Our Energy Coverage Gets A Power-Up
DISCLAIMER: This note is intended for US recipients only and, in particular, is not directed at, nor intended to be relied upon by any UK recipients. Any information or analysis in this note is not an offer to sell or the solicitation of an offer to buy any securities. Nothing in this note is intended to be investment advice and nor should it be relied upon to make investment decisions. Read our full disclaimer, here.
Introduction
by Alex King, CEO, Cestrian Capital Research
We provide three types of investment research here at Cestrian, which can be used separately or, better, in combination.
- Technical analysis or the Art Of Drawing Not Just Any Old Lines, and
- Algorithmic analysis otherwise known as the Endless Hunt For The Hidden Markov Model, and
- Fundamental analysis - comprising sector understanding and financial insight. Our sector knowledge comes not from an LLM nor a human cosplaying as an LLM, aka a finance professional, but instead from actual sector experts with actual experience of a senior career in the field. Yes I know. In biotech for instance we have Nathan Brinkman, who has generated some wonderful opportunities for our Inner Circle members, in semiconductor, systems and software we have many folks who have entered the chat in recent years, and now in energy I am delighted to welcome Preston Schultz. He sets out his stall below, in advance of coverage proper beginning in our Inner Circle service.
And Now The Main Event
by Preston Schultz
I am a renewable energy developer by trade. I spend my working life assembling land, fighting interconnection queues, negotiating with utilities, and arguing with county planning departments. I have worked projects from initial land lease through to commercial operation across solar, storage, and increasingly the full energy campus stack — gas, renewables, and data center co-location together.
That background means my read on the energy sector is not theoretical. It is built from watching how power actually gets built, permitted, and connected to the grid, which turns out to be directly relevant to some of the most interesting investment questions of the moment. I am not going to pretend to have deep O&G expertise — XOM and CVX are not my world. Where I think I can add genuine value is at the intersection of power infrastructure, renewables, storage, and the AI/data center buildout.
The Core Thesis: This Is a Physics Problem
The dominant narrative has been "AI needs power." That is true but incomplete. The more actionable version is: AI needs power that can be delivered, on a timeline that matters, to locations that will permit it. Those three constraints — deliverable, timely, permittable — disqualify most of the obvious answers and make the real opportunities harder to find, more durable when you find them, and still underappreciated by a market mostly buying the narrative rather than the physical reality.
The numbers support this. Jefferies published a data center supply/demand model in early June that I find credible. Their bottom-up view: GPU and XPU chip shipments from NVDA, AVGO, MRVL, AMD, and others imply ~19 GW of North American AI power demand in 2026 alone, against ~10.3 GW of new data center capacity actually expected to come online. The cumulative unmet gap reaches 27 GW by 2027 and 50 GW by 2028. The industry exited 2025 with a 12 GW deficit between leases signed and capacity delivered — triple the 4.3 GW gap in 2024.
This is not a demand problem. Hyperscaler capex is tracking toward ~$770B in 2026, up 74% year-on-year and nearly 5x the $156B deployed in 2023. Combined cloud backlog hit $2T in Q1 2026, growing 3x faster than capex. The problem is purely physical: you cannot build data centers faster than you can pour concrete, wire switchgear, procure cooling, and get an interconnection agreement. And right now every one of those steps is constrained simultaneously.
The Physical Bottlenecks — And Who Profits From Each
EPC/Labor is the binding constraint in 2026 at a ~10.4 GW ceiling. Quanta Services ($PWR) reported record Q1 2026 revenue of $7.87B (+26% YoY) with record backlog of $48.5B; MasTec ($MTZ) posted record Q1 revenue of $3.83B (+34% YoY) with an 18-month backlog up $4.4B YoY. The pattern is consistent: backlog grows faster than revenue conversion, meaning labor — not contracts — is the actual bottleneck.
Primoris ($PRIM) is a preferred value play here, still discounting prior execution concerns while the narrative shifts toward gas generation and power delivery. SOLV Energy ($MWH) is worth a look as a newer entrant in this space — a large-scale solar and storage EPC that has been expanding its data center-adjacent power work and whose self-perform labor model gives it a structural advantage in a market where subcontractor availability is becoming the rate-limiting factor. Both $PRIM and $MWH sit in the part of the market where the work is guaranteed for years but the valuation has not yet caught up with the backlog.
