Four demand engines are firing at once and they all draw on the same finite pool of physical inputs. That overlap — not any single application — is what creates durable, multi-year bottlenecks. The investable edge is in the choke points: the input or piece of equipment that every builder needs and that cannot be scaled quickly.
US data-center power demand runs ~31 GW (2025) → 66 GW (2027), rising from ~4% to 8.5% of peak grid. Each campus is a compound draw on power, copper, transformers, cooling, and water.
Missiles, drones, missile defense, and a Navy shipbuilding push (JPMorgan floated a $1.5T financing initiative) all consume rare-earth magnets, specialty steel, and secure chips.
Fab capacity expansion pulls wafer-fab equipment, specialty gases, ultrapure water, and enormous baseload power into the same regions competing with data centers.
Grid, EVs, and industrial load compound the copper and electrical-equipment deficit. S&P flags a widening structural copper shortfall driven by AI + defense.
The scarce things aren't the chips — they're what lets a chip run: firm power (gas turbines & nuclear), grid iron (transformers, switchgear — Goldman notes >40% of DC capacity risks delay on equipment/labor), copper, rare-earth magnets (now a weaponized supply chain after China's 2025–26 controls), and cooling + water. Own the bottleneck, not the headline.
Every name below is profitable or strategically underwritten, positioned in a genuine bottleneck. The Value column is my read of price vs. business quality today — click any header to sort. "Conviction" blends bottleneck strength, balance-sheet solvency, and margin of safety (higher = better risk-adjusted setup on a value lens).
| Ticker | Bottleneck | Price | P/E | Mkt Cap | Value | Conviction | Entry |
|---|
P/E is trailing GAAP from live feed; negative or n/m marked accordingly. Market cap in $B. Conviction is my qualitative score (1–10), not a model output.
You can't run a GPU you can't power. Firm, dispatchable generation and the iron that moves it are the tightest link in the chain. Gas turbines and grid equipment have multi-year backlogs; the equipment makers have pricing power the hyperscalers don't.
Hyperscalers want 24/7 carbon-free baseload, and nuclear is the only source that delivers it at scale (Microsoft–Three Mile Island, Meta–Oklo). The investable spectrum runs from cash-generative uranium/utilities to pre-revenue SMR stories. Value discipline matters most here — the fuel-cycle names carry the theme's richest multiples.
The physical supply layer under everything above. Copper is the connective tissue of electrification (S&P flags a widening structural deficit); diversified majors give you copper + iron + critical-minerals exposure in one solvent, dividend-paying vehicle; lithium is the battered battery-metals leg; and royalty/streaming is the capital-light, lower-risk way to own the whole complex. This is the cheaper, more cyclical corner — real value exists, but so does commodity beta.
One-ticker exposure to copper + iron ore + critical minerals, with the balance sheets and dividends to survive the cycle. Gold majors add a monetary/safe-haven leg and screen cheap here.
Lithium is deep in a price trough — cheap on assets, little-to-no earnings today. A contrarian, higher-beta bet on the storage buildout that firms intermittent power for data centers. Size for the volatility.
These firms finance miners in exchange for a cut of production — commodity upside and exploration optionality with far less operating and cost-inflation risk. They earn premium multiples for a reason: tiny headcounts, fat margins, and diversified asset books. The quality-compounder corner of mining.
China's 2025–26 export controls turned NdFeB magnets into a national-security chokepoint — they're in F-35s, submarines, Tomahawks, drones, and every EV/wind/robotics motor. Washington responded with a landmark deal that de-risks the economics of the sole US scaled player. This is the purest "policy tailwind" vertical in the map.
$110/kg NdPr price floor for 10 years (≈2× market, DoD covers the gap) · DoD takes Series A preferred + warrants to ~15% (potential largest shareholder) · 100% offtake of 7,000 MT/yr new magnet capacity + $140M/yr earnings guarantee · "10X" facility scales capacity 3,000 → 10,000 MT. This transfers commodity-price risk off the equity — a structurally different setup than a normal miner, even at a rich headline multiple.
Rack densities have blown past air cooling — liquid and direct-to-chip are now mandatory, and that pulls in water treatment, cooling chemistry, and thermal hardware. Ecolab's CoolIT acquisition put a water-chemistry major squarely into the DC cooling supply chain; Xylem sees AI turning into water demand.
The full picks-and-shovels stack under every AI chip — five distinct sub-layers, each its own chokepoint: lithography (ASML, sole EUV maker), deposition/etch/implant (AMAT, LRCX, ACLS), process control (KLAC, ONTO, CAMT), test (TER), design software (SNPS, CDNS — the EDA duopoly), and consumable materials (ENTG). Deeper moats than almost any vertical here — but AI enthusiasm has pushed nearly all of them rich, so entry discipline matters most.
