I remember standing in a repurposed warehouse outside Prague in 2017, watching a roomful of developers try to reconcile the ICO frenzy outside with the trustless ideals we were teaching inside. No tokens to promote. No allocations. No airdrops. Just whiteboards and a stubborn conviction that decentralization was a moral architecture, not a marketing category. Forty of those developers went on to build legitimate open-source projects. The rest of the market went on to do what markets do.
I think about that warehouse whenever I read a story that frames a new technology as a straight line to riches. The latest one crossed my desk this week: Siemens Energy, the German gas turbine and grid technology giant, reported record industrial profit, and the company — along with a chorus of market commentary — pointed to AI data center electricity demand as a primary driver. A gas turbine maker is now an AI beneficiary stock.
Your first instinct might be to file this under unlikely corporate synergies. But this deserves a longer look, because the profit number is just the surface of a much deeper structural shift. It is not really about gas turbines, or about Siemens Energy, or even about AI chips. It is about the second bottleneck of the intelligence economy: electricity. And once you see that bottleneck, the entire landscape of who controls, who profits, and who gets left behind starts to reorganize.
Let me explain what I mean, and what I think most coverage is missing.
The Setup: Compute Got Thirsty
First, the context. Modern AI training clusters are not your father's server room. Rack power density has climbed from roughly ten kilowatts per rack a decade ago to fifty, one hundred, even higher kilowatts per rack for the newest GPU systems. Total facility capacity that used to land in the ten-to-fifty-megawatt range is now regularly planned at one hundred megawatts and above. This is not a linear trend; it is a step change. And it collides with a physical grid that was never designed for this.
Here is the part that market commentary tends to compress: new data centers need interconnection — permission to tap into the grid — and in regions like Northern Virginia, the interconnection queue now runs three to seven years. Some projects wait even longer. For a hyperscale cloud provider spending tens of billions on AI infrastructure, a seven-year wait is not an inconvenience; it is a competitive death sentence. So the industry has begun doing what resource-constrained industries always do: build around the constraint.
Gas turbines fit that need because they are the fastest large-scale power source you can deploy in roughly eighteen to thirty months. They are dispatchable, meaning they can ramp up and down to match demand, and they run 24/7, unlike solar and wind, which wait for the sky. The key players are Siemens Energy, GE Vernova, and Mitsubishi Heavy Industries, and the demand has pushed them into what analysts call a seller's market. Microsoft, Google, Amazon, and Meta have begun locking down electricity supply the way they once locked down semiconductor wafers — through nuclear power purchase agreements, gas capacity reservations, and site selection strategies built entirely around grid access.
None of this is exactly a secret. But the reframing matters: AI has become a physical industry. The same way Bitcoin mining taught us that digital assets are really electricity transmuted into settlement assurance at the margins of the energy system, AI is teaching us something adjacent but different — that intelligence itself has a power bill, and that bill is now large enough to move the earnings of an industrial giant. That is why this story matters to anyone who thinks about infrastructure, not just people who hold energy stocks.
What Record Profit Actually Means
Let's dig into the core fact: Siemens Energy recorded record industrial profit. The market heard "AI equals electricity equals Siemens Energy" and pushed the stock narrative accordingly. Based on my years watching project valuations versus protocol reality, and a professional life spent separating signal from incentives, I can tell you that the profit is real, and yet the narrative around it deserves a serious audit.
First, what "record industrial profit" actually means. Siemens Energy is a diversified business. It sells heavy-duty gas turbines, but it also runs grid technologies and owns Siemens Gamesa, a wind turbine manufacturer that has spent years bleeding money. The "industrial profit" line typically refers to its industrial operations before group-level costs and tax effects — a critical distinction. In previous years, wind losses dragged the entire group into negative territory. A record industrial profit can coexist with an underwhelming group net result. Investors who lap up the headline without opening the segment table are buying into precisely the kind of partial truth that has always lived in the margins of corporate reporting. I am not saying the numbers are false; I am saying the story the market tells about the numbers is one selected frame among many.