Cooling becomes the binding constraint from 2028 onward. Vertiv ($VRT) reported Q1 2026 orders up 71% YoY but raised full-year guidance only modestly — manufacturing throughput, not demand, is the governor on growth. That gap between order growth and revenue guidance tells you everything you need to know about the supply tightness. Carrier Global ($CARR) is worth watching here on relative value versus Trane Technologies ($TT) despite comparable data center cooling exposure.
Power transformers have 24–36 month lead times and compete directly with grid expansion and renewables for the same manufacturing slots. GE Vernova ($GEV) reported Q1 2026 orders of $18.3B (+71% organically) with data center equipment orders of $2.4B in Q1 alone — more than all of FY2025 combined. Siemens Energy ($ENR) reported record Q2 FY2026 orders of $17.7B with total backlog of $154B and a book-to-bill of 1.72x. Forgent Power Solutions ($FPS) is the higher-beta pure-play in this category — ramping transformer supply from a ~$1.5B run-rate toward $5B by end of decade, with the capital expenditure build-out for manufacturing concluding mid-2026.
Power availability is the structural backdrop. The US interconnection queue holds over 2,000 GW of proposed generation with a median queue-to-COD of five years nationally and eight years in PJM. Natural gas in the queue surged 86% YoY to 253 GW, reflecting data center demand for dispatchable firm baseload. Regulated utilities building new generation as rate base are an interesting play. NiSource ($NI) in Indiana is the cleanest example — its innovative GenCo structure lets it build generation for Amazon and Alphabet and earn premium regulated returns.
The Solar Thesis — A Contrarian Play Worth Taking Seriously
Here is something that cuts against the current policy narrative: solar may be one of the most realistic near-term solutions to the data center power problem, and the market is not fully pricing that.
The logic is this. A 200–500 MW behind-the-meter solar plus battery energy storage system (BESS) stack can be operational in 12–24 months on land that solar developers already control. You run the data center on solar+BESS first, avoid the interconnection queue entirely, layer in gas generation as turbines become available, and eventually tie to the grid in year 3–5 when interconnection clears. Solar is the bridge technology that lets a hyperscaler break ground in 2026 rather than waiting until 2030 for the full power stack to be ready. Google recently struck a deal for a 30 GWh Form Energy iron-air battery project to power a data center in Minnesota — exactly this construct in practice.
The land optionality piece reinforces this. Developers who have been assembling 5,000–12,000 acre blocks for solar projects happen to have land that overlaps with fiber networks and gas pipeline corridors, which makes them uniquely positioned to develop full energy campus stacks over time. The solar land play and the gas generation play often end up being the same land play. From a policy optionality standpoint, solar also benefits asymmetrically: the behind-the-meter logic works under current policy constraints, and it benefits further under any administration change.
$FSLR is the obvious public name here. I am not a natural FSLR bull — cadmium telluride has efficiency and local-acceptance headwinds, and the stock took a predictable hit when the current administration dialed back the IRA domestic content adder. But FSLR has always played in utility-scale solar so the data center procurement bridge is credible, and they have shown they know how to reposition around the prevailing demand narrative. The chart post-2025 bottom is potentially interesting as a trade. Not a conviction hold, but not without merit either.
Bloom Energy and the Fuel Cell Question
$BE deserves its own section because it captures the near-term data center power story more directly than almost any other name — and because it keeps coming up in community discussion for good reason.
The basic insight is physical: connecting a new AI data center to the power grid typically takes 3–7 years of regulatory approvals, utility negotiations, and infrastructure construction. Bloom Energy's solid oxide fuel cells can be deployed at a data center site on a meaningfully compressed timeline — months rather than years. If you believe AI buildout is accelerating and the grid genuinely cannot keep up, a company that can deliver firm, dispatchable, on-site power fast has structural pricing power.
The Oracle data center deal was the visible catalyst that brought BE into the wider market's frame. What I find more credible than the narrative, though, is the installed base argument: Bloom has hundreds of operational fuel cell systems at industrial and commercial sites built up over years. That means real manufacturing relationships, real EPC execution track record, and real customer references that a startup pivoting into this space cannot replicate. That is a genuine moat. The question I watch is whether Bloom can scale manufacturing fast enough to match the demand narrative — the bottleneck going forward is production capacity, not customer interest. I would want to see manufacturing scale evidence before treating this as a straight compounder rather than a momentum story.
