On 14 October 2026, Cummins published an announcement declaring that its new 3,500 kVA generator sets would be on their way to customers “this summer”. October. In Britain. The leaves are coming off the trees, the heating is going back on, and Cummins is still confidently selling August. The machinery is supposed to protect mission-critical infrastructure from catastrophic interruptions. Apparently, the first casualty of the launch was the fucking calendar.
The announcement concerned the Cummins QSK78 generator, a 78-litre, 18-cylinder diesel-powered beast intended to keep data centres operating when their normal electricity supply fails. Manufactured in Daventry, the latest Centum Series package promises enormous power in a compact design, sophisticated controls and an astonishing proposition about lifetime maintenance costs.
The following day, Cummins published another article celebrating its involvement in the artificial intelligence gold rush. This one opened with a little history lesson, informing readers that Bell Labs invented the transistor in 1948.
It was invented in December 1947. Bell Labs announced it publicly in June 1948. Even the historical record makes the distinction.
Two corporate publications in two days. One apparently trapped in the wrong season, the other starting its grand technological sermon with the wrong year. For a manufacturer selling reliability to an industry obsessed with uptime, somebody might consider introducing a fucking proofreading schedule.
Still, historical accuracy is small change compared with the spectacular financial bonfire into which Cummins is enthusiastically positioning its equipment.
The Artificial Intelligence Cash Furnace
Artificial intelligence has become the corporate world’s favourite justification for spending money at a velocity previously associated with electrical faults. Billions are disappearing into processors, buildings, cooling equipment, electricity infrastructure and increasingly elaborate promises about the future. Every expenditure arrives wearing the same reassuring label: investment.
Unfortunately, the difficulty begins when somebody asks what the investment is expected to earn.
On 14 October, technology research company Gartner published an assessment of AI infrastructure economics. Its projections suggest that more than $3.5 trillion in AI infrastructure spending could face impairment by 2030 if present investment patterns continue.
That’s $3,500,000,000,000. Enough zeroes to give Cummins’ Destination Zero department a fucking erection.
Meanwhile, Gartner estimates that AI providers will spend nearly $1 trillion on infrastructure in 2026 alone. Its forecast adds another $6 trillion between 2027 and 2030, creating an investment programme so enormous that even the most enthusiastic corporate accountant should be reaching for a stiff drink.
Yet the anticipated revenue presents a rather more complicated picture.
Trillions In, Trillions Missing
Gartner calculated that AI data-centre owners would need approximately $8.5 trillion in cumulative revenue through 2030 to cover announced capital spending, financing costs and a 12% return on invested capital. Its forecast for enterprise and consumer AI revenue attributable to those data centres was no more than $3.6 trillion.
That leaves a projected $4.9 trillion gap between the required revenue and the expected revenue.
Somewhere, an investment banker is surely preparing a PowerPoint presentation explaining why this is actually tremendously exciting.
Furthermore, Gartner predicted that more than 80% of specialist AI cloud providers, commonly called neoclouds, would cease operating independently by 2030. In another assessment published on 19 October, the research company reported that just 13% of surveyed chief information officers said their organisations had achieved significant or transformational value from AI.
The likelihood of success for individual AI use cases remained approximately one in five.
Consequently, the spectacle unfolding around Cummins becomes even more extraordinary. Industries are constructing vast technological estates to accommodate demand whose long-term financial returns face serious scrutiny from the very analysts advising corporate management.
And while everyone else argues over the economics, Cummins is standing beside the building site measuring up for generators.
The Gold Rush Has A Diesel Department
Cummins doesn’t need artificial intelligence to become profitable for every developer. Instead, the manufacturer sees an opportunity to supply equipment that protects valuable computing infrastructure from power interruptions.
The financial exposure is not identical. Cummins sells the machinery; developers carry the responsibility for making the wider investment economics work. Meanwhile, every additional facility needs electrical infrastructure, and reliability becomes an expensive priority when thousands of processors depend on an uninterrupted supply.
There is something magnificently cynical about the arrangement, even if the commercial logic makes perfect sense.
AI companies can spend years attempting to establish whether their business models will generate enough revenue to justify the infrastructure. Their suppliers, however, can begin collecting payments long before those arguments reach a conclusion.
Welcome to the artificial intelligence gold rush, where the prospectors have borrowed billions, the gold is still theoretical, and the fucking shovel salesman has already submitted his quotation.
For Cummins, the interesting question isn’t whether every speculative AI venture will survive. It’s how much machinery the company can supply while the industry keeps pouring concrete.