Second, the timing problem. Gas turbine deals are long-cycle. The record profit Siemens Energy reported this quarter does not come primarily from orders signed in the last year; it comes from an order backlog built over years, with multi-year delivery schedules. Orders booked today show up in revenue and profit years from now — a lag that the conventional "AI demand is paying off" narrative usually ignores. Had the quarter's profit directly reflected the latest AI orders, it would have required an unusually fast project cycle. The more honest reading is that record profit is the sum of a long tail of older orders, service contracts, and price adjustments — with AI demand contributing at the margin and, more importantly, turbocharging the outlook for the next several years.
Third — and here is where I want to be careful — the profitability engine of turbine makers has a name: the Long-Term Service Agreement, or LTSA. When a utility or hyperscaler buys a gas turbine, the real margin often lives in the service contract that follows: maintenance, monitoring, spare parts, digital diagnostics. These contracts are high-margin, sticky, and recurring. Even if hardware margins compress under raw material costs and inflation, LTSA portfolios smooth the earnings path. That is what a mature turbine maker is today — less a manufacturer of boxes and more a subscription service for spinning metal. What is happening in this AI boom is that each turbine sold today becomes an annuity lasting twenty-plus years.
Pause on this structure, because it will look familiar to anyone who has spent time in crypto: sell the hardware at thin margin, earn the recurring yield in services. Validator infrastructure, staking pools, and even DeFi lending protocols share the same playbook — the same long-tail revenue logic, where retention beats acquisition and trust beats marketing. The difference is that in the turbine world, the maintenance schedule is enforced by physics, not by tokenomics. The blades wear out, and the service contract is the only thing between a spinning machine and a malfunction.
The Real Choke Point Is a Transformer
Now, the less-examined news that deserves more attention than the already-covered profit number: the choke point in this new AI-power economy may not be gas turbines at all. It is the humble electrical transformer.
I am not talking about the machine-learning kind. I am talking about the multi-ton units that step voltage up and down across the grid. Grid interconnection relies on large power transformers, and in many jurisdictions, their delivery time has stretched to two or even three-plus years — often longer than the turbine itself. I have seen reports of hyperscalers trying to pull forward transformer deliveries at any cost, because without these units, all the gas turbines in the world cannot get electrons where the compute actually lives. If you want the honest bottleneck analysis: the turbine is the visible spearhead of the AI power rush; the transformer is the actual gate. Behind the transformer, you will find switchgear, high-voltage cables, and cooling systems — each with its own queue.
This matters because it changes the strategy map. Companies that manufacture large transformers, or that have secured supply agreements, are effectively controlling a strategic bottleneck that no amount of venture capital can code around. You cannot compress the transformer delivery queue with a software update. You have to build physical capacity, which requires plant construction, skilled labor, and time.
This is the essential lesson of the AI energy story: at a certain scale, software's promise of zero marginal cost collides with the reality of finite physical infrastructure. I lived through the crypto mining arms race — watching miners locate next to hydroelectric dams and stranded natural gas flaring — and the pattern is unmistakable. The intelligence economy is not about to float free of the earth.
A Seller's Market for Power Machinery
The competitive dynamics are worth examining too, because they are changing. Siemens Energy's record profit is a symptom of a broader shift: large gas turbine manufacturers have moved into a seller's market, where the competitive focus has turned from price to delivery speed, manufacturing capacity, and service network depth. GE Vernova and Mitsubishi Heavy Industries are equally central to this story, yet the coverage of the AI-energy theme tends to treat them as footnotes. In a capacity-constrained market, whoever has spare production slots and a global service network wins disproportionate share.