The demand side of the resource pull — and, interestingly, where the cheapest quality in the whole map sits. Missiles and missile defense consume rare earths and secure chips; the Navy shipbuilding push is a decade-long backlog. These primes trade at market-or-below multiples despite record backlogs.
The second-derivative, under-owned corner. Silver runs a sustained structural deficit (solar, electronics, EVs) yet the miners screen far cheaper than the rest of the theme. Gallium, germanium, and antimony are China-controlled specialty inputs to chips and defense optics — a watch-list for policy-driven re-rating.
Every ounce, ton and megawatt above has a financial counterpart — the firms that clear the risk, own the royalties, and supply the capital for the grid, power and data-center buildout. This is the annuity layer: exchanges taking a toll on commodity volatility, royalty/mineral owners with zero capex, and the real-asset managers financing the physical iron. Quality is high and cyclicality lower than the miners — so the entry question is usually multiple, not survival.
If energy is the spine of the AI build-out, insurance is its risk-transfer layer — inseparable from it the way marine and cargo cover is inseparable from oil shipping. Every hyperscale campus, substation and power plant needs construction-all-risk, property, energy, equipment-breakdown, business-interruption and cyber cover; the large, concentrated, novel risks flow to specialty / E&S underwriters and the reinsurers who backstop them, placed by wholesale brokers. The read is contrarian: the underwriters and reinsurers screen cheap on through-cycle ROE, while the beloved brokers are richly priced.
The map above is about owning scarcity. This section is the opposite bet — owning the escape from scarcity. If a company can genuinely relieve one of the four constraints with a proprietary, defensible solution, its IP re-rates violently and it can invert the entire thesis for the incumbents it disrupts. This is higher-risk, higher-reward, and demands watching the private/pre-commercial edge as closely as the public names.
A solver is upside if you own it and tail risk if you own the incumbent it displaces. Niron's rare-earth-free magnet is a call option on disruption — and a quiet threat to MP Materials' scarcity premium. Microsoft's in-chip microfluidics could compress the very cooling shortage that makes Vertiv a 65× stock. Track solvers not just to buy them, but to know when a "bottleneck" name's moat is quietly eroding.
The real bottleneck isn't generation capacity — it's the multi-year wait for grid interconnection. The solvers bypass the grid entirely with on-site generation.
Conventional transformers have multi-year lead times. Solid-state transformers (SSTs) collapse size, speed deployment, and add software control — the potential leapfrog. The purest players here are still private.
The disruptive bet is designing the constraint out — magnets with no rare earths at all, or superconductors that move far more power through far less copper. If either scales, it partially unwinds the scarcity thesis for miners.
Rack density has outrun air. The incremental solvers sell liquid-cooling hardware (public, investable); the changes-everything solver cools the chip from inside the silicon — and it's owned by a hyperscaler, not a pure-play.
The most radical "changes-everything" bet: put the data center in space. Orbit offers uninterrupted solar power (no grid queue) and radiative cooling into vacuum (no water) — attacking the two hardest constraints simultaneously. Long-dated and speculative; the pure-plays are private/pre-revenue, so the investable expression today is the launch layer.
"Private" = not publicly traded (accessible only via pre-IPO/secondary markets or venture, if at all). "Embedded" = the solution sits inside a large diversified company, so you can't buy it as a pure-play. Valuations for public solvers are live as of 22 Oct 2026; several trade at extreme multiples that price in success — size accordingly.
| Intrinsic / target | |
| Range | |
| Vs. price | |
| Key lever | |
| Guardrails |
Full Damodaran-guardrails model (per model type). Snapshot as of 22 Oct 2026. Research & education — not investment advice.
How to maintain this dashboard. The valuation flags and prices are a snapshot as of 22 Oct 2026. To refresh: re-pull live quotes/fundamentals for the tickers in the ranked table, update the px/pe/mc fields in the DATA array at the bottom of this file, and re-check the "Value" bucket against the new multiples. The theses/catalysts change less often — revisit on major news (new DoD/hyperscaler deals, China control changes, backlog updates).
Method & caveats. "Value" buckets are judgment, not a DCF — priced-for-perfection names (VRT, PWR, CCJ) and pre-earnings stories (OKLO, SMR, MP on GAAP) are flagged as such deliberately. Commodity names (FCX, silver) carry cyclical beta. Nothing here is a recommendation or sizing guidance; it's a structured universe to run independent FCFF/entry work against.
Primary sources: Goldman Sachs (US DC power), S&P Global (copper deficit), CSIS & FAS (rare earths / MP–DoD terms), Utility Dive (GE Vernova backlog), Silver Institute (silver deficit), company disclosures. Live fundamentals via brokerage feed.