Sheephill Group

AI Infrastructure

Power Analysis

Why AI scaling hits a power wall in 2029—and how behind-the-meter gas generation with CCS becomes the only viable path to sustain compute scaling growth prior to 2030

For Discussion Purposes Only • 1Q 2026

Executive Summary

The AI Power Wall

AI compute scaling requires 100x more power by 2030 — the grid cannot deliver it — behind-the-meter gas with CCS is the only viable path

The Demand
1 GW → 100 GW
100x power increase by 2030
→
The Bottleneck
39.4 GW/yr GT capacity
Sold out through 2030
→
The Failure Point
−20 GW shortfall in 2029
OOM scaling breaks at 108
→
The Solution
BTM Gas + CCS
12–18 mo deployment, $25/MWh

2025 Hyperscaler CapEx

$379B
Combined Big 5 — up 47% YoY

Total DC Capacity Tracked

44.9 GW
30 projects across 6 hyperscalers

2029–30 Power Gap

20–54 GW
Addressable shortfall range

The Argument in Five Points

1
AI capability scales in orders of magnitude (OOMs) of effective compute. Each OOM requires ~10x more power. Through 2027, the grid can keep pace. By 2029, demand hits 50 GW.
2
Gas turbine manufacturing is the binding constraint. GE Vernova, Siemens, and MHI produce 39.4 GW/yr combined. Only ~19.7 GW/yr is available for AI after existing commitments. Sold out through 2030.
3
The grid path is broken. PJM has 81+ GW in queue but only 9.8% converts to construction. Wait times: 4–8 years. Nuclear SMRs won't deliver before 2030+.
4
Behind-the-meter gas + CCS is the only path that delivers GW-scale power on the 12–18 month timeline hyperscalers need — already proven at 1.55 GW across xAI and Stargate deployments.
5
CCS net of 45Q credits produces $25/MWh LCOE — 55% below IPP gas. PPA pricing at ~$82/MWh matches dirty alternatives while delivering net-zero power. 45Q self-funds the capture infrastructure.
Year OOM GW Req’d GW Avail. Result
2026 103 1 15.8 PASS (+14.8)
2027 104 3 23.0 PASS (+20.0)
2028 106 10 23.0 PASS (+13.0)
2029 108 50 30.0 FAIL (−20.0)
2030 1010 100 45.9 FAIL (−54.1)
Probability-Weighted Base Case
10–20%
Overbuild
~30%
Adequate
~50%
Underbuild
The Gap = The Opportunity

The delta between announced and deliverable datacenter power — backed by $379B/yr in hyperscaler CapEx searching for power solutions — is the investment thesis.

Core Thesis: AI infrastructure demand is real and accelerating. The power delivery bottleneck creates a multi-year structural advantage for companies that can deliver behind-the-meter gas generation with carbon capture at scale. The companies that own this bridge infrastructure capture a generational opportunity.

Source: Project Olara Consolidated Model — OOM Executive Dashboard, AI OOM Scenarios, Turbine Summary, PJM Queue Data; Leopold Aschenbrenner, Situational Awareness (2024); Lazard LCOE+ v18.0; GE Vernova SEC filings

Sheephill Group

Primary Investment Thesis

ExxonMobil (XOM) — Integrated BTM Gas + CCS Platform

The only company positioned to deliver gas supply, pipeline, generation, CCS, and EPC execution at GW+ scale in 12–18 months

Four Structural Competitive Moats

Moat 1: Balance Sheet & Financing
Aa2 / AA-
Prefers to fund with balance sheet cash — does not need financing but can borrow at corporate credit rating, which is nearly equivalent to the U.S. sovereign rating. $10.7B cash on hand, $52B annual operating cash flow, net-debt-to-capital of just 11%. No project finance needed — eliminates lender due diligence delays and SPV structuring friction. 4.5% WACC vs. 7.7% for IPP project finance — saves $53M/yr per GW in CapEx recovery costs.
Moat 2: Land & Site Control
Existing Permian Basin acreage and industrial site footprint provides pre-permitted land adjacent to gas production. Brownfield site development cuts permitting timelines by 6–12 months vs. greenfield. No land acquisition risk — XOM already owns or controls the sites where BTM generation would be deployed.
Moat 3: Gas Transport & CCS Infrastructure
Permian wellhead gas at $1.75/MMBtu (49% below Lazard baseline). Existing gas gathering and transport pipelines eliminate midstream dependency. ~1,300 miles of CO₂ pipeline (Denbury acquisition, 2023) plus Bayou Bend CCS hub already in place — accelerates CCS permitting and sequestration delivery by 2–3 years vs. greenfield.
Moat 4: OEM & EPC Relationships
Deep, decades-long relationships with GE Vernova (turbine OEM) and Kiewit / Bechtel (EPC contractors) for islanded power solutions. In a supply-constrained market where GEV is sold out through 2030, relationship priority = execution priority. These firms will prioritize XOM over unproven datacenter developers. XOM’s AA- counterparty credit and project track record reduce EPC execution risk and accelerate delivery schedules.