More interesting, and largely underreported, is the power structure shift hiding inside the demand curve. Hyperscale cloud providers may begin buying power plant equipment directly from manufacturers, bypassing traditional engineering-procurement-construction (EPC) contractors. That would be a significant restructuring of the industry's value chain — and it mirrors something we have seen in blockchain: validators and staking pools buying hardware directly from manufacturers to cut out intermediaries. When the customer becomes sophisticated enough to order a complete power island the way it orders a GPU cluster, relationships change. Long-term capacity reservation agreements, or even equity investments by clouds into turbine makers, become logical next steps.
There are also regional wrinkles. Much of this AI-driven gas turbine demand is concentrated in the United States, where grid interconnection is slow and gas is abundant and relatively cheap. In Europe, the story is different — tighter gas supply, carbon pricing, and a stronger political push toward renewables and nuclear. In China, the energy mix is dominated by coal and grid expansion, so data center power demand takes a different shape entirely. A global headline about "gas turbines for AI" can obscure the fact that this is, to a large degree, a North American phenomenon driven by a specific grid and regulatory environment. Regional capacity and trade policy will shape who actually benefits.
The Centralization Problem Nobody Wants to Name
Now we reach the deeper argument. For someone whose career has been built on decentralization, the spectacle of hyperscale clouds and gigawatt energy procurement is troubling — not because the demand is illegitimate, but because of the concentration it implies.
Electricity, by its nature, is relational. Its generation, transmission, and distribution involve rights of way, cooling water, transmission corridors, and land-use approvals — all intensely local resources. Data center siting decisions in Northern Virginia, Ohio, or the desert outskirts of Phoenix are reshaping local grid loads and electricity prices. The same dynamic appears in the turbine market: only a handful of firms make the giant machines, each with limited production slots. Cloud giants can pay a premium and reserve capacity, which in turn squeezes everyone else out of the market — smaller utilities, municipal power providers, independent developers. In this context, "AI's thirst for power" is not a neutral physical phenomenon. It is a bidding contest that consolidates access to resources in the hands of the largest balance sheets.
During the Prague Consensus workshop series, one of the principles we taught was that trustless systems fail when physical resources are unevenly distributed. The same principle does not stop applying at the borders of AI. If the cost of intelligence is a wholesale electricity market where data centers bid up capacity and residential rates follow, then we are not merely building a new industry; we are quietly redistributing a shared resource toward the most capitalized actors.
That is a hard thing to say in a bull market. Nobody wants to be the killjoy at a party where the music is AI-generated and the drinks are paid for by venture capital. But my experience running the Reclaim peer-support network in 2022 — watching developers burned out by a sudden bull-to-bear reversal — taught me that the real victims of infrastructure scarcity are not the ones who profit from it. They are the ones who cannot command the resources to participate. That is why the phrase "build for humans, not just nodes" is not a slogan; it is a design constraint. It applies to protocol governance, and it applies to interconnection queues.
The Contrarian Reading: Maybe the Turbine Is the Moral Choice
Now let me steelman the other side, because I do not think the answer is "gas turbines are evil." The contrarian view is more subtle: the gas turbine boom might be the most rational available path to enable something genuinely transformative, and condemning it from a position of comfort in a wealthy European city would be a bit rich.
Look at the arithmetic honestly. Nuclear and renewables plus storage are the long-term endgame — everyone with a spreadsheet knows this. But small modular reactors are not yet commercially deployed at scale. Batteries cannot run a hundred-megawatt facility through day-long training runs. Wind and solar require grid upgrades that lag demand. In the interim, the fastest way to get reliable, dispatchable, 24/7 power to an AI facility is a gas turbine. If you believe AI can accelerate scientific discovery, drug development, or material science, then a bridge fuel for the AI buildout is not irrational — it might, in a narrow consequentialist sense, be a moral good.
