“Test, time, scaling is going to go through the roof.”
San Jose, CA — November 5, 2025. Pebble has joined the NVIDIA Developer Ecosystem, bringing the company’s power-aware optimization platform closer to NVIDIA’s AI infrastructure stack. The membership gives Pebble access to NVIDIA’s developer tooling, reference designs, and validation programs — accelerating deployment of Pebble across customer environments running NVIDIA GPUs.
The energy crisis context
Data center investments by Google, Microsoft, and Amazon exceeded U.S. oil and gas industry spending in 2023, representing about 0.5% of U.S. GDP. ARM estimates that U.S. data centers will expand from 1.5% of national energy consumption in 2021 to over 20% by 2030. Without a structural response, the AI build-out will collide with grid capacity long before it collides with model capability.
What Pebble brings
Pebble profiles every AI workload in real time and finds two thresholds for each job: the efficiency knee — the power cap where tokens-per-watt peaks — and the SLA threshold — the lowest cap that still hits the workload’s SLAs. Operators use the margin between them two ways:
- Pebble Sonar caps each GPU at the efficiency knee, reclaiming 20–30% of stranded TDP. That headroom becomes capacity for additional GPUs inside the same kilowatt envelope.
- Pebble Flex uses the SLA-threshold margin to release power back to the utility on grid-signal, qualifying clusters for flex-power contracts that bypass the 7-year interconnect queue.
Technical integration
Pebble’s optimization layer runs alongside production inference stacks — vLLM, SGLang, TRT-LLM — without requiring application changes. NVIDIA Developer Ecosystem membership enables deeper integration with the NEMO framework, NIM microservices, and the NVIDIA Agentic framework for deployment on cloud-managed services and edge environments.
About Pebble
Pebble is the intelligence layer between AI infrastructure and the power grid. The company helps operators pack more GPUs into the same power envelope, or unlock flex-power contracts that turn workload SLA-headroom into grid-flexible capacity — all without code changes to the underlying serving stack.
Media contact
MyPebble, Inc.
333 West San Carlos Street, Riverpark Tower, 4th Floor
San Jose, CA 95110
1-888-314-1019 · tech@gopebble.com