Where industrial capital went this year

Capital has a tell. It commits before the consensus catches up, and the size of the cheque says more than the press release around it. Read the flows of the past eighteen months and a single pattern shows through: money moving back into the physical economy, at a scale that the last cycle reserved for software.

Four currents carried most of it.

Defence tech stopped waiting for grants

The clearest signal came from European defence technology, a category that spent years proving it could build hardware and win contracts while the question of whether it could attract serious private capital stayed open.

That question closed in July 2026, when Munich’s Quantum Systems raised a $1.2bn Series D at a post-money valuation of roughly $8bn, the largest private defence-technology financing raised in Europe to date, co-led by Blackstone, Noteus, Airbus and Advent (Quantum Systems). It did not arrive alone. Helsing, the other German defence-AI firm, raised €600m in June 2025 at a €12bn valuation (Helsing). Across the Atlantic, Anduril raised $2.5bn in June 2025 at a $30.5bn valuation, more than doubling its prior mark (TechCrunch).

Three rounds, three continents of investor, one reading. Institutional and crossover capital now underwrites defence hardware at growth-equity scale. Government grants no longer carry the category alone.

The power build-out became a hardware problem

The second current runs into electricity, and it has moved from a demand story to a supply-side scramble.

The demand is the AI data centre. Grid Strategies projects US peak electricity demand rising by about 166 GW over the five years to 2030, more than four times the equivalent 2023 estimate, with data centres accounting for roughly 55% of the growth (Grid Strategies, via Utility Dive). Demand of that shape resolves in the equipment that connects generation to load, and that equipment has become the binding constraint.

The capital is following the constraint. Hitachi Energy broke ground on 29 June 2026 on a $457m plant in South Boston, Virginia, described as the largest facility for large power transformers in the United States, adding about 825 jobs, and forming part of a wider grid-equipment investment of more than $1bn (Hitachi Energy). Transformers carry some of the worst lead times in the whole build-out. A plant of this scale is capital betting that the bottleneck is durable enough to underwrite a factory.

The same logic runs through the chips that manage the power. Infineon opened a €5bn power-semiconductor fab in Dresden on 2 July 2026, the largest single investment in its history, backed by around €1bn of public money under the EU Chips Act and the IPCEI programme, and creating about 1,000 jobs (Infineon, via PR Newswire). Power semiconductors sit between the grid and the server rack, and between the battery and the wheels. The fab is a bet on both curves at once.

Reshoring turned policy into concrete

The third current is the return of manufacturing to home soil, and here the flow shows up in poured foundations before it shows up in funding rounds.

US manufacturing construction spending peaked at roughly $235bn in 2024, having tripled from the level of a few years earlier (Wolf Street, citing Census data). The catalyst was policy made concrete: the CHIPS and Science Act authorised $52.7bn for domestic semiconductor manufacturing (FRED / CHIPS Act), and the private capital that stacked on top of it dwarfed the public sum.

The honest note is that the curve has rolled over. Spending has eased from the 2024 peak, and the semiconductor-heavy segment has cooled hardest as the first wave of fabs finishes construction. The reshoring flow is real and it is maturing, moving from breaking ground to filling the buildings that the ground broke for.

Robotics priced the labour it replaces

The fourth current is the narrowest in company count and among the steepest in valuation. Physical AI needs a body, and capital is paying up for the ones that look production-bound.

Figure exceeded $1bn in a Series C in September 2025 at a $39bn post-money valuation, backed by a syndicate including Nvidia, Brookfield, Intel Capital and Qualcomm Ventures (Figure). The valuation only makes sense against the size of the market a general-purpose humanoid would address if it works. Investors are pricing the option on physical labour, with current revenue a footnote.

The industrial incumbents are buying into the same thesis by acquisition. onsemi agreed in June 2026 to buy Synaptics for roughly $7bn in an all-stock deal, reaching beyond its power-and-sensing base for the edge-AI compute, human-machine interface and connectivity layers that turn a component into a system (onsemi, via Yahoo Finance). A power company paid its record price for capabilities it could not build in time. That is what conviction looks like on a balance sheet.

The pattern the flows describe

Set the four currents side by side and they point the same way. Defence, power, reshoring and robotics all draw on one shared base: the physical, capital-intensive, hard-to-copy end of the economy that the software cycle spent twenty years routing around.

The capital moving into that base is funding the equipment, the fabs, the foundations and the bodies that a rebuilt industrial economy runs on. The rounds are the visible edge of a slower shift in where serious money believes the next decade of value gets made.

ISI tracks these flows across nine industrial domains every week. The Raise, our Friday issue, follows the capital into the rounds that fund the build.