Share Price

$146
NYSE: XOM • Feb 17, 2026

Market Cap

$629B
~4.22B shares outstanding
Balance Sheet Snapshot — Q4 2025
Cash & Equivalents $10.7B
Total Obligations1 ~$70B
Net Obligations ~$59B
Total Equity2 $266.6B
Oblig. / Capital ~20.8%
FY25 Operating CF $52.0B
1 Conservative total obligations: $34.2B long-term debt + $8.8B pension & postretirement liabilities + $26.1B other long-term obligations = ~$69.1B. We include the $26.1B of other long-term obligations (asset retirement, deferred tax, operating leases) that are excluded from standard debt-to-capital calculations to present a conservative view of total enterprise liabilities.
2 Book equity value: $266.6B. Market capitalization: $629B (~4.22B diluted shares × ~$149). Market value of equity is 2.4x book — reflects franchise premium. Market data as of 2/18/26.
Credit rating: Moody’s Aa2 / S&P AA- — nearly equivalent to U.S. sovereign rating. Industry-leading balance sheet.
5 GW Illustrative TAM — Incremental XOM Value
Total CapEx (5 GW)
$11.3B
$2.25B/GW × 5
XOM Equity at Risk
$6.3B
$1.25B equity/GW × 5
Incr. Annual Earnings
$1.9B
30% ROE on equity
Incr. EPS (5 GW)
+$0.45
$1.9B / 4.22B shares
At 22x P/E
+$9.90/sh
~6.8% above current price
Incr. Market Cap
+$41.8B
5 GW alone — scalable
Assumes 30% ROE on $1.25B equity/GW, current P/E of 22.2x, 4.22B diluted shares. 45Q credits ($153M/yr per GW) included in ROE. 5 GW represents ~1% of 2025 CapEx budget — de minimis capital commitment relative to enterprise.

Illustrative Project Finance Structure — JV BTM Power Plant

XOM sells Permian gas to a project finance JV that builds, owns, and operates the BTM power plant with CCS. The JV sells power to hyperscalers under a long-term PPA. XOM captures value at every link in the chain.

Equity Sponsor
ExxonMobil (XOM)
Gas supply + CCS + equity
Manages operations & offtake
Equity Co-Sponsor
Blackrock / Infra Fund
LP equity capital
Long-duration infra allocation
Equipment & EPC
GE Vernova / Kiewit
Turbine supply + EPC
Long-term service agreement
↓ Equity + Gas
↓ LP Capital
↓ Turbines + EPC
Special Purpose Vehicle
Project Co. — BTM Power Plant (1–5 GW)
Assets: CCGT + CCS
Revenue: PPA + 45Q
Debt: Ring-fenced project debt
↑ Gas Supply Agreement ↓
↓ 25-yr PPA @ ~$82/MWh
↓ CO₂ sequestration
XOM Upstream
Permian Gas @ $1.75
Long-term gas supply agreement
Locked-in offtake for associated gas
Hyperscaler Offtaker
MSFT / GOOG / AMZN
25-yr PPA @ ~$82/MWh
Net-zero 24/7 baseload power
CCS Sequestration
45Q → $153M/yr/GW
Denbury CO₂ pipeline network
$85/ton × 1.79M tons captured
Why This Structure: The JV isolates project risk from the XOM parent, attracts infrastructure LP capital at scale, and lets XOM capture margin at every layer: gas supply, equity returns, CCS/45Q, and long-term service. The hyperscaler gets net-zero baseload at IPP pricing ($82/MWh) with an AA- counterparty and 12–18 month deployment. The JV structure is infinitely replicable across sites.

Source: Project Olara — 1GW Natgas Unit Economics, Behind Meter Analysis, Revenue Flow Diagram; XOM Q4 2025 earnings release (Jan 30, 2026); IRA §45Q ($85/ton)

Sheephill Group

Supporting Investments

Opportunity Map — By Valuation & Conviction

Positioned across the AI infrastructure power value chain by risk/reward profile

1 • Fairly Priced
Near-term beneficiaries of AI CapEx spending — reasonable valuations with identifiable catalysts
Ticker Company Price Mkt Cap P/E Thesis
XOM ExxonMobil $146 $629B 22x Integrated BTM gas+CCS platform. Only co. delivering GW-scale power in 12–18 mo. See prior page for full thesis.
GEV GE Vernova $802 $221B 44x Largest GT OEM. 83 GW backlog sold out through 2030. 15.3 GW/yr capacity, guiding 20 GW by mid-2026. ~33% SRAs from DC customers.
BHP BHP Group $37 $186B 20x Copper mispricing. DC copper intensity ~25 MT/MW. Copper “too small a %” of revenue — repricing likely as structural deficit widens.
RIO Rio Tinto $81 $131B 15x Parallel copper thesis. New mines take 10+ yrs. Chinese stockpiles nationally strategic and unlikely to reach global markets.
ATX ATEX Resources Junior • Speculative Speculative copper development. Leveraged play on structural copper repricing from DC infrastructure demand.
SOL Solstice Adv. Materials Honeywell spinout • Newly listed Sole U.S. supplier of UF₆ — irreplaceable nuclear fuel precursor. Monopoly positioning as 2.3 GW SMR pipeline converts to demand.
2 • Monitoring for Reset
Strong long-term prospects but speculatively priced — likely to reset on missed execution targets

Cameco (CCJ)

$115 • $49B mkt cap • 117x P/E

World’s largest public uranium company. Nuclear fuel supply chain is the durable long-term bottleneck for SMR scaling. However, priced as if 20+ GW of nuclear capacity is already online and operating. Hyperscaler nuclear PPAs total just 2.3 GW with delivery 2028–2030+ at the earliest. At 117x trailing earnings, any delay in SMR deployment timelines (likely) triggers a significant multiple compression. Wait for a valuation reset below 40–50x on execution disappointment, then re-evaluate.

