TREND TRACKING · AI Infrastructure
A 702 MW PPA and 420 MW Campus Push AI Real Assets Into Contracting—Cash Flow Still Lies Ahead
Physical Economy & Real Assets | Saturday Signal Scan
Research watch: American Electric Power / AEP · Alphabet / GOOG / GOOGL · Bloom Energy / BE · Anthropic · EverWind · Powered Land
AEP plans about $4 billion for 702 MW of fuel-cell generation under a 20-year Google PPA. EverWind has an Anthropic lease for the initial phase of a planned 420 MW Point Tupper campus. Contracted capacity advances, but permits, funding, energization, utilization and FCF/MW remain unproven.
AEP capacity and Google PPA term
AEP investment; simple $5.70m/MW, generation only
Point Tupper phases for late 2027 and 2028
Developer-disclosed campus and wind investment
Research view: contracts and asset structure advance materially | 2027–2028 delivery, funding cost and social license govern cash conversion
01 · RESEARCH BRIEF
The one-minute brief
The material change is not another announced-capacity headline. AI demand is beginning to bind customer contracts, dedicated power, industrial land and community-cost commitments into the same asset structure. AEP plans about $4 billion for a 702 MW fuel-cell project near Cheyenne under a 20-year Google PPA, a simple generation-only intensity of about $5.70 million/MW.[1] EverWind plans a 420 MW Point Tupper campus: an initial 95 MW phase targeted for late 2027 and a 325 MW expansion targeted for 2028. It disclosed more than C$8 billion for the campus, another C$2 billion for wind, and an initial-phase long-term lease with Anthropic.[2][3] Both projects move Contracted MW toward Power-Secured MW, not yet Revenue-Generating MW.
Audio transcript
This week's physical-economy shift is not another announced gigawatt. Customer contracts, power projects, industrial land and community-cost commitments are entering the same asset structure. AEP plans 702 megawatts of fuel-cell power for Google, while EverWind advances a 420-megawatt Point Tupper campus for Anthropic. Contract evidence strengthens, but permits, funding, energization, utilization and free cash flow per megawatt remain the next gates.
Known facts and open questions
- Confirmed | AEP contracted power
- Google will buy power from Cheyenne Fuel Cell under a 20-year PPA. AEP controls about 92%; Basalt-advised funds own nearly 8%.[1]
- Confirmed | AEP delivery path
- Construction is planned for 2026 and full operation for end-2028. AEP previously disclosed an approximately $2.65 billion fuel-cell purchase agreement.[1][6]
- Confirmed | Point Tupper lease and phases
- EverWind signed a long-term initial-phase lease; Reuters identifies Anthropic. The campus plans 95 MW first and 325 MW more.[2][3]
- Confirmed | Community-cost commitments
- EverWind says it will pay attributable grid costs, add storage, use closed-loop cooling and fund community benefits.[2]
- Open | Cash return
- Neither project fully discloses pricing, IRR, debt, interest cost, utilization, revenue/MW, FCF/MW or stabilized return.
