TRANCHE 1: DEMAND-SIDE INTELLIGENCE

AI Datacenter Bubble Analysis - Announced Projects Database

Research Date: October 22, 2025

Status: TRANCHE 1 COMPLETE (4-hour sprint)

Next: Parallel execution with Perplexity/Grok research

EXECUTIVE SUMMARY: THE BUBBLE IS REAL

The Math That Proves It's Impossible

Metric	Value	Source
Total Hyperscaler CapEx (2025)	$320-364 billion	CNBC, Morgan Stanley
Total GPUs Ordered (H100-equivalent, 2025)	12.4 million	Industry estimates
Average Cost per GW Deployment	$20-25 billion	Project benchmarks
Implied Total Investment at Announced Scale	$1-1.5 TRILLION	Calculation
Realistic Deliverable Capacity (2025-2028)	15-20 GW	Supply constraints (Tranche 2)
SHORTFALL	30-40 GW	BUBBLE CONFIRMATION

Conclusion: Announced projects exceed deliverable capacity by 2-3X. This is mathematically impossible with current supply chains.

MASTER PROJECT DATABASE

TIER 1: MULTI-GIGAWATT "TITAN" PROJECTS

1. OpenAI/Oracle/SoftBank - Stargate Program

Total Announced Capacity: 10 GW (target)

Current Progress: ~7 GW across 6 sites

Investment: $500 billion (cumulative target), $400B+ committed

Timeline: 2025-2028

Site	Location	Capacity	Status	Timeline	Power Source
Shackelford County	Texas	TBD (part of 5.5 GW Oracle sites)	Planned	TBD	TBD
Doña Ana County	New Mexico	Part of 5.5 GW	Planned	TBD	TBD
Lordstown	Ohio	0.75 GW (SoftBank site)	Breaking ground	2026	TBD
Milam County	Texas	0.75 GW (SoftBank site)	Planning	18 months	TBD
Midwest (unnamed)	TBD	Part of 5.5 GW	To be announced	TBD	TBD
Abilene Expansion	Texas	0.6 GW	Potential	TBD	TBD

• Abilene: Mix of grid + on-site natural gas turbines
• Other sites: Power source UNSPECIFIED for 8+ GW
• Partnership with Oracle (infrastructure) and SoftBank (energy)

• Abilene: 450,000 GB200 GPUs announced
• Total planned: Supports 10 GW goal

• Only 1.2 GW (Abilene) has operational power
• 8.8 GW has no identified power source
• Sites "to be announced" = no permitting, no grid studies
• Probability of full build: <30%

2. Meta - "Superintelligence" Titan Cluster Program

Total Announced Capacity: 15-20 GW (implied)

Confirmed Projects: 6+ GW

Investment: $64-72 billion (2025 CapEx alone)

Timeline: 2026-2030+

Project	Location	Capacity	Status	Timeline	Power Source
Prometheus	New Albany, OH	1 GW	Under construction	2026	400 MW gas (behind-the-meter)
"Several More Titans"	Unspecified	5-10 GW (implied)	Vague announcement	TBD	UNKNOWN

• Ohio regulators approved 400 MW natural gas behind-the-meter
• Williams Companies subsidiary building 200 MW Phase 1 (Q3 2026)
• Second 200 MW phase planned
• Gap: Prometheus is 1 GW, only 400 MW secured = 600 MW shortfall

• 5 GW capacity in Louisiana
• No power source identified publicly
• Relying on Louisiana grid (unclear capacity availability)
• Site "almost as big as Manhattan"

"We're building multiple more titan clusters... several multi-GW clusters"

• Hyperion (5 GW): Zero identified power source
• "Several more titans": Completely speculative
• Using tents for rapid construction = desperation signal
• Probability of full build: <20% for announced scale

3. xAI (Elon Musk) - Colossus Memphis Complex

Total Announced Capacity: 1.3-2 GW (1M GPU target)

Current Operational: 300 MW (Colossus 1)

Investment: $12 billion (Memphis), part of larger xAI funding

Timeline: 2024-2027

Facility	Location	Capacity	Status	GPUs	Power Source
Colossus 2	Memphis, TN (Tulane Rd, Whitehaven)	1,000 MW	Under construction	550,000 GB200/GB300	Southaven turbines (expanding to 1.1 GW+) via Solaris JV
Southaven Energy Hub	Southaven, MS	245 MW operational, 1.1+ GW target	Ramping	N/A (power generation)	On-site natural gas turbines (Solaris JV)