Catalyst to Watch

NRC licensing milestones for Westinghouse AP300 and NuScale VOYGR. Any slippage in the 2028–2030 SMR deployment window compresses CCJ’s speculative premium. The fuel supply thesis is correct — the price is not.

3 • Short Candidates
Very low likelihood of delivering on stated goals

Oklo (OKLO)

$68 • $11B mkt cap • N/M (pre-revenue)

SMR developer with zero operating units, zero revenue, and no NRC license. $11B market cap priced entirely on optionality. Prior NRC application was rejected. Delivery timeline is 2028 at the earliest, but more realistically 2030+. In a world where gas turbines deliver GW-scale power in 12–18 months, the probability that Oklo delivers on stated goals within a commercially relevant timeframe is extremely low. Short thesis: execution failure is the base case.

Risk Factors

NRC licensing uncertainty. FOAK construction risk. No proven supply chain. Sam Altman-affiliated — headline risk on AI sentiment shifts. Down ~65% from 52-wk high ($194). Borrow availability may be limited.

4 • Emerging Trends — Watchlist
New categories with early-stage investable themes

Next-Gen Semiconductor

New chip architectures optimized for inference workloads at lower power consumption. Companies building custom silicon (ASICs) for hyperscaler-specific AI workloads. Power efficiency gains extend the OOM curve and buy time for grid/GT supply chain to catch up.

Direct Liquid / Immersion Cooling

Datacenters shifting from air to direct-to-chip liquid cooling and full immersion cooling to manage thermal loads at higher rack densities. Reduces cooling energy by 30–40%, enabling more compute per MW. Key enabler for high-density GPU clusters at scale.

AI Application Layer Disruptors

Companies leveraging foundation models to disrupt traditional business models: legal, financial services, healthcare, enterprise software. The infrastructure thesis creates the substrate; the application layer is where value accrues to end users.

Note on Foundation Model Layer: Sheephill Group believes Amazon (AMZN), Alphabet (GOOG), and Microsoft (MSFT) are best positioned at the foundation model layer due to their intelligence community contracts and hyperscaler infrastructure advantages. However, detailed analysis of the foundation model competitive landscape is beyond the scope of this report, which is focused on the physical infrastructure layer of the AI stack.

Source: Project Olara Consolidated Model; company SEC filings; Bloomberg; Robinhood; StockAnalysis.com. Market data as of Feb 18, 2026. All P/E ratios trailing twelve months.

Sheephill Group

Section 1

The AI Scaling Curve

Leopold Aschenbrenner's Orders of Magnitude Framework — Situational Awareness (2024)

The OOM Framework

AI capability scales in orders of magnitude (OOMs) of effective compute. Each OOM requires ~10x more compute, translating directly into exponential power demand. The path from today's AI to superintelligence requires scaling from 100 MW to 100 GW of datacenter power—a 1,000x increase.

Power Scaling Required

100 MW → 100 GW
1,000x power increase needed to reach 10^10 OOM

Equivalent Reference

20%+ of US Grid
100 GW = more than entire PJM interconnection capacity
Target Year OOM Level Power (GW) Intelligence Equivalent Grid Feasibility Status
2022 100 90 MW Smart highschooler Yes ✓ Achieved
2024 101 100 MW — Yes ✓ Achieved
2026 103 1 GW Smart undergrad Stressed ✓ On Track
2027 104 3 GW Agentic AI (remote worker) Stressed ⚠ Grid Constrained
2028 106 10 GW Human genius Timing Tight ⚠ Dependent on GT supply
2029 107–108 30–50 GW — Not Feasible ✗ ~100–150% of GT capacity
2030 1010 100 GW Automated AI research Not Feasible ✗ Not possible before 2040
Key Takeaway: We are currently on track through 103–104 OOMs (1–3 GW). The constraint wall hits at 106 (10 GW) in 2028–2029 when power demand exceeds the entire global gas turbine manufacturing supply chain.