From Strategy To Standby, From Speculation To Sales
Cummins’ 15 October article, From strategy to standby, introduces Brandon Ewing, the company’s Senior Strategic Director – Data Centers. Ewing explains how the business becomes involved in planning power infrastructure, sometimes years before equipment reaches a site.
His team examines available electricity, water supplies, fuel availability, regulatory requirements and generator configurations. Moreover, the company’s sales pitch includes the Centum Series, QSK95 generators, supporting equipment and tailored maintenance arrangements.
It’s an industrial shopping list for the future, and the future apparently requires an extraordinary quantity of metal.
The article describes close collaboration with customers, extensive technical planning and solutions designed around individual facilities. Yet Cummins identifies no specific AI data-centre customer, named installation or measurable commercial outcome from those collaborations.
Plenty of discussion about the machinery. Considerably less about who has actually bought it.
Cummins knows how to produce customer success stories when it wants to. Its newsrooms contain named organisations, identifiable projects and enthusiastic testimonials. Consequently, the company should understand why such details matter when promoting its involvement in one of history’s largest infrastructure investment cycles.
Which customers are buying these systems? How much generating capacity have they contracted? Where are the installations, and what operational evidence demonstrates the advertised performance?
A glossy description of an engineering capability is not the same thing as a documented customer result.
No amount of corporate storytelling changes that.
The Cummins QSK78 Maintenance-Free Miracle
Now we arrive at the part where the engineering brochure apparently acquires supernatural powers.
Cummins’ 14 October announcement describes the QSK78 Centum Series generator as a 3,500 kVA, 50 Hz package built around an 18-cylinder diesel engine, integrated controls, a sophisticated alternator and the Centum Transmission Drive.
There is plenty of genuine machinery to discuss. High-pressure fuel injection, turbocharging, cooling circuits, electronic control modules and the mechanical demands of reliably producing enormous electrical output all deserve detailed engineering explanation.
Instead, Cummins offers this remarkable phrase:
“Little to no maintenance cost over the lifetime of the generator set”.
Read that again.
A vast diesel-powered industrial generating system, packed with rotating equipment, lubrication systems, filters, coolant, electronic controls and associated components, apparently capable of delivering a lifetime of service with little or no maintenance expenditure.
Well, fuck me. Somebody tell the maintenance engineers they’re being replaced by a press release.
The claim concerns the benefits of a simplified system-level design. Fair enough, show us the calculation. Define the assumed service life, identify the operating hours and establish which components require replacement.
Better still, explain whether the cost estimate includes inspections, labour, lubricants, batteries, load-bank testing and consumables.
Because industrial machinery, unlike corporate optimism, does not lubricate itself with adjectives.
Engineers, Meet The Invisible Service Bill
Every serious generator installation carries an operational history. Engineers inspect components, test electrical performance, manage fluids, monitor batteries and maintain the equipment so it will actually start when the electricity disappears.
In addition, standby generators need attention even when they spend most of their lives waiting for an emergency. Infrequent operation doesn’t magically eliminate deterioration, testing requirements or the possibility of component failure.
So how does Cummins arrive at almost no lifetime maintenance cost?
Perhaps its calculation assumes relatively few operating hours. Alternatively, the company may be describing savings associated with particular components rather than the complete maintenance burden. Either way, customers need the underlying figures before they can judge the claim.
What happens when something fails outside warranty? Who pays for servicing, and which expenses sit outside the calculation? More importantly, how much would a customer actually spend across the generator’s anticipated service life?
These aren’t obscure questions from somebody trying to catch the marketing department out. They’re ordinary questions a competent buyer asks before spending serious money on mission-critical equipment.
Yet Cummins apparently considers the phrase sufficiently impressive to publish without an accompanying cost breakdown.
The brochure promises a maintenance miracle.
The engineers might appreciate seeing the fucking arithmetic.
The Next Day, Cummins Starts Selling Maintenance
A particularly entertaining detail appears in the following day’s corporate article.
While the QSK78 announcement promotes remarkably low lifetime maintenance costs, From strategy to standby advertises customised maintenance programmes, maintenance agreements, spare-parts availability and around-the-clock field support.
Outstanding.
On Wednesday, the industrial machine apparently costs next to nothing to maintain. On Thursday, Cummins is enthusiastically discussing the infrastructure for maintaining industrial machines.
The articles concern different elements of the company’s data-centre offering. Nevertheless, their proximity makes the unanswered commercial question impossible to ignore: what does the complete lifecycle of Cummins backup-power equipment actually cost?