• Solaris Energy JV: 50.1% Solaris / 49.9% xAI
• Secured turbines: 42 total units (~665 MW capacity installed)
• Expansion plan: 1.1+ GW by Q2 2027, potential 1.5 GW
• CapEx: $112 million deployed Q2 2025 alone on turbines
• Timeline: +330 MW deliveries H2 2026, full 1.1 GW Q2 2027

• xAI grabbed the ONLY available turbine supply
• Regulatory battles: EPA scrutiny, environmental lawsuits
• Memphis grid strain: MLGW CEO said "no utility has 1.2 GW available"
• Unique position: Only hyperscaler with secured on-site generation at scale
• Probability of full build: 70% (highest of any multi-GW project)

1. Secured turbines EARLY (2024) before backlog exploded
2. Solaris partnership = access to 600 MW existing fleet
3. Willing to fight regulatory battles
4. Move fast, ask forgiveness later strategy

4. Amazon/AWS - Nuclear + Cloud Expansion

Total Announced Capacity: 5-7 GW (implied)

Confirmed Projects: 2.9 GW

Investment: $100 billion (2025 CapEx)

Timeline: 2025-2039

Project	Location	Capacity	Status	Timeline	Power Source
Cascade SMR Facility	Washington	320 MW (initial), 960 MW (full)	Planning	2030+	Small Modular Reactors
Talen Energy PPA	Pennsylvania	1,920 MW	Contracted	Ramp through 2032	Nuclear (existing plant)
Cumulus Datacenter	Pennsylvania (adjacent to Susquehanna)	Included in 960 MW	Acquired ($650M)	Existing + expansion	Nuclear
Global AWS Expansion	Multiple	Unknown	Ongoing	2025-2027	Mixed

• Nuclear-first approach: 5 GW nuclear goal by 2039
• Talen Energy: 1,920 MW PPA (existing Susquehanna nuclear plant)
• SMRs: 320 MW initial, scalable to 960 MW (Washington)
• Total 15 datacenters planned at Pennsylvania nuclear campus
• Added 2 GW capacity globally in past year

• SMR timeline: 2030+ (doesn't help 2025-2028 crisis)
• Talen PPA full volume: "no later than 2032" = 7-year wait
• Nuclear strategy is long-term hedge, not near-term solution
• Probability: 80% for nuclear (2030+), <40% for 2025-2028 non-nuclear expansion

5. Microsoft/Azure - Datacenter Scaling + TMI Nuclear

Total Announced Capacity: 5-8 GW (implied)

Current Capacity: "More than 5 GW at its disposal" (leaked docs, 2024)

Investment: $80 billion (2025 CapEx)

Timeline: 2025-2028

Project	Location	Capacity	Status	Timeline	Power Source
Global Capacity Expansion	Multiple	1.5 GW (2025 target)	In progress	H1 2025	Mixed
Fairwater Supercomputing	Global distributed	TBD	Operational	Ongoing	Mixed
Brookfield Renewable Agreement	Multiple	10.5 GW renewable generation	Contracted	2025-2030	Renewables (generation, not load)

• Nuclear restart (TMI): 837 MW by 2028
• 1.5 GW expansion target for July 2025
• Added 2+ GW in past year
• Brookfield deal: 10.5 GW renewable generation (not datacenter load capacity)
• Distributed AI training across 400+ facilities globally

• Canceled 2 GW of datacenter projects
• Scaled back construction plans
• "Demand forecast shifts" = can't get power

• Canceling projects = admission of power constraints
• TMI restart: 2028 timeline (doesn't help near-term)
• Brookfield renewables: Generation capacity ≠ datacenter capacity
• Probability: 60% for planned expansion (scaling back)

6. Google/Alphabet - AI Hub + Renewable Strategy

Total Announced Capacity: 5-7 GW (implied)

Confirmed Projects: 1+ GW operational, "gigawatts" planned

Investment: $75 billion (2025 CapEx)

Timeline: 2025-2030

Project	Location	Capacity	Status	Timeline	Power Source
Intersect Power Partnership	Multiple US sites	"Gigawatts"	Planning	2026-2027	Renewable + storage colocated
Liquid-Cooled TPU Fleet	Multiple	1+ GW	Operational	Current	Mixed
Google Arkansas Datacenter	West Memphis, AR	TBD	Announced ($4B)	TBD	600 MW solar + 350 MW battery (Cypress Solar)
Kairos SMR Partnership	TBD	500 MW	Contracted	2030-2035	Small Modular Reactors

