Why this exists.
The power layer inside every server was engineered for reliability, decades before anyone considered it a target. Today it sits inside the most expensive infrastructure ever deployed, and nothing in the modern security stack watches how it behaves. PowerGuard exists to close that gap, to make the physical layer of AI infrastructure visible, accountable and defensible.
This problem required a founding team that could see it from both sides of the stack simultaneously.
Shimon Eidelman
CEO Israel & CTO
Air Force veteran, 20 years in microelectronics, from semiconductor fabs to Hitachi, Amdocs and ECI. He built PowerGuard to working hardware, proven detection and a filed U.S. provisional patent, as sole inventor, before raising.
Alexander Goldinstyn
CEO US
Vice President at JPMorgan Chase in New York, leading regulatory reporting and risk controls for one of the world's largest banks.
Omer Schneider BOARD ADVISOR
Co-founder and CEO of CyberX, the OT security company acquired by Microsoft
Yury Kreimer TECHNICAL ADVISOR
CTO, FortifyIQ, silicon side-channel security
Yigal Unna STRATEGIC CONSULTANT
Former head of Israel's National Cyber Directorate
Plain answers.
What exactly do you install, and when?
In the first phase, nothing. Discovery reads your fleet through its existing management interfaces. If you continue, the second phase runs PowerGuard software on your servers in a record-only mode, it observes and logs the decisions it would have made, and changes nothing. Enforcement is a third, separate step, switched on only after you have reviewed that record. Each step has a condition agreed before the previous one starts.
Does this void our hardware warranty or support contract?
The discovery phase touches nothing, so nothing is affected. Before any software is installed in later phases, support and warranty implications for your specific vendors and platforms are reviewed together and agreed in writing as part of the pilot terms.
What load does it add to our systems?
Discovery adds only standard management-interface reads. The observation software is designed to run at the management layer, outside your workload path. Measured load on your own hardware is part of the pilot's agreed success criteria, we publish the numbers to you rather than asking you to trust a brochure.
What happens if it gets a decision wrong?
Enforcement never acts on a statistical anomaly alone, action requires physical or protocol evidence, and it runs under your own policy: the verdict is carried into controls you operate, under authority you set and can revoke at any time. Before enforcement is ever enabled, the record-only phase shows you every decision the system would have made on your own workload, so a wrong tendency is visible before it can act. A legitimate load spike is examined, not punished.
Can it run with no outbound connectivity?
Yes. Detection and response run locally, and the analysis tier can be deployed entirely on premises. Air-gapped operation is a supported configuration, not a special case.
What data leaves our environment?
None has to. The system reads machine telemetry: voltage, current, temperature, power, and management-plane commands. No personal data is involved. If you choose a fully local deployment, everything stays inside your environment.
Which hardware do you support today?
Validation to date is on our own instrumented hardware. Support for your production platforms is validated jointly during the pilot, on your own or production-identical machines, that is what the pilot is for.
What does a pilot involve, and what does it cost?
A focused pilot on 1–5 servers over 4–6 weeks, with success criteria agreed in advance. It starts with the nothing-installed discovery phase. Commercial terms are discussed directly, talk to us below.
Talk to us about a pilot.
Tell us about your environment and timeline. Shimon Eidelman will reply within 24 hours.