The week's most interesting AI bet was not on a software company.
It was on a 606-megawatt gas plant outside Houston.
It was not alone. Investors also acquired a company operating more than 2,000 mobile generators, agreed to buy an electrical-services business, and combined two building-commissioning platforms.
The loud question is which AI model will win.
The useful question is what every AI system still needs before it can scale.
Power. Permits. Qualified people. Buildings that work.
The party is upstairs.
The bottleneck is in the basement.
In this issue
Five deals worth knowing before Monday
Why being near a bottleneck is not the same as controlling it
What Airtable's two prices say about optionality
Three questions for underwriting scarcity
The Tape
LS Power agreed to acquire the Brazos Valley Energy Center: The 606-megawatt natural-gas plant outside Houston is expected to change hands in the fourth quarter, subject to regulatory approval. Why it matters: operating capacity can be bought now. New generation can take years to build. Source
KKR agreed to acquire Medicover's India hospital business: The announced enterprise value is approximately €1.2 billion. Why it matters: scale is the thesis. Hospitals expose the trade-off: revenue can grow with the footprint while complexity grows with it. Source
I Squared agreed to acquire Cella: Five Indonesian logistics and cold-storage assets will seed a platform intended to grow organically and through M&A. Why it matters: the assets are the starting point. The repeatable operating and acquisition system is the platform thesis. Source
Leeds Equity completed the sale of OptionMetrics: Equality Asset Management is the buyer, and founder and CEO David Hait is making a significant reinvestment. Leeds invested in 2021 to expand products, datasets, distribution, and M&A; by exit, OptionMetrics had acquired Woodseer and its portfolio also included OptionStrat. Why it matters: Equality is inheriting a specific product-expansion test: use IvyDB's 30-year historical core to reach more markets and users without weakening the data quality that earns adoption. Current sale · 2021 investment · Company context
P&G agreed to acquire Thorne from L Catterton in a deal reported at $3.8 billion: L Catterton took Thorne private in October 2023 at an approximately $680 million transaction value. Thorne now says it reaches more than seven million consumers, tens of thousands of healthcare professionals, and more than 100 professional sports teams from a vertically integrated South Carolina manufacturing base. Why it matters: the reported price is roughly 5.6x the 2023 transaction value. P&G is not just buying supplements; it is buying the practitioner and athlete trust that lets Thorne occupy premium wellness. The test now is whether P&G can widen distribution without weakening the credibility that supports the premium. Thorne announcement · Reported price · 2023 completion · P&G FY2026
The One Thing
A bottleneck can make every company around it busier.
It does not give every company power.
LS Power is buying operating generation. Gravity, acquired by Firefly Capital, operates more than 2,000 mobile power units across 13 states, according to the company. Veritas is acquiring Saber Power Services, which helps engineer, build, and maintain electrical infrastructure.
LS Power owns capacity. Gravity owns a fleet and the ability to move power to the point of need. Saber sells the technical execution required to make electrical infrastructure useful.
A service provider can grow revenue and costs together. A company that controls scarce capacity, permission, qualification, or workflow may gain leverage as the queue grows.
That is why “AI infrastructure” is too broad to be an investment thesis.
The same test applies downstream. Facility Grid and PingCx work in building commissioning. Trinity works in compliance and technical consulting. Command Alkon provides software for construction-materials workflows.
The releases establish proximity to a constraint, not control of it. The underwriting question is whether that position produces retention, pricing power, or capital efficiency, or merely more work.
Even genuine control can be a bad investment at the wrong price. A high entry valuation, leverage, new capacity, or customer insourcing can erase the return while the scarcity remains real.
So the thesis is not “buy the physical layer of AI.”
It is this:
Being near a bottleneck creates demand. Controlling it can create leverage.
The work is determining which is which, then deciding whether the price already assumes the answer.
From the Deal File
Airtable and the optionality tax
Airtable is the inverse side of the argument.
It had cash. Its announced sale to Bending Spoons asks whether that cash ever became durable strategic control.
The announced terms produced two different values: one for the equity and one for the operating business after cash.
Capital buys time. Time should improve your options. When it does not, the gap eventually appears in the exit.
The Operating Edge
If I were underwriting a company exposed to the AI buildout, I would ask three questions.
What is actually scarce? Not what is popular. What takes years, permission, capital, qualification, or trust to reproduce?
Who controls it? Does the company own capacity, access, certification, or the customer workflow? Or does it add labor whenever demand rises?
Does control show up in the economics? Look for pricing power, utilization, retention, recurring work, or capital efficiency. Then test what those gains cost in labor, maintenance, working capital, leverage, and the entry price.
The thesis weakens if capacity arrives quickly, customers bring the capability in-house, or growth requires proportional additions of people and capital. It fails if today's price already capitalizes the entire future.
AI demand can be real while AI-adjacent returns disappoint. If the thesis needs every AI forecast to work, the asset carries macro exposure, not a durable moat.
In partnership with
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
Sources
If you saw a deal, exit, or operating pattern worth studying, reply and send it my way.
Nick
Founder, The Multiple
The Multiple is an independent publication about private equity, exits, and the operating work behind enterprise value. Analysis is for informational purposes only.
If someone forwarded this to you, subscribe at themultiple.fm.