After all, customers aren’t buying a metaphor about minimal intervention. They’re purchasing expensive equipment that must perform under precisely the conditions when failure becomes unacceptable.
If the QSK78 really delivers the advertised lifetime economics, put the numbers on the table. Publish the service schedule, assumptions, exclusions and independently verifiable total cost of ownership.
Let engineers examine the engineering.
Until then, the most impressive component in this particular generator may be the extraordinary carrying capacity of its advertising copy.
Destination Zero, Via Eighteen Cylinders
There’s another magnificent piece of corporate theatre taking place here.
Cummins promotes its Destination Zero strategy as part of its commitment to reducing emissions and helping customers navigate the energy transition. Its published climate plans contain targets covering operational emissions, product emissions and longer-term decarbonisation.
Meanwhile, the company is enthusiastically expanding its position in the market for diesel-powered standby generation serving energy-hungry data centres.
The International Energy Agency projects that global data-centre electricity consumption will roughly double to around 945 terawatt-hours by 2030. Artificial intelligence is a principal driver of that increase, alongside growing demand for other digital services.
That is an enormous electrical appetite. Feeding it requires generation capacity, grid infrastructure and backup systems, much of which carries an environmental cost.
Cummins is perfectly comfortable discussing the commercial opportunity. Its generator announcement highlights compatibility with hydrotreated vegetable oil, alongside claims about efficient combustion and low emissions.
Wonderful. Now publish the operating assumptions.
How much fuel does a complete QSK78 installation consume under the advertised load conditions? What are its measured carbon dioxide and nitrogen oxide emissions? Which fuels are customers actually using, and how frequently will the systems operate?
Most importantly, what lifetime emissions result from the operating conditions Cummins expects its customers to encounter?
Those figures would be rather more illuminating than another corporate recital about helping the world become greener.
The $1.675 Billion Elephant In The Engine Room
Cummins has an especially compelling reason to demonstrate transparency over emissions.
In January 2024, the company agreed to a $1.675 billion civil penalty in the United States following proceedings involving illegal emissions-control defeat devices and undisclosed software features in Ram diesel pickup trucks.
The Environmental Protection Agency identified approximately 630,000 vehicles equipped with illegal defeat-device software. Hundreds of thousands of additional vehicles carried undisclosed auxiliary emission-control devices.
As a result, the settlement concerned emissions-control software affecting nearly one million vehicles and produced the largest civil penalty in the history of the US Clean Air Act.
Nearly a billion dollars would have been an astonishing penalty. Cummins managed to clear that benchmark with another $675 million on top.
That’s quite a financial achievement for a company whose corporate literature spends so much time celebrating environmental responsibility.
The settlement concerned Ram engines rather than the QSK78, but Cummins’ emissions record remains relevant when assessing the company’s environmental marketing. A manufacturer that has paid a record environmental civil penalty can hardly expect future emissions claims to command automatic reverence.
Destination Zero is a lovely slogan. Measured emissions are rather more useful.
And when you have already paid $1.675 billion over environmental compliance failures, perhaps the corporate communications department should spend less time polishing the halo and more time publishing the engineering data.
Daventry Builds The Machines, Wall Street Builds The Fantasy
The QSK78 announcement also provides a welcome reminder that actual industrial manufacturing remains a serious business.
Cummins identifies Daventry as the British manufacturing base for the Centum 78 product supplied to markets outside North America. Furthermore, its announcement highlights the integrated design and presents the UK operation as an important centre of expertise.
Engineers, technicians, production systems and complex mechanical components sit behind this product. Their work produces something tangible, with an identifiable function and measurable output.
Contrast that with the financial circus surrounding artificial intelligence.
One side produces enormous machinery capable of delivering thousands of kilovolt-amperes of electrical capacity. The other produces financial projections whose viability may depend on customers eventually discovering enough profitable reasons to use the computing infrastructure currently under construction.
Engineers can test the generator under load. Unfortunately, the business model may require another fucking funding round.
There’s an appealing absurdity in watching a traditional industrial manufacturer position itself to benefit from a technological revolution whose investment economics increasingly resemble an exercise in collective wishful thinking.
Cummins doesn’t have to invent artificial general intelligence, solve the AI industry’s revenue problem or establish whether another hundred billion dollars of processing capacity will deliver commercially worthwhile applications.
It needs equipment that customers want to buy.