Source: Project Olara Consolidated Model — AI Compute Timeline Analysis; Leopold Aschenbrenner, Situational Awareness (2024)

Sheephill Group

Section 2

Where We Are on the OOM Curve

Current datacenter announcements place us firmly in the 103–104 transition zone

2025 Hyperscaler CapEx

$389B
Across 30 projects tracked

Total DC Capacity Tracked

44.9 GW
Power generation required (all stages)

2026 Hyperscaler CapEx

~$500B
2026 capex estimated based on combined hyperscaler guidance (AWS, MSFT, GOOG, META, OCI)

Hyperscaler Capacity by Company

Google
13.0 GW
Microsoft
9.6 GW
Amazon
9.2 GW
Oracle
5.1 GW
xAI
4.2 GW
Meta
3.7 GW

OOM Mapping: Where Announcements Land

Total tracked capacity of ~45 GW across all hyperscalers maps to somewhere between 107 and 108 OOMs—but most projects are in early stages. Actually deliverable capacity through 2028 is far less.

Deliverable by 2026: ~6.3 GW (103 OOM) ✓
Deliverable by 2027: ~18.0 GW cumulative (104) ✓
Deliverable by 2028: ~30.0 GW cumulative (106) ⚠
Required for 108: 50 GW ✗

Announcements ≠ Delivery

55% of tracked MW at HIGH or MEDIUM permitting risk. Only 45% in LOW-risk (ERCOT/BTM) paths.

Key Takeaway: Hyperscalers have announced enough capacity for 107–108, but deliverable power through 2028 only supports 106. The gap between announced and deliverable is the core investment thesis.

Source: Project Olara — Summary Dashboard, Project Details (US/Global), Power Generation tabs; SEC 10-K filings, company earnings guidance

Sheephill Group

Section 3

Gas Turbine Supply Chain Reality

Current manufacturing capacity and the path to 60 GW delivery by 2030

2025 YE Global Production Capacity

GE Vernova 15.3 GW
Siemens Energy1 15.1 GW
Mitsubishi (MHI)1 9.0 GW
Total Global Capacity 39.4 GW/yr

1 Siemens Energy and MHI do not publicly disclose annual GW production capacity. Estimates derived from GE Vernova's disclosed unit orders/GW ratio applied to Siemens and MHI reported orders & estimated deliveries.

Allocation After Existing Demand

Utility 30%
Industrial 20%
AI/DC 50%

Only ~19.7 GW/yr of global gas turbine production is available for AI datacenter power after existing utility and industrial commitments.

GE Vernova Pipeline (Q4 2025)

83 GW
Total contracted backlog (51 GW firm + 29 GW Slot Reservation Agreements + Q4 bookings), stretching into 2029
Datacenter % of SRAs ~33%
SRA-to-firm conversion rate ~3 GW/qtr
Sold out through 2030

SRA (Slot Reservation Agreement): A binding deposit-backed agreement reserving a manufacturing slot for a gas turbine unit. SRAs are the key leading indicator for tracking conversion from order placed to turbine delivered, as they represent committed demand ahead of firm purchase orders.

Scaling to 60 GW: Margin of Error

Reaching 60 GW/yr delivery by 2030 requires ~70% capacity increase across all three OEMs. GE Vernova has guided to 20 GW annualized by mid-2026, stretching to 24 GW by mid-2028 at existing facilities. When asked at the December 2025 Investor Day whether surging demand visibility would push the company toward a significant new greenfield production investment, CEO Scott Strazik was measured:

"We do not anticipate having to address this in the next 18 months."

— Scott Strazik, CEO, GE Vernova Investor Day (Dec. 9, 2025)

Translation: even with an 83 GW contracted backlog sold out through 2029, GE Vernova sees no urgency to build new factory capacity. Manufacturing lead time for new capacity: 3–5 years. Even with full commitment today, 60 GW delivery is a 2031–2032 reality at best.

Labor & EPC constraints compound the bottleneck. Kiewit and Bechtel are the primary EPC contractors GE works with for large-frame gas turbine installations. Qualified craft labor, site supervisors, and commissioning engineers are in short supply—adding physical limits to how fast even manufactured turbines can be installed and brought online.

Order-to-Commercial Operation: 42–57 Month Critical Path

2–4 mo
Contract Execution
→
48–60 mo
Manufacturing & Delivery
→
12–18 mo
Construction (parallel)
→
6–9 mo
Installation
→
3–6 mo
Commissioning
Key Takeaway: The 48–60 month manufacturing lead time is the critical bottleneck. Orders placed today won't deliver power until 2030–2031. The supply chain cannot be wished into scaling faster.

Source: Project Olara — Turbine Summary, Pipeline Tracking Q3 2025, Phase Timeline Estimates; GE Vernova SEC filings, Q4 2025 earnings, Investor Day (Dec. 9, 2025); Utility Dive; IIR News Intelligence

Sheephill Group

Section 4

The Grid Is Already Maxed

PJM Interconnection Queue Analysis — the largest US grid operator is overwhelmed

PJM Queue: Demand vs. Deliverability

PJM, serving 65 million people across 13 states, has an active interconnection queue of 81+ GW of generation projects. But the funnel from application to actual construction reveals massive attrition.

Active Queue 81.3 GW
Feasibility Study Complete 3.0 GW
Interconnection Agreement 14.5 GW
Proceeding to Construction 7.9 GW

Queue-to-Construction Conversion

Only 9.8% of the active PJM queue progresses to Engineering & Procurement. Of 81 GW in the active queue, only 7.9 GW has a Construction Service Agreement. Wait times: 4–8 years.

