- How we know ↓
Data-center capacity is finite, and idle or poorly allocated compute leaves less capacity for useful work on the same infrastructure.
A production scheduler must make dependable allocation decisions across changing workloads at very large scale.
AlphaEvolve used language models, automated evaluation, and evolutionary search to produce a scheduling heuristic for Google's Borg system.
What was shown, and what wasn’t
Shown
Google says the production heuristic continuously recovers, on average, 0.7% of its worldwide compute resources.
Not shown · limits
- This is a company-reported, Google-specific operational result.
- The announcement does not quantify electricity, water, carbon, reliability, or independent-audit outcomes.
Bright editorial interpretation
Recovering capacity can allow more tasks to run on an existing computational footprint.
Still open
Would the heuristic transfer to other scheduler designs?
What are the measured energy and reliability effects under representative workloads?
How we know1 source · checked 2026-09-07 · no corrections
Original sources
Institutions: Google DeepMind · Google
- Maturity
- Deployed
- Source published
- 2025-05-14
- Captured
- 2026-09-07
- Last source review
- 2026-09-07
- Editorial method
- AI-assisted source review
Bright compared this account with the linked original and supporting sources and kept reported, budgeted, projected, and observed claims distinct. Bright did not independently audit the underlying records.
Maturity describes the tested or operational setting. Confidence describes support for the particular claim; one does not determine the other.
Revision & correction history
2026-09-07 · Recovering capacity can allow more tasks to run on an existing computational footprint.
No corrections recorded.
Keep exploring
Explore the shared question in another setting. These connections do not imply replication.