CorrectionOur 15 July correction to the 14 July editions itself carried an incorrect figure — August TTF settled at €53.06/MWh on 14 July, not €44.18. The cause was a stale exchange-data feed, now fixed. Read the full account →
Fluence Energy's 98% rally bets the grid bottleneck outlasts chip supply
Battery storage contracts with hyperscalers signal that power infrastructure, not GPUs, now sets the pace for AI data centre expansion.
Fluence Energy's stock climbed 98% in the week ending Thursday (2026-05-21) after the battery storage company disclosed a record backlog and signed master supply agreements with two major hyperscalers.3,2 The move was sharp enough to lift the company's market cap to $3.6bn, up from a 52-week low of $4.40 a share.1
The jump confirms what grid planners have been arguing for months: the binding constraint on AI data centre deployment has shifted from GPU availability to the power infrastructure needed to connect those facilities to the grid. Capital is rotating accordingly, with nuclear and renewable baseload generation among the cleaner candidates to fill the gap.3
Management reaffirmed a 2026 revenue target of $3.2bn to $3.6bn, with 85% of the midpoint already contracted.2 Supply chain disruptions delayed roughly $80m in second-quarter billings, and analysts expect those deferred shipments to flow into third-quarter results as delivery schedules normalise.2 Adjusted gross margins improved despite the revenue miss, and the master supply agreements with two hyperscalers mark a deliberate expansion into the data centre vertical.2
Fluence's situation is not isolated. Panelists at the BloombergNEF Summit in New York in April cited high battery pack prices and global shipping bottlenecks as the main drag on deployments, even as utility demand for energy storage grows.6 The supply chain has not yet caught up to the pace at which hyperscalers are committing.
For gas markets, the connection runs through the gap that storage and interconnection delays leave open. Where battery or grid buildout cannot keep pace, data centre operators need firm generation to bridge the interval until grid upgrades are complete. Crusoe Energy has already secured 4.5GW of gas-powered capacity for US data centre infrastructure, with OpenAI's Stargate venture among the potential customers.4 AI systems could account for close to half of total data centre power consumption by the end of 2025, according to estimates by Alex de Vries-Gao of Digiconomist, which gives some scale to how much firm capacity that implies.4
The urgency is unevenly distributed by geography. In China, industrial electricity costs run at roughly half the rate paid by many American facilities, according to official figures.7 Beijing sets residential power prices independently, which limits political resistance to energy-intensive infrastructure. UBS analyst Ken Liu expects China to add another 25GW of AI data centre capacity by 2029, five times the 5GW it built over the prior two years.7
The US faces a different arithmetic. Grid interconnection queues are measured in years, and analysts expect listed US power companies' sales to grow at an annual rate of 6% between 2025 and 2028 in nominal terms, up from 4% since 2022 — though analysts describe even that acceleration as no bonanza.5 One bank puts the potential capacity shortfall at 45GW by 2028, raising the prospect that sold chips sit idle if the power does not arrive.5 Alphabet and Microsoft are spending tens of billions on GPUs; hardware that cannot be energised does not generate returns.5
Fluence's valuation is pricing in the assumption that storage can close part of that mismatch. But the company remains lossmaking, and the secondary offering of 20 million Class A shares announced in mid-May (2026-05) introduces dilution pressure against the rally.1,2 At a price-to-sales multiple of 1.23 on a $3.2bn enterprise value, the stock is pricing delivery execution, not earnings.1
The forward question for energy infrastructure investors is whether US grid interconnection lead times begin to compress. EIA data on interconnection filings and construction starts would be the earliest visible signal. If waits stay measured in years rather than quarters, equipment makers with contracted backlogs — not the data centres themselves — will capture the margin from this infrastructure cycle regardless of how much additional GPU capacity Nvidia ships.5,3