PJM Generation by Projected Year (MW)

2026
15,462 MW
2027
16,793 MW
2028
9,433 MW
2029
6,435 MW
2030
6,020 MW
50% Expected Attrition

Net of attrition: only ~28.2 GW cumulative through 2032

HIGH Risk PJM Projects

AWS Pennsylvania (Talen/PJM): Backlogged, 4–8 year wait, 100 GW+ queue
Google PJM Expansion: Backlogged, 4–8 year wait

6.3 GW at HIGH risk of delay (14% of total tracked capacity)

LOW Risk: BTM Bypasses Grid

Stargate Abilene: 360 MW, TCEQ approved, operational
Stargate Shackelford: 700 MW, 100% off-grid
xAI Colossus: 247 MW, permitted

~1.55 GW BTM permitted — bypasses queue entirely

Key Takeaway: The grid interconnection path is broken for datacenter-scale power. PJM's 4–8 year queue timelines are incompatible with AI scaling needs. Behind-the-meter generation is the only path delivering power on schedule.

Source: Project Olara — PJM Analysis, PJM Queue Data (9,255 records), Summary Dashboard, Federal Permitting Matrix

Sheephill Group

Section 5

OOM Scaling Fails in 2029

Summary Dashboard — Datacenter capacity vs. AI compute power requirements

Capacity Delivery vs. OOM Demand

Year† OOM Target† GW Required† GW Deliverable Result
2026 103 1 GW 19.7 GW PASS (+18.7)
2027 104 3 GW 23.0 GW cumul. PASS (+20.0)
2028 106 10 GW 23.0 GW cumul. PASS (+13.0)
2029 108 50 GW 30.0 GW cumul. FAIL (−20.0)
2030 1010 100 GW 45.9 GW cumul. FAIL (−54.1)

† Year, OOM Target, and GW Required are based on the scaling framework and estimates from Leopold Aschenbrenner, Situational Awareness (2024). GW Deliverable and Result are Sheephill Group estimates.

The 2029 Inflection Point

−20.0 GW
Shortfall at 108 OOM target in 2029

Constraint

Gas Turbines + Grid
~100% of global GT capacity needed for AI alone

Why 2029 Is the Wall

Jumping from 10 GW (106) to 50 GW (108) requires 5x the power in just 12 months. The gas turbine supply chain delivers ~20 GW/yr to AI. Even with 20% capacity growth, the math doesn't work. Orders placed today arrive 2030–2031.

Scenario Analysis: Probability-Weighted Outcomes

Overbuild Case
10–20%
Can't reach AGI due to constraints; industry absorbs excess capacity
Adequate Build
~30%
Reach 106 by 2028–30 but stall on path to 1010
Underbuild (Base Case)
~50%
Power generation constraints slow AI development vs. SA curve
Key Takeaway: AI scaling will most likely fail to maintain the Situational Awareness curve by 2029. The constraint is not compute hardware or algorithms—it is physical power generation. This creates a massive, urgent demand signal for non-grid power solutions.

Source: Project Olara — OOM Executive Dashboard, Summary Dashboard, AI OOM Scenarios; Situational Awareness (2024)

Sheephill Group

Section 6

The Only Solution at Scale

Integrated behind-the-meter natural gas power generation with CCS

Why Behind-the-Meter Gas + CCS

With grids maxed out, interconnection queues at 4–8 years, and turbine manufacturing sold out through 2030, only one power delivery model can meet hyperscaler timelines: islanded, behind-the-meter natural gas generation with carbon capture.

Grid Interconnection 4–8 years
Nuclear (SMR) 2030+ earliest
BTM Nat Gas 12–18 months

BTM is the only path that can deliver GW-scale power on the 2026–2028 timeline hyperscalers need.

XOM's Integrated Advantage

Upstream Gas Supply
Permian Basin integration — own the molecule from wellhead to generator
Midstream Pipeline
Direct pipeline infrastructure to DC campuses — no utility dependency
Power Generation
On-site CCGT / reciprocating engines — behind the meter, no grid queue
Carbon Capture & Storage
Integrated CCS leveraging existing geological expertise — ESG credibility

Proof of Concept: BTM Already Working

xAI Colossus 1: 247 MW on 15 gas turbines, permitted July 2025
Stargate Shackelford: 700 MW on 210 generators, 100% off-grid
Total BTM permitted: ~1.55 GW and growing

Scalability

BTM gas generation is modular and repeatable. Each 1 GW campus can be replicated at any site with gas pipeline access. XOM's upstream-to-power integration eliminates multi-party coordination risk.

Economics

Islanded power economics are compelling at datacenter utilization rates. Hyperscaler willingness to pay is high given AI revenue opportunity cost. The grid alternative isn't cheaper—it simply isn't available.

Key Takeaway: XOM's vertically integrated BTM gas + CCS model is uniquely positioned to solve the power bottleneck. No other company can deliver gas supply, pipeline, generation, and carbon capture at the scale and speed required. This is an infrastructure monopoly in formation.