01 | Land Control
Control of developable industrial land
02 | Interconnection
Credible grid-upgrade path
03 | Power Secured
Deliverable generation, fuel and storage
04 | Permitted
Construction, environmental and local approvals
05 | Contracted
Customer contract with term and enforceability
06 | Funded
Equity, debt and interest cost can be carried
07 | Socially Accepted
Power, water, land and community terms accepted
08 | Energized to Cash
On-time power, utilization and FCF conversion
| Project | Confirmed | Simple intensity | Still missing |
|---|---|---|---|
| AEP Cheyenne Fuel Cell | 702 MW; about $4bn; 20-year Google PPA; end-2028 target | About $5.70m/MW, generation only | Permits, fuel/carbon-capture detail, funding, returns and delivery |
| EverWind Point Tupper | 420 MW; 95+325 MW; initial Anthropic lease; >C$8bn campus investment | >about C$19.05m/MW, developer basis | Lease economics, permits, funding, utilization, revenue and FCF/MW |
| Gate | AEP 702 MW | EverWind 420 MW |
|---|---|---|
| Permit | Not fully disclosed | Subject to commercial, technical and regulatory conditions |
| Interconnection | Not fully disclosed | Will self-fund connections/upgrades; completion not disclosed |
| Power | Bloom fuel cells selected; fuel and carbon-capture path open | Wind, storage and curtailment design; delivery open |
| Contract | 20-year Google PPA | Initial Anthropic long-term lease |
| Construction | 2026 start planned; full operation by end-2028 | 95 MW late-2027 target; 325 MW 2028 target |
| Funding / Interest | AEP about 92% and Basalt nearly 8%; debt and returns undisclosed | Privately financed claim; sources and interest cost undisclosed |
| Social License | Emissions objective disclosed; limited community evidence | Grid cost, water, community fund and Indigenous participation explicit |
02 · FACTS → IMPACT → VIEW
Why does this change matter?
Executable capacity moves from land story to contracted asset
- What is confirmed
- Many AI campuses consisted mainly of land, planned capacity or nonbinding power interest.
- Why it matters
- AEP combines 702 MW, about $4 billion, a 20-year Google PPA and 2028 delivery. EverWind combines an initial lease, phased MW, power and community commitments.[1][2][3]
- ACIS view
- Contracted-MW evidence strengthens, but only permits, funding, energization and utilization can create Revenue-Generating MW.
Social license becomes a project-economic term
- What is confirmed
- Power prices, water and community opposition often sat outside the core model.
- Why it matters
- EverWind self-funds grid upgrades, uses closed-loop cooling, funds community benefits and discusses Indigenous ownership. India's new framework reviews power, water, siting, value and governance together.[2][4]
- ACIS view
- Social license belongs in time-to-power, capitalized interest and project IRR.
Public funding cannot be assigned automatically to data-center cash flow
- What is confirmed
- A large utility loan can be read as AI capex.
- Why it matters
- USDA lent a record $3.3 billion to Oglethorpe Power, but the facility is not dedicated to a data center and serves 38 rural cooperatives plus broad demand.[5]
- ACIS view
- Treat it as financing-environment evidence, not data-center contracts, revenue or executable MW.
03 · EVIDENCE & ANALYSIS
Evidence and analysis
01｜Why AEP is a contract-grade power case
AEP combines large-scale fuel cells, ownership, customer, term and delivery timing. About $4 billion divided by 702 MW is roughly $5.70 million/MW, but it covers generation only and cannot be mixed with total data-center CapEx/MW. Full operation is targeted for end-2028; revenue is not yet converted.[1]
02｜Why Point Tupper is closer to a complete powered-land case
Point Tupper uses an existing industrial site, plans 420 MW and has an Anthropic lease for the initial 95 MW. Wind, storage, load flexibility, closed-loop cooling and community benefits sit in the same plan. Permits, funding, lease pricing, utilization and stabilized cash return remain open.[2][3]
03｜Policy starts quantifying social license
NITI Aayog's 14 October framework asks states to evaluate power, water, siting, economic value and governance. India's capacity rose from about 375 MW in 2020 to roughly 1.5 GW, shifting policy from investment attraction to investment readiness and measurable public value.[4]
04｜The boundary around a $3.3 billion public loan
USDA's loan to Oglethorpe Power may improve Georgia infrastructure, but the facility is not dedicated to data centers. It shows utility balance sheets expanding for a large-load era; it is not a data-center contract, revenue stream or dedicated award.[5]
05｜Who may capture AI real-asset value
Demand can be captured by equipment vendors, generators, developers, landowners, network operators and financiers. Corporate value depends on contract protection, cost pass-through, capital structure and execution; investor return also depends on valuation. Economic Value Created, Corporate Value Captured and Investor Return remain separate.