There is also the efficiency trajectory to consider. The current scramble assumes an exponential demand curve, but history offers a caution: compute efficiency improves under price pressure. Model distillation, sparse architectures, and specialized ASICs could flatten the power curve sooner than doomsayers expect. The currently fashionable projections of AI electricity demand doubling every year are extrapolations, not laws of physics. If the demand curve bends, the gas turbine order boom could peak and fade within a handful of years — leaving a stranded asset problem very similar to the overbuilt fiber-optic networks of the early 2000s. Buyers of "AI power" stocks today should be asking which scenario is priced in: the endless boom or the brisk correction.
And the environmental contradiction deserves a full accounting. Natural gas is a fossil fuel. It burns cleaner than coal, but it still emits carbon dioxide, and upstream methane leaks add to its footprint. Simple-cycle gas turbines, used for peaking power, are less efficient and more carbon-intensive per megawatt-hour than combined-cycle configurations — a detail that rarely appears in the promotional material. We also do not know whether the turbines being ordered for data centers are operating as baseload generation or just as backup and peak shaving. That distinction dramatically changes the emissions and the economics.
The tech companies at the center of this narrative remain committed, at least in public, to aggressive climate pledges. The expansion of gas-powered data centers sits in direct tension with those pledges, and over time, carbon pricing or regulation could raise the fuel cost significantly. Hydrogen-ready turbines are in development — Siemens Energy and Mitsubishi Heavy Industries are both working on them — but today's hydrogen supply is nowhere near what a five-hundred-megawatt data center campus would require. The gap between the clean-energy story and the physical reality is exactly the kind of gap that produces stranded assets.
There is also a media narrative caveat. The source report, coming out of a crypto-focused outlet, is part of a pattern: the attention economy rewards stories that connect AI hype to speculative entry points. Especially in a bull market — whether crypto or AI or both — stories that can be converted into a buy-the-infrastructure thesis travel faster than careful analysis. To my own readers, I would say: be humble about your sources. Order growth and revenue recognition are not synonyms. "AI demand" and "excluding AI demand" are rarely separated in the deck. Read the footnotes. Education is the ultimate yield — that lesson from DeFi Summer's whitepaper translation work, when I spent weeks simplifying Aave's liquidation mechanics for non-technical users, applies just as much to energy stocks today. The person who understands the difference between orders and profits, between a headline and a balance sheet, will not be swept away by the next narrative tide.
What This Means for the Decentralization Ethos
There is a thread connecting this story to the broader value system I care about. If the scarcest resource in the intelligence economy is electricity, then whoever controls power infrastructure controls the future of computation — not just in a commercial sense, but in a civic one. Communities hosting data centers have a legitimate claim to a voice in how those facilities are sited, who receives the economic spillovers, and how the grid is managed. That is not an anti-growth position; it is a pro-governance position.
In my work with the EU regulatory task force on inclusive protocols, one theme kept surfacing: technical systems inherit the values of whoever controls the access points. Interconnection queues, transformer allocations, and turbine order books are access points. They deserve as much governance attention as any smart contract. We spent years building on-chain dispute resolution mechanisms to protect retail users; applying the same logic to energy infrastructure would be a natural extension. "Community first" does not mean no data centers. It means data centers that pay a fair community price, and a transparent process when local grid capacity is redirected.
So where does that leave us? The gas turbine boom is not a mirage, but it is not a prophecy either. It is the physical manifestation of a bottleneck, and bottlenecks attract both profit and distortion. The question that matters is whether the energy required for intelligence can be provisioned in a way that is fair, transparent, and accountable to the communities that host it. Otherwise, we will simply repeat the oldest pattern of industrial history: extracting local resources for remote benefit.
Build for humans, not just nodes. That line is easy to say in a conference keynote and hard to live when the humans in question are paying higher electricity bills or breathing near a new gas plant they had no voice in approving. And education — real education about how infrastructure actually works — remains the ultimate yield. I do not know whether gas turbine orders will keep growing for twenty years or stall in five. But I do know this: the future belongs not to those who bid highest for power, but to those who understand what it is for. The second bottleneck is not just electricity. It is comprehension. And comprehension is one bottleneck we can still choose to solve together.