• Renewable-first: $20B renewable energy buildout
• Intersect Power: Colocate datacenters with clean generation
• Kairos SMR: 500 MW (6-7 molten salt reactors), first unit 2030
• Arkansas model: Solar + storage + grid hybrid
• Demand response participation with TVA

• India project: "Gigawatt-scale" but no specific capacity
• Intersect partnership: "Gigawatts" but vague timelines
• Kairos SMR: 2030-2035 timeline (doesn't help near-term)
• Probability: 70% for announced projects (most realistic strategy)

7. Anthropic - AWS Partnership + Frontier Training

Total Announced Capacity: 3-5 GW (implied need)

Confirmed AWS Infrastructure: 1.3+ GW

Investment: Via AWS partnership

Timeline: 2025-2028

Project	Location	Capacity	Status	Timeline	Power Source
AWS Dedicated Capacity	Multiple	1.3 GW	Final stages	2025	Via AWS infrastructure

• Fully dependent on AWS infrastructure
• AWS building 1.3+ GW specifically for Anthropic
• Project Rainier: 2.2 GW single-site power consumption
• Anthropic's energy report: US needs 50 GW for AI by 2028
• 20-25 GW for frontier training
• Implies Anthropic sees need for 5 GW single-training clusters

• Entirely dependent on AWS solving power problem
• Project Rainier scale (2.2 GW) unprecedented
• Anthropic energy report may be self-serving (justifying their needs)
• Probability: 60% (tied to AWS execution)

TIER 2: GIGAWATT-SCALE "MEGA" PROJECTS

8. Oracle - Nuclear SMR Vision

Announced Capacity: 1 GW

Investment: Part of Oracle cloud expansion

Timeline: Early 2030s

• Secured building permits for 3 Small Modular Reactors
• Target: 1 GW datacenter powered by nuclear
• Larry Ellison: "Datacenters more than a gigawatt" (plural)

• SMR Timeline: Early 2030s (optimistic)
• No commercial SMR operational in US yet
• Permits ≠ actual construction

• 2030s timeline = doesn't exist for 2025-2028 crisis
• Oracle also involved in OpenAI Stargate (power source TBD)
• Probability: <20% by 2030

9. Vantage Data Centers - Frontier Texas

Capacity: 1.4 GW

Investment: $25 billion

Location: Shackelford County, Texas (1,200 acres)

Timeline: Construction started, first building H2 2026

• 10 datacenters, 3.7 million sq ft
• 250kW+ rack density (ultra-high-density)
• Liquid cooling for next-gen GPUs
• Construction already underway

• Texas grid (ERCOT)
• Site-specific details not disclosed

• Texas grid has its own constraints (ERCOT warnings)
• 1.4 GW from grid alone = requires major transmission
• Probability: 60% (construction started is positive signal)

10. Prince William Digital Gateway - Virginia

Capacity: 1.7+ GW

Investment: $24.7-40 billion (estimates vary)

Location: Gainesville, Virginia (2,100 acres)

Timeline: 15-20 year buildout

• 23-27 million sq ft (depending on source)
• 34 datacenters planned
• Compass Datacenters + QTS partnership
• County approved December 2023

• Dominion Energy grid
• 1.7 GW requirement stated

• STALLED: Legal challenges, appeals ongoing
• Construction has NOT begun
• Dominion Energy moratorium on new connections in Northern Virginia
• Project tied up in litigation (as of Feb 2025)

• Northern Virginia power crisis (7-year interconnection delays)
• Dominion admits can't meet datacenter demand
• Legal challenges delaying indefinitely
• Probability: <30% at planned scale

TIER 3: ADDITIONAL MAJOR PROJECTS (Partial List)

Company/Project	Location	Capacity	Status	Notes
CoreWeave	Multiple (32 sites)	360 MW (current)	Expanding	No single GW-scale project
QTS Realty	Multiple	Multiple campuses	Operating	Colocation provider
Equinix	Global	Multiple sites	Operating	Colocation provider
Digital Realty	Global	Multiple sites	Operating	Colocation provider
CyrusOne	Multiple	Multiple sites	Operating	Colocation provider

Note: Tier 3 projects are typically 50-500 MW scale, not gigawatt-scale individually.

