The CEO of supply chain giant Flex says infrastructure and the grid needs to scale with AI—or else ‘Forget all the frontier models’

Flex CEO Revathi Advaithi warned that U.S. power infrastructure may not keep pace with AI-related electricity demand, predicting the country could remain in a power deficit over the next decade without faster innovation and approvals. A 2025 Berkeley study estimates data centers could consume around 11.8% of U.S. electricity by 2030; Advaithi said building a capable grid takes decades, compared with about five years to develop a chip.

Analysis

The investable implication is a shift in AI infrastructure bottlenecks: incremental value may accrue less reliably to compute suppliers and more to equipment, generation, and grid-enabling projects. Eaton, Vertiv, Schneider Electric, GE Vernova, and Quanta Services are plausible beneficiaries if utilities and data-center operators convert concern into funded orders; permitting and interconnection delays could, however, defer rather than accelerate revenue. Data-center developers may increasingly favor on-site generation, storage, and long-term power contracts, changing the mix of demand across utilities and equipment suppliers.

For Flex, the comments are not evidence of incremental orders or earnings. Its exposure is potentially relevant, but verify segment revenue, order growth, and customer concentration before treating the issue as a company-specific catalyst. The CEO’s warning may also serve Flex’s commercial narrative; industry-level constraints are credible, but neither the quoted forecast nor the cited power-demand estimate establishes a near-term shortfall at any particular project.

Days: likely a modest narrative rotation, not a standalone reason to reprice AI leaders. Over 1–3 months, track utility capex plans, interconnection queues, power purchase agreements, and equipment backlog/guidance. Over 6–18 months, sustained grid constraints could delay data-center energization and monetize power availability rather than planned compute capacity. Contrarian risk: large operators may finance behind-the-meter supply and improve utilization, keeping AI deployment growing despite grid limits. The thesis weakens if permitting, transmission awards, and energization timelines improve while equipment orders fail to follow.

Key Decisions for Investors

• Prefer a measured relative-value watchlist: long diversified electrical-equipment and grid-construction exposure against a basket of AI data-center buildout beneficiaries, rather than shorting AI compute outright. Initiate only on confirmed order/backlog strength; cap risk if equipment guidance or backlog rolls over.
• Treat FLEX as an alert, not a direct buy signal. Verify data-center/power-infrastructure segment exposure, customer-linked orders, and margin contribution in filings or guidance before assigning value to the infrastructure thesis.
• Over the next 1–3 months, monitor interconnection approvals, utility capex updates, data-center power contracts, and equipment lead times. Evidence of faster energization without rising supplier orders would challenge the expected beneficiaries.
• Avoid a broad short of hyperscalers on this evidence alone: their ability to secure long-term supply or deploy on-site generation could shift costs and timing without materially impairing AI demand. Reassess if major projects are delayed or capex guidance is cut.