04 · INVESTMENT IMPLICATIONS
Industry and asset implications
Power utilities | Longer contract duration
Long PPAs convert some demand risk into execution, funding and asset-recovery risk.
Powered-land developers | Integrated delivery is the gate
Site, power, storage, cooling, community terms and tenant contracts must close together.
Governments | Public value moves upfront
Grid cost, water, jobs, tax and community benefits affect permission and time-to-power.
Investors | Move from announced MW to cash conversion
Focus on energization, lease quality, utilization, revenue/MW, FCF/MW and incremental ROIC.
This scan strengthens evidence of physical AI demand and contracting. It is not a rating, trade or position change for AEP, Alphabet, Bloom Energy or any related asset.
05 · VALIDATION & RISKS
What to watch next
AEP | 2026–2028 delivery
Permits, construction, equipment and fuel support full 702 MW operation by end-2028.
What would weaken the view: Permitting, supply, carbon capture or funding cost delays operation or lowers return.
EverWind | Initial 95 MW
Funding, approvals, grid upgrades, construction and Anthropic occupancy support late-2027 service.
What would weaken the view: The lease does not convert into funding, construction or energization.
Social license | No cost spillover
Grid, water, storage and community commitments become enforceable contracts and permit conditions.
What would weaken the view: Power-price, water or community disputes delay delivery.
Cash conversion | Real-asset return
Utilization, pricing, revenue/MW, FCF/MW, interest cover and incremental ROIC become visible.
What would weaken the view: MW and CapEx grow while cash returns remain absent.
What would change our view?
Long-term contracts may contain conditions, delay remedies, cost reopeners or credit support. Developer investment headlines are not funded or spent capital or asset book value. Fuel supply, carbon capture and lifecycle emissions could change fuel-cell economics. If community commitments do not become enforceable permit and tariff terms, social license can still fail.
06 · FAQ
Key questions
Is 702 MW already supplying Google?
No. Construction is planned for 2026 and full operation for end-2028.
Can $5.70m/MW be used as data-center build cost?
No. It is generation only and excludes buildings, servers, network, cooling and fit-out.
Is all 420 MW leased to Anthropic?
Public evidence confirms the initial phase. Expansion tenant coverage and utilization are not fully disclosed.
Is USDA's $3.3bn loan a data-center subsidy?
No. It serves 38 rural cooperatives and broad demand; the facility is not dedicated to a data center.
07 · TERMS & SOURCES
Terms, sources and related research
Key terms
- PPA
- Power Purchase Agreement. Term, cost recovery and protections determine contract quality.
- Contracted MW
- Capacity backed by an enforceable customer contract; not the same as energized or revenue-producing capacity.
- FCF/MW
- Free cash flow per energized megawatt, used to test cash conversion.
- Social license
- Continuing acceptance by communities, politics and public-resource systems beyond formal permits.
[1] American Electric Power｜702MW Cheyenne fuel-cell project and 20-year Google PPA｜2026.10.14 ↗
[2] EverWind｜Point Tupper data-centre campus and renewable-energy portfolio｜2026.10.16 ↗
[3] Reuters｜Anthropic signs lease for first Canadian data center｜2026.10.16 ↗
[4] NITI Aayog / Press Information Bureau｜Data Centre Evaluation Framework｜2026.10.15 ↗
[5] Reuters｜USDA record rural-utility loan may indirectly benefit data centers｜2026.10.16 ↗
[6] American Electric Power｜Form 10-Q, solid-oxide fuel-cell purchase agreement｜2026 Q1 ↗
This report uses first-party company and regulatory disclosure plus Reuters reporting available through 17 October 2026. About $5.70m/MW and more than C$19.05m/MW are simple headline-investment-to-planned-capacity ratios with different scopes and are not directly comparable. Planned investment, PPAs and leases are not recognized revenue, profit or free cash flow. The Oglethorpe Power loan is financing-environment evidence only. Research and education only; not investment advice.