Hyperscale capital spending has turned the world's largest technology companies into industrial buyers, while refineries, plants, warehouses and quarries are purchasing simulation, autonomy and machine learning as operating equipment. Robotics venture funding has already passed $18.8 billion this year, above the whole of 2025. This is the territory BrentDesk covers, and the standard it applies to it.
On 5 August, Siemens Energy reported a record order backlog of 51 billion euros in its grid technologies business, with quarterly orders in that segment up 28 percent to 5.4 billion euros and most of the increase coming from large power transformers. In the same year, the four largest buyers of data centre capacity have guided to capital spending approaching three-quarters of a trillion dollars, against roughly $410 billion between them in 2025. The growth of the world's most valuable technology companies is now being recorded, quarter by quarter, in the order books of firms that wind copper coils and pour concrete.
That is one half of a trade, and the less examined half is the return leg. Emerson's AI hybrid models are running inside the planning system that decides what Aramco's refineries process, not beside it. PepsiCo is sizing capital expenditure for United States plants and warehouses against simulation models built with Siemens and NVIDIA before anything is installed. ADNOC has a Taurob robot walking inspection rounds at its Taweelah gas compression plant. Sawmills and food plants are buying vibration monitoring that writes its own work orders. None of these are information technology projects. They sit in the operating chain and the capital approval process, which is where equipment sits.
Put the two halves together and the shape of the market changes. The physical economy has become a technology market, and technology has become a physical business. The claim is not that industry has gone digital, which has been asserted every year for two decades and has usually meant a dashboard. It is narrower and more testable: technology purchases now sit inside industrial operating and capital decisions, and the technology industry's own expansion is now rate-limited by industrial supply.
The capital markets have already repriced this. Robotics companies have raised $18.8 billion in venture funding so far in 2026, against $15 billion across the whole of 2025 and $14.1 billion in 2021, the previous peak. Apptronik added $520 million in February to a round that now exceeds $935 million. Mind Robotics, spun out of Rivian, closed a $500 million Series A in March and raised a further $400 million in May. Money that a decade ago went to software with no factory attached is now going to companies whose product has to be manufactured, shipped, certified and serviced.
The vendors have repositioned to match. At GTC in San Jose in March, Jensen Huang, founder and chief executive of NVIDIA, said that physical AI had arrived and that every industrial company would become a robotics company. That is a sales pitch, and should be read as one. What is harder to discount is who stood behind it: FANUC, ABB, KUKA and Yaskawa, which between them account for most of the world's installed industrial robot base, all committed to putting NVIDIA's Omniverse and Isaac simulation stack inside their own development and validation systems. Siemens and NVIDIA are building what they call an industrial AI operating system, and Siemens is putting the tooling used to assemble digital twins onto its Xcelerator marketplace this year. Sold that way, the capability stops being an integration project and becomes a line item, which changes who is able to buy it at all.
The installed base is a useful corrective to the enthusiasm. The International Federation of Robotics counted 542,000 industrial robots installed in 2024 at a record market value of $16.7 billion, with an operational stock of 4.66 million machines and 54 percent of new installations going to China alone. That last figure is the one to hold onto. Robot demand tracks where manufacturing capacity is being built, not where labour is most expensive. The market for physical automation is largest where physical plant is largest, which is a different proposition from the one usually sold.
In logistics the shift is furthest advanced and least discussed. Amazon operates close to a million robots. DHL runs more than 7,500 autonomous warehouse robots, has at least one automated system in over 90 percent of its warehouses, and with Locus Robotics has passed a billion cumulative picks across more than 40 sites. UPS finished last year with 127 automated buildings after adding 57 in a single quarter, plans another 24 this year, and expects to run 68 percent of its United States volume through automated facilities by the end of it. That is network design, not a pilot programme.
Earthmoving is following, with clearer economics than most people expect. Komatsu's connected construction platform now covers more than 40,000 sites. Heidelberg Materials moved over two million tons of limestone at its Lake Bridgeport quarry in Texas using a mixed autonomous hauling fleet drawn from both Komatsu and Caterpillar, an interoperability result that matters more than either vendor's individual product claim. Caterpillar, working with NVIDIA, has extended autonomy and machine assistance across excavators, dozers, compactors, loaders and haul trucks. Humanoid robots, the most heavily promoted category, are at an earlier and more honest stage: Agility Robotics reports its Digit machines have accumulated more than 65,000 operating hours across nine customer sites including GXO, Schaeffler and Toyota Motor Manufacturing Canada, and BMW is moving from testing to a pilot at Leipzig on high-voltage battery assembly. Sixty-five thousand hours is a real number and a small one.
Every one of these deployments imports a liability along with the capability. Dragos counted 1,140 ransomware incidents involving industrial organisations in the second quarter of this year, 747 of them in manufacturing, and the striking feature of the worst cases is that the attackers never reached a control system. West Pharmaceutical Services shut its plants worldwide for more than three weeks in May to contain an intrusion in its enterprise systems. Connectivity is the enabling condition of remote monitoring, digital twins, planning models and autonomous inspection alike; none of them work air-gapped. Security is therefore not a separate programme to be argued about afterwards. It is part of the running cost of the equipment.
The constraint running the other way is more stubborn still. Data centre construction has moved its bottleneck off the semiconductor and onto grid connections, transformers, cable and cooling plant, where lead times are measured in years and cannot be shortened by capital alone. The reason Siemens Energy's backlog is a record is the same reason hyperscale campuses are now sited by where a connection can be obtained rather than where land is cheap. Whoever can build and energise capacity holds the scarce position in this cycle, and that is an industrial competence, not a software one.
Saudi Arabia and the Gulf sit at the intersection of both directions, which is why this publication is based where it is. Aramco has five plants recognised in the World Economic Forum's Global Lighthouse Network. Alat, the Public Investment Fund's industrial technology company, carries a $100 billion mandate and is building robots in Riyadh with SoftBank rather than only buying them, though the gap between the Kingdom's most instrumented plants and its Technology Vendors Learned What Saudi Contractors Will Actually Buy
From Robots to Concrete Sensors: 10 Technologies Changing ConstructionThe Physical World Is Becoming Software-Defined, One Component at a TimeProject Tracking Tools Push Into Materials and Supplier DataTen Problems a Saudi Contractor Has, and What on the Big 5 Floor Answers Them