That distinction helps explain why infrastructure suppliers can occupy attractive commercial positions even when the wider investment boom generates nervous conversations in finance departments.
Nevertheless, the machinery must justify its own claims. An impressive factory doesn’t establish negligible lifetime maintenance costs, while a sophisticated generator cannot make a financially unsustainable data centre profitable.
It can only keep the lights on.
A Brochure Is Not A Customer Case Study
For all the talk of cooperation, Cummins’ two publications offer remarkably little information about actual AI data-centre customers.
The company explains its expertise, advertises its technology and describes the planning process. However, readers receive no named deployment demonstrating the particular commercial benefits being promoted.
There is no representative customer’s maintenance expenditure, no documented lifecycle comparison and no detailed installation showing how the advertised operating economics perform outside the marketing department.
Instead, the reader gets another confident account of industrial capability accompanied by carefully polished claims.
For a business supplying mission-critical equipment, that’s an unnecessarily thin foundation on which to build sweeping commercial expectations.
Cummins possesses substantial engineering resources, decades of manufacturing experience and access to technical information that could establish precisely how its products perform.
So demonstrate it.
Name a representative customer willing to discuss an installation. Identify the generator configuration, service requirements and measured operating performance. Above all, publish a realistic lifecycle cost calculation that includes the assumptions and exclusions.
The numbers might even support the sales pitch.
Imagine that. Corporate marketing supported by the very evidence it keeps telling everyone to trust.
The Corporate Brochure Needs An Accountant
Taken together, these two publications offer an inviting mixture of claims about data-centre partnerships, technical capability, equipment efficiency, remarkable maintenance economics and the future of artificial intelligence.
Yet neither provides the detailed, independently testable account of commercial performance that would make the most ambitious statements genuinely informative.
For a company operating in a sector where reliability is supposed to be everything, this is an extraordinary preference for presentation over measurement.
We have an October publication promising summer deliveries. A technological history lesson that misdates the invention of the transistor. An industrial generator carrying an almost supernatural maintenance-cost proposition. Meanwhile, the company’s wider service offering continues to promote maintenance agreements and round-the-clock support.
Quite the fucking double act.
None of these questions requires revolutionary technology or artificial intelligence. Instead, they require competent documentation, accurate editing and a willingness to substantiate what the company has chosen to advertise.
The manufacturer can surely manage a few numbers without dressing them up as a miracle.
And if somebody finds the spreadsheet, perhaps they could ask the communications department to stop treating it as classified military intelligence.
The Generator Starts. The Arithmetic Doesn’t.
Artificial intelligence may yet deliver extraordinary scientific advances, substantial productivity improvements and profitable new industries. That possibility doesn’t eliminate the ludicrous economics currently projected across parts of the infrastructure market.
Gartner has supplied the warning signs. Trillions in planned expenditure, trillions less in forecast attributable revenue, substantial potential asset impairments and a looming consolidation of specialist providers.
Cummins has supplied the commercial response.
More generators. More industrial capacity. More engineering partnerships. More promotional copy about machines that apparently cost next to nothing to maintain.
The company’s executives can discuss the technological revolution while its sales organisation concentrates on the one certainty behind every data centre: the electrical infrastructure must actually work.
There is something almost beautiful about that arrangement.
A generation of technology companies is spending extraordinary sums attempting to create machines capable of increasingly sophisticated reasoning. Cummins is helping construct the infrastructure required to keep those machines alive, while publishing its own questionable dates and magnificently ambitious maintenance claims.
Perhaps somebody could ask the artificial intelligence to check the brochure.
Or, better still, let a human engineer review the numbers before the next press release escapes.
The AI industry is building cash furnaces on an industrial scale. Cummins is positioning itself to sell the emergency power.
And if the investment case finally trips the breaker, at least somebody will be standing nearby with a diesel generator and a fucking invoice.
Lee Thompson – Founder, The Cummins Accountability Project
Sources
- Cummins – Cummins introduces QSK78 Centum Series generator set (14 October 2026)
- Cummins – From strategy to standby (15 October 2026)
- Gartner – Over $3.5 Trillion of AI Infrastructure Spending Could Be Exposed to Impairment by 2030 (14 October 2026)
- Gartner – Only 13% of CIOs Report Significant or Transformational AI Value (19 October 2026)
- International Energy Agency – Energy and AI: Energy Demand from AI
- Computer History Museum – 1947: Invention of the Point-Contact Transistor
- US Environmental Protection Agency – 2024 Cummins Vehicle Emission Control Violations Settlement