The broader fuel cell space has other names worth knowing. Plug Power ($PLUG) has well-documented execution issues and remains speculative — the technology thesis is sound, but the business has struggled to translate it into reliable earnings. Cummins ($CMI) is a more durable business straddling backup generators and hydrogen fuel cells; their Power Systems segment posted record Q1 2026 revenue of $1.96B (+19% YoY) driven by data center backup power demand, and they are scaling hydrogen electrolysis manufacturing. Less exciting than BE as a story, but meaningfully less binary as a risk.
The Creative Plays
These are names that follow directly from the infrastructure-over-applications logic but are less obviously in the standard "AI trade" framing that most investors are running.
**$COHR and $LITE — optical networking inside the box.** As AI clusters scale to thousands of GPUs operating in parallel, the interconnects between chips and between servers become a critical bottleneck that gets less attention than the chips themselves. Coherent Corp. and Lumentum make the transceivers, optical amplifiers, and components that run inside and between data centers. Leopold Aschenbrenner — whose Situational Awareness fund built a concentrated picks-and-shovels infrastructure book starting in late 2024 — entered LITE around $40 (now ~$885) and COHR around $60 (now ~$318). The thesis is intact because AI scaling laws imply dramatically more intra-cluster data movement, not less. Jefferies covers Credo Technology ($CRDO) in a similar vein, flagging early-innings AEC adoption across XPU and CPU platforms with a meaningful ASP uplift from 1.6T AEC ramp in late 2026/early 2027.
Bitcoin miner → AI HPC conversions: this isn’t a new theme, but I think it has plenty to run yet. $CORZ, $WULF, $CIFR, $HUT This is the most creative play in the energy/data center nexus from where I sit. Bitcoin miners have two things that every AI data center developer desperately wants and cannot easily get: energized sites and established grid connections. When you are facing an 8-year interconnection queue in PJM, a miner who already has a signed interconnection agreement and a working substation is sitting on an asset worth multiples of their market cap. Core Scientific ($CORZ) was the pioneer — the first to pivot with a 590 MW development for CoreWeave, and it still trades at a discount to peers despite having the largest powered land bank in the group. TeraWulf ($WULF) is converting BTC mining land sites into turnkey data centers for neoclouds and hyperscalers; management's background in power utilities is exactly the right DNA for navigating utility negotiations and interconnection. The risk is straightforward: BTC mining economics are volatile, conversion timelines slip, and these are not quiet businesses. But the underlying energized-site asset value is real and does not require any view on bitcoin to appreciate.
What I Am Cautious On
Enphase Energy $ENPH: Residential and small commercial microinverters — almost the opposite design philosophy from what a 100 MW data center power stack requires. Their new IQ Solid-State Transformer announcement is a Bloom-style pivot attempt, but Bloom had an actual industrial installed base to bridge from. NEM 3.0 and AB 205 in California, plus the expiry of the federal residential clean energy credit at year-end 2025, continue to pressure the core business. Show-me story for a while.
”Not in my backyard” (NIMBY) risk broadly — an underappreciated headwind that I want to name explicitly because I see it on the ground. National support for data center construction has reportedly collapsed from 65% to 36% in thirteen months. Over $156B in data center projects were blocked or delayed in H2 2025. Over 300 state-level bills were filed in the first six weeks of 2026. This does not stop the buildout but it slows it, raises costs, and redirects capital toward low-density, constructive-regulatory geographies — Texas, Wyoming, Kansas, Indiana, Louisiana, Oklahoma. The xAI Colossus situation near Memphis — running unpermitted reciprocating engines in close proximity to residential communities — is the cautionary tale for what happens when you move fast and ignore the community relations side. Regulators noticed, and they will keep noticing.
The Bigger Picture
The Aschenbrenner organizing framework — electricity, compute, networking — is the right lens for this whole space. Where I would add from the development side: the land and permitting stacks are the longest lead-time items of all, and the developers who have already assembled large, gas-adjacent, fiber-adjacent, grid-adjacent land banks are sitting on optionality that does not show up on a DCF. That is usually where the interesting money gets made — before the asset gets a Bloomberg ticker and a sell-side initiation.
The constraint is not going away. AI chip-implied power demand exceeds new North American supply additions by 1.9x in 2026, 3.2x in 2027, and 4.8x in 2028. The companies that solve pieces of that constraint — whether through fuel cells, gas turbines, optical networking, land assemblage, or EPC execution — have genuine pricing power that is structural, multi-year, and not policy-dependent.
*Preston Schultz | June 2026*
DISCLOSURE: The author may hold long and/or short positions in any stocks and/or ETFs named in this note.