Koch’s $15B Edged Bet Shows AI’s Real Bottleneck Isn’t Chips
A data-centre developer may fetch more than $15 billion because AI has turned boring things—power, water and permits—into gold. The chatbot is the sexy bit. The infrastructure bill is where the money is.
A data-centre developer may fetch more than $15 billion because AI has turned boring things—power, water and permits—into gold.
The chatbot is the sexy bit. The infrastructure bill is where the money is.
A $15 billion reminder that software does not float in the cloud
Koch Inc. is reportedly exploring a sale of Edged, the data-centre developer it cofounded, in a process that has drawn multiple bids and could value the business at more than $15 billion. Koch Real Estate Investments is working with Goldman Sachs and Newmark to assess interest. Nothing is signed, and the talks could still fall apart. But that caveat does not make the number less revealing. ([news.bloomberglaw.com](https://news.bloomberglaw.com/mergers-and-acquisitions/koch-said-to-weigh-15-billion-sale-of-data-center-player-edged))
Edged is not selling a magical new foundation model. It is not a consumer app with a billion users. It is in the brutally physical business of finding land, arranging electricity, building facilities and stopping high-density computing equipment from cooking itself.
That may sound less glamorous than AI models. It is also exactly why it matters.
For the past few years, plenty of people have treated AI as a software story: buy GPUs, train models, sell subscriptions, ring bell, become rich. That is only half the picture. The other half is a nasty real-world queue involving substations, transmission lines, construction crews, cooling systems, water rights, planning approvals and local politics.
You can order chips. You cannot order a new electricity grid from Amazon Prime.
Koch’s reported move puts a big, fat price tag on that reality. A potential valuation above $15 billion says capital markets are prepared to pay handsomely for a company that has already worked out how to turn AI demand into functioning, financeable infrastructure. That is not a verdict that every data-centre deal is sensible. It is a verdict that the hard bits have become scarce. ([news.bloomberglaw.com](https://news.bloomberglaw.com/mergers-and-acquisitions/koch-said-to-weigh-15-billion-sale-of-data-center-player-edged))
Edged is selling the part of AI everyone would rather ignore
Koch says it created Edged with data-centre entrepreneur Jakob Carnemark after building expertise across land, power, construction and cooling. Its projects built or in development have included a 96-megawatt facility in Chicago, a 24-megawatt facility in Kansas City and a 168-megawatt facility in Atlanta. ([kochinc.com](https://www.kochinc.com/news/koch-real-estate-investments-is-helping-major-tech-companies-meet-the-demand-for-new-data-centers?ktm_campaign=ki_awareness_krei_edged_investment&ktm_channel=organicsocial&ktm_contentcategory=value&ktm_platform=sproutsocial&ktm_source=linkedin&utm_source=openai))
Those numbers matter because megawatts are the currency beneath the AI hype. A model demo can go viral overnight. A 168-megawatt campus cannot. It takes years of planning, capital and negotiations before a site can reliably feed that kind of load.
Edged’s pitch is not merely that it can build sheds full of servers. The company claims its closed-loop ThermalWorks cooling system uses zero water for cooling and supports high-density equipment. Koch has said the technology can handle densities of up to 200 kilowatts per rack. ([kochinc.com](https://www.kochinc.com/news/koch-real-estate-investments-is-helping-major-tech-companies-meet-the-demand-for-new-data-centers?ktm_campaign=ki_awareness_krei_edged_investment&ktm_channel=organicsocial&ktm_contentcategory=value&ktm_platform=sproutsocial&ktm_source=linkedin&utm_source=openai))
Its Mesa, Arizona facility, opened in April, is a useful example. Edged says the 36-megawatt site is built for AI training and inference, can support more than 400 kilowatts per rack with liquid cooling, and is expected to conserve more than 138 million gallons of water a year. The company says its portfolio-wide designed power-usage effectiveness, or PUE, averages 1.15. ([edged.us](https://edged.us/news/edged-us-opens-waterless-ai-ready-data-center-in-the-heart-of-arizonas-silicon-desert-tech-corridor?utm_source=openai))
That PUE figure is not a vanity metric. PUE measures how much extra energy a facility uses beyond the computers themselves. A theoretical score of 1.0 means nearly all energy goes straight to IT equipment. The closer you get to that number, the less power gets chewed up by cooling and other overhead.
When electricity is constrained, a more efficient facility is not just greener marketing. It can mean more usable compute from the same power allocation. In a market where power capacity is the bottleneck, that is commercial leverage.
The contrarian point: this is not really a data-centre story
The lazy take is that a possible $15 billion valuation proves investors have lost their minds over AI. Maybe some have. There is plenty of circular financing, heroic forecasts and expensive optimism sloshing around this sector.
But writing off every big infrastructure valuation as a bubble is equally lazy.
The more useful interpretation is that Edged sits at the intersection of three shortages: available power, buildable sites and acceptable cooling. AI did not create those constraints. It merely took a flamethrower to them.