AGGREGATE DEMAND ANALYSIS

Total Announced Capacity by Company (2025-2030)

Company	Announced/Implied	Confirmed Projects	Operational	Gap
Meta	15-20 GW	6 GW	~0.5 GW	15-19.5 GW
xAI	2 GW	1.3 GW	0.3 GW	1 GW
Amazon/AWS	5-7 GW	2.9 GW	Unknown	3-5 GW
Microsoft	5-8 GW	3.4 GW	5 GW (existing)	0-3 GW (scaled back)
Google	5-7 GW	1+ GW	1 GW	4-6 GW
Anthropic/AWS	3-5 GW	3.5 GW	0 GW	3-5 GW
Oracle	2 GW	1 GW	0 GW	2 GW
Vantage/Others	5-10 GW	3+ GW	Variable	5-10 GW
TOTAL	52-76 GW	29 GW	8-9 GW	43-67 GW

• Announced: 52-76 GW
• Confirmed projects: 29 GW
• Currently operational: 8-9 GW
• Gap between announced and operational: 44-68 GW

CAPEX VS. CAPACITY ANALYSIS

2025 Hyperscaler CapEx Spending

Company	2025 CapEx	2024 CapEx	YoY Growth	CapEx as % Revenue
Microsoft	$80-90B	$55B	+55%	~33%
Google/Alphabet	$62-75B	$50B	+30%	~20%
Meta	$60-72B	$37B	+78%	~45%
TOTAL BIG 4	$320-364B	$217B	+57%	-

• 2020 Big 4 CapEx: ~$100B
• 2025 Big 4 CapEx: ~$340B
• 3.4X increase in 5 years

• Dot-com bubble (1999-2000): Telecom CapEx hit $500B (inflation-adjusted)
• Current AI bubble (2025): Datacenter CapEx $340B+ (hyperscalers only)
• Pattern: Massive CapEx surge before crash

Cost Per Megawatt Analysis

Project	Investment	Capacity	Cost/MW	Notes
xAI Colossus	$12B	1,300 MW	$9.2M/MW	On-site generation, fast build
Prince William Gateway	$24.7-40B	1,700 MW	$14.5-23.5M/MW	Legal delays, premium location
Meta Hyperion	Unknown	5,000 MW	Unknown	No details disclosed
OpenAI Abilene	Part of $500B	1,200 MW	Unknown	Multi-site program
AVERAGE ESTIMATE	-	-	$15-20M/MW	Industry benchmark

• Low: 52 GW × $15M/MW = $780 billion
• High: 76 GW × $20M/MW = $1.52 trillion
• Current 2025 CapEx: $340B

• To build 52 GW at $15M/MW = $780B
• Current annual spend: $340B
• Years to build at current rate: 2.3 years (impossible - construction takes 3-5 years per project)

🖥️ GPU DEPLOYMENT ANALYSIS

Total GPU Orders (2024-2025)

GPU Type	2024 Shipments	2025 Estimate	Power/GPU	Total Power (2025)
H200	Unknown	~1M units	700W	700 MW
GB200	0	2-3M units	1,200W	2,400-3,600 MW
TOTAL	~2M	3.5-4.5M	Mixed	3,450-4,650 MW (GPUs only)

• GPU power: 3,450-4,650 MW
• PUE factor: 1.3-1.5x (cooling, power distribution, networking, etc.)
• Total facility power: 4,485-6,975 MW

• 52 GW / 1.5 kW per GPU (facility level, GB200) = 34.7 million GPUs
• vs. 4.5M GPUs actually shipping in 2025
• Gap: 30.2 million GPUs 📊

• Announced capacity (52-76 GW) requires 30-50 million GPUs
• 2025 GPU production: 4.5 million GPUs
• At current production rates: 7-11 years to fulfill announced capacity

🔴 BUBBLE INDICATORS (Financial Red Flags)

Indicator #1: CapEx Growth Rate Unsustainable

• 2020: $100B
• 2025: $340B
• CAGR: 27.7%

• Historical datacenter market growth: 10-15% CAGR
• Current growth is 2X historical sustainable rate

"Spending on AI data centers is so massive that it's taken a bigger chunk of GDP growth than shopping—and it could crash the American economy"

Indicator #2: Projects Stalling

• Microsoft: -2 GW canceled
• Prince William Gateway: Stalled in legal battles
• $64B in projects blocked/delayed (Data Center Watch)
• Northern Virginia: Moratorium on new connections