Source: Project Olara — Behind Meter Analysis, Summary Dashboard (BTM projects), Federal Permitting Matrix, 1GW Natgas Unit Economics

Sheephill Group

Section 7

The Exxon Advantage: Power Economics

Cross-validated against Lazard LCOE+ v18.0 and EIA AEO2025 capital cost data

Two Structural Cost Moats

Moat 1: Permian Wellhead Gas
$1.75 vs $3.45/MMBtu
Internal transfer price for Permian associated gas vs. Lazard baseline. 49% fuel cost discount. Marginal lifting cost likely $0.50–0.75/MMBtu—gas was worthless (negative Waha pricing 47% of 2024 trading days) before datacenter demand.
Note on $1.75 pricing: This is not an opportunity cost concession—it is a deliberate commercial strategy. XOM is profitable on Permian gas well below $2/MMBtu. Below-market gas pricing is the wedge that brings hyperscalers to the table to negotiate long-term PPAs. XOM trades short-term margin on the molecule for 25 years of recurring revenue across the integrated power stack (generation + CCS + 45Q) at 30% ROE. The value creation is in owning the full vertical, not maximizing price on any single link.
Moat 2: AA-Rated Balance Sheet
4.5% vs 7.7% WACC
Corporate balance sheet financing (Aa2/AA-) vs. IPP project finance (Lazard: 60% debt @ 8%, 40% equity @ 12%). $53M/yr per GW CapEx recovery savings.
LCOE Comparison — $/MWh
XOM Standalone
$32/MWh
XOM + CCS (net 45Q)
$25/MWh
◀ Lowest cost
XOM + CCS (gross)
$52/MWh
IPP (no CCS)
$56/MWh
Lazard CCGT range
$48–$109
Firmed Solar (24/7)
$89–$164
CCS net of 45Q is cheaper than standalone

$152.5M/yr in 45Q credits more than offset $113M incremental CCS cost

From Cost Floor to Contract Price: PPA Pricing Bridge

LCOE is what it costs XOM to produce power. PPA price is what hyperscalers pay. The bridge shows how XOM earns a 30% ROE while matching IPP pricing.

XOM Standalone (No CCS)
$32
LCOE
+$66
30% ROE
−$7
−$3
$89
PPA
PRICE
Cost
Floor
+ ROE
Target
− Gas
Margin
− T&D
Saved
Contract
Price
XOM + CCS (Net of 45Q)
$25
LCOE
+$66
30% ROE
−$27
45Q
−$9
Gas+T&D
$82
PPA
PRICE
Cost
Floor
+ ROE
Target
− 45Q
Credit
− Gas
+ T&D
Contract
Price
IPP
Reference
$83
IPP
PPA
No CCS
12% ROE

The punchline: XOM's CCS offer lands at $82/MWh — virtually identical to an IPP's $83/MWh dirty gas price. The hyperscaler gets net-zero power at the same cost as conventional gas, with 12–18 month deployment, AA- counterparty, and GW+ scale. Meanwhile, XOM earns a 30% ROE on $1.25B equity at risk because 45Q credits self-fund the CCS infrastructure.

What Big Tech Is Actually Evaluating — $/MWh All-In Comparison

Option LCOE
$/MWh		 Est. PPA
$/MWh		 Time to
Power		 24/7
Baseload		 Net-Zero
Ready		 Counter-
party		 Scale
Available
XOM Standalone $32 ~$89 12–18 mo ✓ ✗ AA- GW+
XOM + CCS (net 45Q) $25 ~$82 18–24 mo ✓ ✓ AA- GW+
IPP Gas (No CCS) $56 $65–80 24–36 mo ✓ ✗ BBB Limited
Firmed Solar (24/7) $89–164 $100–180 24–48 mo ~93% firm ✓ Multi-party MW-scale
Nuclear SMR $141–221 $150–250 5–10+ yrs ✓ ✓ FOAK risk 0 operating
Grid Interconnect $48–109 $60–120 4–8 yrs ✓ Varies Utility Queue: 2,600 GW
✓ Green = advantage ● Amber = conditional ✗ Red = disadvantage PPA prices derived from XOM Power Economics Model (30% ROE, 25-yr term). Lazard LCOE+ v18.0, EIA AEO2025.

Cost Advantage

55%
CCS net LCOE ($25) vs IPP ($56)