A hyperscaler can have the money to buy Nvidia hardware and the customers to use the computing power. That still does not guarantee it can get a site online where it wants, when it wants, with enough electricity and without causing a local revolt over water consumption or grid pressure.
That is why the phrase “the cloud” has always been a bit of marketing nonsense. The cloud is warehouses, transformers, generators, fibre, pipes and enough electrical engineering to make your eyes bleed.
Koch itself describes the development problem in plain English: identify land, secure power and deal with the water requirements of conventional cooling. It says Edged can also draw on Koch capabilities for behind-the-meter power solutions, including on-site solar, battery storage and natural-gas generation to help address supply gaps. ([kochinc.com](https://www.kochinc.com/news/koch-real-estate-investments-is-helping-major-tech-companies-meet-the-demand-for-new-data-centers?ktm_campaign=ki_awareness_krei_edged_investment&ktm_channel=organicsocial&ktm_contentcategory=value&ktm_platform=sproutsocial&ktm_source=linkedin&utm_source=openai))
That does not eliminate risk. In fact, it highlights it. The further data-centre developers move into on-site generation and complicated power arrangements, the more they become infrastructure businesses carrying infrastructure-grade risks: regulation, fuel costs, utility negotiations, project delays and enormous upfront capital requirements.
Anyone valuing these companies purely as an “AI multiple” is missing the plot. Their economics depend on long contracts, creditworthy customers, disciplined project delivery and whether the power genuinely turns up when promised.
Why the $15 billion figure should make founders uncomfortable
If you are building an AI company, the Edged story should be mildly confronting.
It says the value in your industry may not sit where you think it does. Your clever interface may be copied. Your model provider may cut prices. Your distribution channel may change the rules. But the company that controls a permitted site with power, cooling and an operating track record owns something that cannot be replicated by a weekend of prompt engineering.
I have seen this in business repeatedly. People chase the visible layer because it is easier to explain at a barbecue. The real money often accrues to the constraint underneath it.
In spirits, it might be supply, distribution or shelf space—not the flashy bottle design. In AI infrastructure, it is increasingly the right to consume huge amounts of electricity in the right location without getting strangled by cooling, construction or community opposition.
That does not mean every founder should run off and build a data centre. Most should absolutely not. Capital-intensive businesses are a brilliant way to go broke if you do not understand the downside.
It means you should ask a harder question: what scarce input does my business rely on, and who actually controls it?
For many AI startups, the answer is compute. For software businesses, it may be distribution or proprietary data. For consumer brands, it may be trusted attention. For operators, it is often great people who can execute without needing a committee meeting to decide which colour the spreadsheet should be.
Find the constraint. Build around it. Or partner with the person who owns it before they realise how valuable it is.
The overlooked risk: infrastructure can become a crowded trade
There is a catch, and it is a big one.
The same conditions that make Edged attractive can encourage too much money into the sector. When investors see billion-dollar data-centre transactions, everybody suddenly discovers they have always been passionate about substations. That is usually when discipline starts packing its bags.
A data centre is not automatically valuable because it has “AI-ready” in the brochure. It needs contracted demand, credible counterparties, construction competence, connectivity, power economics and a clear path through local approvals. A speculative site with no practical route to power is just expensive dirt with a PowerPoint deck.
The reported Edged process is therefore interesting precisely because the company is not being framed as a blank-cheque land bank. It has an operating model built around high-density workloads, waterless cooling and existing projects in major US markets. That is a far sturdier proposition than buying acreage and hoping an AI tenant turns up with a cheque book. ([kochinc.com](https://www.kochinc.com/news/koch-real-estate-investments-is-helping-major-tech-companies-meet-the-demand-for-new-data-centers?ktm_campaign=ki_awareness_krei_edged_investment&ktm_channel=organicsocial&ktm_contentcategory=value&ktm_platform=sproutsocial&ktm_source=linkedin&utm_source=openai))
The distinction matters for investors. In hot markets, good assets and rubbish assets both get called “strategic.” Only one of them survives when financing costs rise or customers delay expansion.
What this means for you
First, stop confusing attention with value. The loudest AI company is not necessarily the one with the best business. Look for the bottleneck it depends on—and work out who owns it.
Second, if you are buying AI tools for your company, do not get hypnotised by the demo. Ask what happens to your cost base if usage grows tenfold, your provider changes pricing, or compute becomes constrained. Build an exit option before you need one.
Third, if you are an investor, separate infrastructure with contracts and operational proof from infrastructure with a sexy story. Revenue quality, power access, project economics and customer concentration matter more than the word “AI” appearing 47 times in a pitch deck.
Fourth, if you are a founder, build a moat that gets stronger as the market gets crowded. That could be proprietary data, embedded workflow, distribution, trust, a regulatory edge or a supply relationship. It does not need to be glamorous. It needs to be hard to replace.
And finally, remember what the potential Edged deal is really telling us: AI is not removing the laws of business. It is making the old ones more expensive. Scarcity still wins. Execution still wins. And the person who controls the unsexy bottleneck usually gets paid before the person giving the keynote.