Positive
Only logged in members can reply and interact with the post.
Join SimilarWorlds for FREE »

May 2026 Inflation Is 10.5% Higher Than April 2026

You can no longer blame this on Biden.

You can't blame it on LA taking too long to count ballots.

Some mote bad news: your 401K is going into financing the debt on data centers and 9 out of 10 data centers are going to be out of business and what you're doing is enabling Musk to become a $ Trillionaire. Enjoy the UFC fights on the White House lawn. Much like Rome enjoyed gladiator fights.
Top | New | Old
Khenpal1 · M
Wall Street analysts have warned that the rapid depreciation of data center hardware (such as chips that become obsolete in 2-4 years) against massive construction costs makes data center financing highly dependent on long-term, sustained revenue from AI.Which brings the central questions here into focus: Will the revenues materialize to service all of it? And, down the line, will the bondholders get paid?
To all appearances, the financing only works if several aggressive assumptions hold simultaneously.

OpenAI needs this much compute long term — and has or will have the revenue to pay for it, from a company that has never turned a profit. Oracle can service tens of billions in debt while its cash flows are, arguably at least, already strained. And the underlying hardware doesn't become obsolete before the debt is paid down, which is a relevant risk given that AI chips turn over roughly every two to three years, even as the bonds carrying this debt have maturities measured over much longer periods.

None of these assumptions are necessarily unreasonable. But all of them have to be right, at the same time, for the math to work — just like many other 144A bonds. The kind that could be finding their way into ordinary Americans’ retirement savings.

And whether this one deal becomes a success story or a cautionary tale, it’s very difficult to think individual investors allocating their savings to bond funds actually want to be holding this kind of debt. Which makes the details revealing and eyebrow-raising, just not for the reasons the players involved might hope.

Put simply: If the most conservative part of your portfolio turns out to hold potentially risky AI debt, then it's arguably not the most conservative part of your portfolio.
Northwest · M
@Khenpal1
Will the revenues materialize to service all of it?

No.

The revenue is set by me, the CONSUMER. Let's say I want an apple, and you tell me that the apple is going to be $1,000. I'll switch to planting my own apple tree.

To recoup the spend on data centers, all these companies will need to charge 10X what I can afford.

The in-depth studies now show that this is the case, so the race AI companies are in, is really how to kill each other, because only one will be left standing.
Khenpal1 · M
There is a severe global shortage of grain-oriented electrical steel (GOES) and copper windings. Historically, the U.S. manufactured only 20% of its own large transformers, relying heavily on imports from places like Mexico, China, and Thailand. The United States' domestic manufacturing capacity was hollowed out over decades, forcing the industry to outbid utilities for limited factory slots.Tech and data center operators are now securing factory production slots before even securing physical site control.Companies are refurbishing old, decommissioned equipment to bridge the gap until new units arrive. Firms are exploring moving to higher direct-current (DC) architectures or deploying on-site solar and battery storage to bypass long-term interconnection waits.Copper loss in a transformer occurs because the copper (or aluminum) wires used for the primary and secondary windings have natural electrical resistance. When electrical current flows through these wires, this resistance generates heat, which causes energy to be lost in the form of
(Joule heating). The AI & Data Center Boom: Hyperscale data centers demand massive amounts of power, causing a ripple effect that requires extensive new substations, cooling systems, and power transformers. An AI training data center can have a copper intensity of up to 47 tons per megawatt installed.New copper deposits are becoming increasingly difficult and expensive to mine due to declining ore grades and 15- to 20-year lead times from discovery to commercial production. Manufacturers are re-engineering the world to run on aluminum wherever physically possible. While aluminum requires larger conductors, its light weight and lower cost make it an appealing bypass.Copper projects are becoming increasingly expensive to build due to inflation, declining ore grades, permitting delays, environmental requirements, and infrastructure costs.We Don’t Have a Transformer Shortage. We Have a System Design Problem.The economics of clean energy are no longer the constraint—they're the catalyst. But there's a stubborn, very physical bottleneck that doesn't respond to price signals nearly as fast: the supply of critical grid equipment. Data centers can't energize new capacity, utilities can't complete interconnections or upgrade substations, and both are increasingly competing for one unglamorous but indispensable asset—the transformer.

U.S. demand for power transformers has surged 116% since 2019. Lead times average 128 weeks. The supply shortfall for power transformers is around 30%. Meanwhile, in-service transformers are approaching the end of their service lives, so the replacement burden on top of new-build demand is enormous. NREL projects transformer demand could reach 260% of 2021 levels by 2050. These aren't alarming statistics in isolation — they're alarming because they're all moving in the same direction at the same time.

When you hear those numbers, the instinct is to say "Build more factories!" And the major OEMs have obliged. Hitachi Energy committed $1.5 billion to transformer manufacturing in 2024, and Siemens, GE Vernova, Hyosung Hico, and WEG are all ramping spending too. But while the dollar amounts look impressive, more factories aren't solving the core problem — how customers procure transformers in the first place.

The development landscape has fundamentally changed. Five years ago, a developer might have managed a handful of projects. Today, the top five solar developers are juggling more than thirty. Hyperscalers are committing to develop power across the country at a pace that would have seemed absurd a decade ago — the Microsoft and Brookfield deal for 10.5 GW of capacity by 2030 is just a sign of the times. These aren't utilities placing one large order every eighteen months. They're operating at a scale where a single delayed piece of equipment creates stranded capital and cascading schedule risk across an entire pipeline. And despite that scale — or perhaps because of it — developers still aren't securing their transformers early enough in the process.