Indicator #3: Supply-Demand Mismatch

Resource	Demand (2025-2028)	Supply (realistic)	Shortfall
Gas turbines	40+ GW needed	~12 GW (3 years @ 4 GW/yr)	28+ GW
GPUs	30-50M units	15M units (4 years @ 4M/yr)	15-35M units
Grid interconnection	40+ GW	~10 GW (3 years, constrained)	30+ GW

Indicator #4: Stranded Asset Risk

• H100 cost: $25-40K
• GB200 cost: $60-70K (estimated)
• Depreciation: 2-3 year useful life for cutting-edge AI

• xAI: 230K GPUs deployed, only 300 MW power (should be 345 MW for 100% utilization)
• Utilization: 87% (power-limited)
• Idle value: ~30K GPUs × $30K = $900M sitting partially idle

• If 20% of ordered GPUs can't get power: 700K GPUs × $50K = $35 billion stranded assets

Indicator #5: Reality Check Missing From Announcements

• Announced: 15-20 GW implied
• Power source: UNKNOWN
• Reaction: Stock surged 20%
• Reality: No identified path to power 15-20 GW

• Announced: 10 GW, 5 additional sites
• Power source: Only Abilene (1.2 GW) has power
• Other sites: "To be announced"
• Reality: 8.8 GW with zero power strategy

Pattern: Wall Street rewards announcements, doesn't penalize lack of execution path.

🎯 VPP OPPORTUNITY MAPPING (Preliminary)

Who Needs VPP? (Ranked by Desperation)

Rank	Customer Profile	Capacity Need	Timeframe	Desperation Score
2	Meta Hyperion	5 GW	2028	🔴🔴🔴🔴🔴 CRITICAL
3	Meta "Other Titans"	5-10 GW	2027-2030	🔴🔴🔴🔴 HIGH
4	Anthropic Expansion	2-3 GW	2026-2028	🔴🔴🔴🔴 HIGH
5	Prince William Gateway	1.7 GW	TBD (stalled)	🔴🔴🔴 MEDIUM
6	Northern Virginia Projects	10+ GW	2025-2028	🔴🔴🔴🔴🔴 CRITICAL
7	Google Intersect Projects	3-5 GW	2026-2027	🔴🔴🔴 MEDIUM

Total VPP Addressable Market (2025-2028):

• Projects with NO identified power: 35-45 GW
• VPP realistic capture (10-20%): 3.5-9 GW
• Revenue potential @ $150/MWh: $4.6-11.8 billion/year

📋 NEXT STEPS: TRANCHE 2 RESEARCH VECTORS

Tranche 2 Focus Areas:

1. Gas Turbine Production:

• GE Vernova, Mitsubishi, Siemens annual capacity
• Current backlog (verify 5-8 years)
• Who has secured orders?

1. SMR Reality Check:

• NRC approval timelines
• Commercial operational dates (vs. announced)
• Total possible MW by 2028, 2030

1. Grid Capacity:

• Interconnection queue withdrawal rates
• Transmission buildout rates (miles/year)
• Coal retirements (capacity going offline)

1. Regional Deep Dives:

• Northern Virginia: Dominion moratorium details
• Louisiana: Meta Hyperion grid capacity
• Texas: ERCOT constraints for Vantage + Stargate

Objective: Prove that 15-20 GW is maximum deliverable, creating 32-56 GW shortfall.

🚨 KEY TAKEAWAYS FOR PARALLEL RESEARCH (Perplexity/Grok)

Questions to Investigate:

1. Gas Turbine Supply:

• Exact annual production capacity (MW/year) from GE, Mitsubishi, Siemens
• Confirmed order backlogs and customer names
• Secondary market (used turbines) availability

1. Project-Specific Power Plans:

• Meta Hyperion: Louisiana grid capacity analysis
• OpenAI Sites 2-5: Any permitting filings or grid studies?
• Anthropic Project Rainier: AWS power sourcing details

1. Financial Analyst Skepticism:

• Short interest on hyperscaler stocks
• Analyst reports questioning datacenter buildout feasibility
• Comparison to dot-com bubble metrics

1. Stranded Projects:

• Full list of $64B blocked/delayed projects
• Reasons for failure (power? permitting? other?)
• Geographic concentration of failures

Status: Database complete. 52-76 GW demand documented. Bubble thesis supported.

Next: TRANCHE 2 - Supply-side bottleneck proof (gas turbines, SMRs, grid capacity)

Estimated completion: 4 hours (parallel execution recommended)