All-in CapEx with CCS

$2.25B/GW
Base $1.25B + CCS $1.0B

Annual 45Q Revenue

$153M/GW
Self-funding: CCS is a revenue stream
Key Takeaway: Phase 1 — deploy standalone at ~$89/MWh PPA (12–18 months). Phase 2 — retrofit CCS and lock in ESG-compliant contracts at ~$82/MWh — matching IPP pricing while delivering net-zero power. 45Q credits more than offset CCS cost, making the net LCOE lower with carbon capture than without it. CCS is not a cost—it is a revenue stream that funds itself.
Notes on 45Q Economics
1
Why CCS net LCOE is lower than standalone. CCS adds ~$113M/yr per GW in incremental costs (additional CapEx recovery ~$67M, heat rate penalty ~$8M, CCS O&M ~$38M). The IRA’s Section 45Q credit generates ~$153M/yr per GW (0.35 tCO₂/MWh × 5,694 GWh × 90% capture = ~1.79M tons × $85/ton = $152.5M). The 45Q revenue exceeds CCS cost by ~$40M/yr, effectively subsidizing not just the capture infrastructure but also the base plant. Even at $0/ton 45Q, CCS LCOE of $52/MWh remains at parity with unabated IPP ($56). Source: IRA §13104; 26 USC §45Q.
2
One Big Beautiful Bill Act (P.L. 119-21, July 4, 2025). The OBBBA modified Section 45Q for facilities placed in service after July 4, 2025. The base credit rate is now $17/ton. The enhanced $85/ton rate requires compliance with prevailing wage and registered apprenticeship requirements during construction and the first 12 years of operation. XOM’s scale and labor practices are expected to qualify for the full enhanced rate. However, the sensitivity analysis (see Excel model) shows viable economics across the full $0–$100/ton range. Under any carbon pricing regime, XOM’s CCS position improves further as unabated competitors face penalties that XOM avoids. Source: P.L. 119-21, Title VII, §70522; IRS Instructions for Form 8933 (Dec 2025).
3
XOM’s qualification pathways to $85/ton. XOM has two independent paths to the enhanced credit rate. Path 1 (Prevailing Wage & Apprenticeship): The OBBBA’s 5× multiplier ($17 → $85/ton) requires Davis-Bacon prevailing wage compliance and registered apprenticeship utilization during construction and the first 12 years of operation. At XOM’s scale, this is standard practice—its EPC contractors (Bechtel, Kiewit) routinely comply with prevailing wage requirements, and apprenticeship programs are readily established on multi-billion dollar capital projects. Path 2 (IRA Grandfathering): The OBBBA’s tightening of begin-construction safe harbors (Notice 2025-42) was scoped exclusively to Sections 45Y/48E (wind and solar). The original IRS begin-construction guidance for 45Q (Notice 2020-12) remains intact, including the 5% safe harbor and the physical work test with a 4-year continuity window. XOM has been actively building CCS infrastructure since well before the OBBBA—the Denbury acquisition (closed 2023; ~1,300 miles of CO₂ pipeline), the Bayou Bend CCS hub, and multiple announced industrial CCS partnerships—establishing begin-construction thresholds under the original IRA framework with a construction deadline of January 1, 2033. Source: IRS Notice 2020-12 (45Q BOC guidance); IRS Notice 2025-42 (limited to §§45Y/48E); 26 CFR 1.45Q-6.

Source: Cross-validated vs. Lazard LCOE+ v18.0 (June 2025), EIA/Sargent & Lundy AEO2025 Capital Cost Study, IRA §45Q ($85/ton), Moody's Aa2 / S&P AA- credit ratings

Sheephill Group

Section 8

Delivery Speed as Competitive Advantage

Hyperscalers who prioritize project delivery speed will be advantaged vs. those stalled by renewable procurement timelines

AI Is Existential for Hyperscalers

For Microsoft, Google, Amazon, and Meta, AI is not a product line—it is the future of their entire business. The annual CapEx commitments tell the story: $379B in 2025 alone. Companies that secure power fastest will lead the AI race.

The question is not whether clean energy is the long-term answer—it is. The question is what bridges the 4–8 year gap between today's power needs and grid/renewable delivery timelines. Natural gas with CCS is emerging as that bridge.

"The cost of AI will converge to the cost of energy... the abundance of it will be limited by the abundance of energy."

— Sam Altman, Senate testimony

The Transition Is Already Underway

xAI / Stargate
Deploying 1.55 GW of gas generators as a bridge to long-term clean energy. Shackelford operating 100% off-grid while renewable procurement proceeds.
Amazon / Microsoft
Nuclear PPAs signed (2.3 GW) but won't deliver until 2028–2030+. Gas bridge essential to maintain project timelines while clean baseload scales.
Google
Kairos SMR won't deliver until 2030+. Google's 2025 CapEx is $75–85B. Interim gas+CCS preserves climate targets while avoiding multi-year delays.
CCS Aligns Climate Goals with Delivery Timelines

Carbon capture enables hyperscalers to deploy gas generation as a bridge fuel while maintaining credible progress toward net-zero commitments—the practical path to scaling AI responsibly

The Delivery Timeline Advantage

$379B
2025 combined CapEx committed
Capital is deployed. Every month of idle capacity is lost competitive position and shareholder value.
4–8 yrs
Grid interconnection queue
Renewable-only procurement creates multi-year delays incompatible with the AI scaling timeline.
12–18 mo
BTM gas + CCS deployment
The only power path that matches AI urgency while CCS preserves the emissions trajectory.
Key Takeaway: Hyperscalers are pragmatically bridging the gap between today's power needs and tomorrow's clean energy infrastructure. BTM gas with CCS allows them to scale AI now while maintaining credible climate commitments. The companies that deliver this bridge infrastructure capture a generational opportunity.

