AI Demand Reaches Storage, Power and Prices

The strongest signal in the past 30 hours is a change in where the AI cycle can be observed. It is no longer confined to cloud revenue, model launches, or accelerator orders. Kioxia’s flash-memory results and Eaton’s electrical backlog show demand reaching storage and power-management suppliers. Eurostat and the U.S. Bureau of Labor Statistics show that energy and labor costs remain difficult. Most unusually, the Bank of Japan now treats AI-related demand as both support for growth and a possible source of semiconductor inflation and asset-price adjustment. The coherent conclusion is not that AI is causing global inflation. It is that AI deployment has become large enough to interact with real capacity, input prices, cash conversion, and the cost of capital.

Kioxia supplied the clearest direct evidence. For the three months ended June 30, the company reported IFRS revenue of ¥1.767 trillion, operating profit of ¥1.270 trillion, and profit attributable to owners of the parent of ¥842.2 billion. Revenue was ¥764.3 billion higher than the previous quarter and ¥1.424 trillion higher than a year earlier. Within that total, SSD and storage revenue reached ¥1.175 trillion, up ¥574.4 billion quarter over quarter. Kioxia attributed the revenue increase primarily to a significant rise in average selling prices resulting from strong demand among data-center customers focused on generative AI. That wording matters: the immediate earnings impulse came principally through price, not a disclosed unit-volume measure.

The quality and durability of Kioxia’s result require qualification. Its ¥1.326 trillion non-GAAP operating profit excludes ¥0.2 billion of purchase-price-allocation effects, ¥19.4 billion of stock compensation, and a ¥36.6 billion litigation-loss provision. The company says those adjusted measures are internal, unaudited, and may not accurately reflect its financial condition; the IFRS operating-profit figure is therefore the cleaner reference. A weaker yen also helped the year-over-year comparison, while memory is a famously volatile industry. Management nevertheless expects September-quarter revenue of ¥2.390 trillion and IFRS operating profit of ¥1.890 trillion because data-center demand should remain strong. That is guidance exposed to pricing, demand, competition, foreign exchange, and execution—not a guaranteed run rate.

Eaton shows the demand spillover on the power side. Second-quarter sales rose 21% year over year to $8.531 billion, comprising 14% organic growth and 7% from acquisitions. Twelve-month rolling orders increased 41% organically in Electrical Americas and 33% in Electrical Global; their respective backlogs rose 33% and 103%, producing 43% combined electrical backlog growth. Eaton called data centers a key growth driver while emphasizing broad end-market strength. This distinction prevents a common overclaim: the results confirm that data-center construction is helping electrical demand, but Eaton does not isolate an audited AI revenue line. Segment margin was 23.1%, down 80 basis points year over year despite sequential improvement.

Eaton’s earnings bridge also shows why physical demand should be assessed through both growth and conversion. GAAP diluted earnings per share were $2.11; adjusted EPS of $3.15 excludes $0.50 for intangible amortization, $0.49 for acquisitions and divestitures, and $0.05 for restructuring. Operating cash flow was $1.127 billion and company-defined free cash flow was $874 million after $253 million of capital expenditure. Full-year organic-growth guidance rose to 11–13%, but the same release warns about raw-material, energy, component, labor, supply-chain, and technology risks. Backlog is valuable evidence of demand; it is not identical to revenue, margin, or cash until orders are delivered without cancellation, delay, or cost overruns.

The macro releases show why those conversion risks cannot be treated as background noise. Eurostat’s July flash estimate put euro-area annual inflation at 2.9%, up from 2.8% in June. Energy inflation accelerated to 10.0% from 8.5%, while services moved to 3.3% from 3.2%; food, alcohol and tobacco eased to 1.2%, and non-energy industrial goods rose to 0.9%. These are estimates for the expanded 21-member euro area and the complete July HICP dataset is not due until August 19. Eurostat does not attribute the rise to data centers or AI. The relevant conclusion is narrower: a capital-intensive AI buildout is occurring while the energy component of a major economy’s price index is already running hot.

U.S. labor costs add a different constraint. The Employment Cost Index rose 0.9% in the June quarter on a seasonally adjusted basis and 3.4% over 12 months on an unadjusted basis. Annual civilian wages and salaries increased 3.2%, benefits 3.8%, and private-industry health benefits 6.0%. Yet inflation-adjusted private wages fell 0.4% over the year. This combination can pressure employers without improving household purchasing power: nominal labor costs continue to rise while real wage gains are absent for the private-sector aggregate. The survey does not identify AI as the cause. It does show that automation investment, data-center staffing, electrical work, and enterprise deployment compete for budgets in a labor market where compensation is not rapidly disinflating.

The Bank of Japan made the AI-to-macro link explicit rather than inferred. It held the uncollateralized overnight call-rate target around 1.0% by an 8–1 vote; the dissenter proposed 1.25% because of upside price risks from overseas demand shocks and financial conditions. In its July outlook, the Bank said global AI-related demand should support Japan’s growth and investment, but that higher semiconductor prices, the yen’s depreciation, and energy costs could lift durable-goods inflation. It also warned that if AI-related profits fail to expand in line with investment, asset-price adjustment pressure could follow. Although Japan’s real rates remain negative and financial conditions are described as accommodative, the Bank says it will continue raising the policy rate as activity, prices, and financial conditions warrant.

There are serious competing interpretations. Kioxia may be capturing a cyclical memory-price surge rather than a permanent change in storage economics, and price-led revenue can invite supply that later reverses the cycle. Eaton’s acquisitions and non-data-center markets explain part of its growth, while backlog can reflect long lead times as well as final demand. European energy inflation is tied to a wider commodity and geopolitical shock, not a measured AI load effect. The BOJ itself lists oil, exchange rates, wages, fiscal policy, overseas conditions, and AI profitability as interacting risks. Simultaneous pressure across storage, electrical equipment, energy, and labor is evidence of a tighter operating environment; it is not proof of a single causal chain.

The next tests are therefore concrete. For Kioxia, watch disclosed bit shipments, average selling prices, the ¥2.390 trillion revenue outlook, capital spending, and whether IFRS cash generation remains strong when memory supply responds. For Eaton, track backlog conversion, cancellations, organic growth, and whether margins recover as Boyd Thermal is integrated. Eurostat’s full July HICP release on August 19 will replace the flash estimate, while the next U.S. ECI arrives October 30. At the BOJ, the issue is whether AI-supported output and investment broaden into productivity before semiconductor, copper, machinery, energy, and wage costs deepen inflation pressure. The AI cycle has escaped the software layer; durable value now depends on physical throughput, disciplined pricing, and returns that can survive a less forgiving capital environment.