The size of the gap becomes clearest when you look at it by segment. The large generator step-up and substation transformer market in the U.S. represents roughly 30–50 GVA of annual capacity. In the medium-voltage padmount segment, the market runs approximately 150–300 GVA per year. Demand is outpacing supply across both, and the traditional procurement system wasn't built to handle either at the current pace.

When a developer waits 128 weeks for a transformer and site conditions or interconnection requirements shift midway through — which they almost always do — the old procurement model has no good answer. You're either locked into a spec that no longer fits the project, eating a redesign timeline, or force-fitting equipment that wasn't built for the updated scope. Neither is fast. Neither is cheap. Both destroy margin and delay interconnection in a market where queue positions are already scarce and hard-won.

Fixing this requires rethinking the system, not just the supply chain. Procurement structures need to change — onboarding new vendors is opaque, blocking flexible suppliers from ever reaching decision-makers. Design needs to be modular, so that when projects evolve through permitting, financing, and site changes, specs can adapt without losing a queue position at the manufacturing plant. And customer relationships need to shift from transaction-by-transaction to portfolio-level partnerships that reflect how energy development actually works today.

A lot of companies are adding capacity. Few are rethinking the procurement system behind the broken ecosystem. What's notable about newer models like Ayr Energy is precisely that shift — from project-level to portfolio-level customer engagement, using standardization and modularity as the forcing function in exchange for much faster equipment access. It's an approach that aligns with how projects are actually developed, from design and quoting through to how long-term customer relationships are structured. Ayr recently announced over $500 million in contracts supporting more than 20 GW of planned U.S. power capacity.

A critical part of that strategy is unlocking supply the traditional procurement system has never accessed. India's transformer industry has built total manufacturing capacity of approximately 400 GVA — with utilization rates of only 60–70%, leaving over 100 GVA sitting idle annually. Securing large generator step-up, substation, and medium-voltage padmount transformers from Indian manufacturers provides an essential bridge for a domestic market that can't wait years for new factory capacity to come online.

Energy demand isn't going away. Every gigawatt of solar, wind, storage, and data center capacity that gets permitted and financed needs equipment to connect it to the grid. The transformer bottleneck is real — but it's as much a failure of procurement design as it is of manufacturing supply. Solving it won't come from factories alone. It will come from fundamentally changing how projects, capital, and equipment come together — before the next 128-week clock starts ticking.
Waveney · M
You can no longer blame this on Biden

But MAGA will nonetheless continue to do so
trollslayer · 51-55, M
@Waveney Or Tim Walz, somehow.
Khenpal1 · M
The constraint is no longer computing chips, but basic physical infrastructure. Severe shortages of electrical components like transformers, switchgear, and backup batteries are stretching timelines, making it impossible to bring massive new server farms online.Power Grid Limitations: Utility grids simply cannot support the immense energy loads these massive facilities require. Grid operators are increasingly pushing back against the speculative "phantom" load requests made by tech companies. Communities and environmental groups are organizing pushbacks against the resource-heavy facilities, citing soaring local utility bills, dwindling water supplies, and noise pollution. In the first quarter alone, over 20 proposed sites were canceled due to local zoning and permitting rejections.Everyone’s talking about the AI boom. The trillion-dollar investments. The model releases. The CEO manifestos about how AI will reshape civilization.

Nobody’s talking about transformers.

Not the cool kind. Not the Optimus Prime kind. I mean the boring, heavy, industrial electrical transformers that sit in substations and convert high-voltage power into something a data center can actually use.

Because right now, those transformers are the single biggest bottleneck in the entire AI industry. And they’re breaking everything.According to Sightline Climate, the market intelligence firm that tracks data center construction across the U.S., approximately 12 gigawatts of data center capacity was announced for completion in 2026.

How much is actually under construction? About 5 gigawatts. One-third.

Bloomberg reported it first. Then Tom’s Hardware, TechSpot, Futurism, and Fortune picked it up. The consensus: roughly 30–50% of planned U.S. data centers for 2026 will be delayed or outright canceled.

Let that sink in. Alphabet, Amazon, Meta, and Microsoft are expected to spend more than $650 billion on AI infrastructure this year. Six hundred and fifty billion dollars. And close to half of the physical infrastructure that money is supposed to build? It’s stalled. Or dead.

This isn’t a minor scheduling hiccup. This is a structural mismatch between the ambition of the AI buildout and the physical reality of actually building things in the real world.I always assumed the bottleneck for AI would be GPUs. Or talent. Or maybe even data. Something sexy and high-tech. Something you’d read about in a TechCrunch headline.

Nope. It’s transformers, switchgear, and batteries.

These are the unglamorous pieces of electrical infrastructure that every data center needs — not just inside the building, but in the surrounding grid. Without them, you can’t deliver power. Without power, you’ve got a very expensive warehouse full of servers that don’t turn on.

Here’s the part that made me pause: before 2020, a high-power transformer in the U.S. had a delivery lead time of 24 to 30 months. Today? That wait can stretch to five years. Five years to get a single transformer delivered.AI data center build cycles are under 18 months. Do the math. You can’t build a data center in 18 months when the transformer it needs won’t arrive for 60.

And it gets worse. The U.S. doesn’t manufacture enough of these components domestically. China accounts for over 40% of U.S. battery imports and roughly 30% in certain transformer and switchgear categories. Imports of high-power transformers from China surged from fewer than 1,500 units in 2022 to more than 8,000 in 2025.

So the country that’s simultaneously trying to restrict Chinese access to AI chips… is deeply dependent on Chinese electrical components to build the AI data centers those chips go into.

 
Post Comment