Source: Project Olara — Summary Dashboard, AI OOM Scenarios; company SEC filings, earnings guidance, Senate testimony

Sheephill Group

Conclusion

The Investment Implication

OOM Scaling
AI requires 100x more power by 2030
→
Grid Reality
Grids maxed, queues 4–8 yrs, turbines sold out
→
Failure Point
OOM scaling fails in 2029 (−20 GW shortfall)
→
The Solution
BTM Gas + CCS is the only scalable path
The Thesis

AI infrastructure demand is real and accelerating. The power delivery bottleneck creates a multi-year structural advantage for companies that can deliver behind-the-meter gas generation with carbon capture at scale.

Total Addressable Power Gap

20–54 GW
Shortfall range 2029–2030

Hyperscaler CapEx Commitment

$379B / yr
Capital searching for power solutions

How to Play the Trend

Near Term (0–2 Years)
BTM Power & Infrastructure Copper
XOM — Integrated BTM gas+CCS (covered above).
Select coal cos. — Bridge fuel optionality.
BHP, RIO — Copper exposure. DC copper intensity (~25 MT/MW) implies massive demand pull. Global supply deficit widening; new mine development takes 10+ yrs. Chinese stockpiles are nationally strategic and unlikely to reach global markets. Both still valued as diversified miners where copper is “too small a % of total revenue”—a mispricing likely to correct as copper reprices.
ATEX — Speculative junior with copper development upside.
Medium Term (2–5 Years)
Nuclear Fuel Supply Chain
SMRs are widely expected for late-stage AI scaling (107+), but OKLO, CCJ (Cameco) are priced speculatively—as if 20+ GW of nuclear is already online and operating. The fuel supply chain is the more durable bottleneck.

Solstice Advanced Materials (Honeywell spinout) — Sole U.S. supplier of UF₆, an irreplaceable precursor in nuclear fuel production. Significant growth expected as SMR pipeline converts to operational demand. A spinout signals Honeywell sees standalone value creation ahead.
Long Term (5+ Years)
Grid Modernization & Clean Baseload
Grid expansion, transmission buildout, and next-gen nuclear (SMRs at scale) become investable as regulatory and construction timelines mature. The near-term gas bridge creates the installed base and site infrastructure that long-term clean energy inherits.
Sheephill Group

Legal & Compliance

Disclaimer

This document is a research report prepared by the Sheephill Group for informational and discussion purposes only. It does not constitute investment advice, a solicitation, or an offer to buy or sell any securities or financial instruments.

Not a Registered Investment Adviser. The Sheephill Group is not a Registered Investment Adviser (RIA), and is not registered with the U.S. Securities and Exchange Commission (SEC), the Financial Industry Regulatory Authority (FINRA), or any state securities regulatory authority. No content in this report should be construed as personalized investment advice or a recommendation to take any specific action with respect to any security or financial product.

Publicly Traded Securities. Publicly traded companies mentioned in this report—including but not limited to ExxonMobil (XOM), GE Vernova (GEV), Siemens Energy, Mitsubishi Heavy Industries, Cameco (CCJ), BHP Group (BHP), Rio Tinto (RIO), and others—are referenced solely for analytical discussion and evaluation purposes. Their inclusion does not constitute a recommendation, endorsement, or solicitation to purchase, sell, or hold any securities of these companies.

No Warranty or Guarantee. The information contained herein is based on sources believed to be reliable, including publicly available filings, government data, industry research, and proprietary financial modeling. However, the Sheephill Group makes no representation or warranty, express or implied, as to the accuracy, completeness, or timeliness of such information. All estimates, projections, and forward-looking statements reflect the authors’ judgment as of the date of this report and are subject to change without notice.

Forward-Looking Statements. This report contains forward-looking statements regarding energy demand, infrastructure capacity, regulatory policy (including Section 45Q tax credits), commodity pricing, and corporate strategy. These statements are inherently uncertain and actual results may differ materially from those projected due to changes in market conditions, government policy, technological developments, or other factors beyond the control of the authors.

Tax & Regulatory Information. Discussion of tax credits, including Section 45Q of the Internal Revenue Code as modified by the Inflation Reduction Act of 2022 and the One Big Beautiful Bill Act (P.L. 119-21, July 4, 2025), is provided for analytical context only and does not constitute tax or legal advice. Recipients should consult qualified tax and legal professionals regarding their specific circumstances.

Conflicts of Interest. The Sheephill Group, its principals, and/or its affiliates may hold positions in securities discussed in this report. No obligation is assumed to update or revise this report in the event of changes in holdings or views.

Distribution & Confidentiality. This report is intended solely for the recipient(s) to whom it is directly provided. Redistribution, reproduction, or dissemination of this report, in whole or in part, without the prior written consent of the Sheephill Group is prohibited.

By receiving and reviewing this report, all recipients acknowledge and agree that they understand the foregoing limitations and that this document is a research report provided for informational purposes only, and not a recommendation or solicitation to engage in any investment activity.

Sheephill Group
Brian McGoldrick
Founder & Portfolio Manager
brian@sheephillgroup.com
+1 860-986-2682
Sheephill Group