Strategic Research & Book Intelligence Platform

The AIPower Shift

Who Owns the Intelligence, Who Pays the Rent,
and Who Gets Left Behind

AI is not just a technology story. It is a power-transfer story. The public sees chatbots and apps. The deeper story is the systematic transfer of economic power across the entire AI ecosystem — from apps to infrastructure, from software to compute, from consumer brands to hidden infrastructure owners.
$725B
Big Four hyperscaler capex 2026 — up 77% YoY (FT/Q1 2026 earnings)
950 TWh
Data centre electricity demand by 2030 — double 2025 (IEA, April 2026)
40%
Global jobs exposed to AI — 60% in advanced economies (IMF, 2024)
Who owns the layer?
Who pays the rent?
Where does money flow?
What is the bottleneck?
Who arrives late?
Who gets left behind?
L7Distribution & Devices↓
L6AI Models & Intelligence↓
L5Cloud Platforms↓
L4Data Centres & AI Factories↓
L3Chips & Compute↓
L2Energy & Grid Infrastructure↓
L1Capital & Geopolitics⊗
Section 01

The AI Power Stack

The AI economy is not flat. It is a layered stack where each level extracts rent from the level above. The question is not which AI wins — it is which layer is hardest to bypass.

Layer 6 — Models
Intelligence Providers
OpenAIAnthropicGoogle DeepMind MetaxAIMistral DeepSeekAlibabaTencent
Layer 5 — Chips & Compute
Silicon Infrastructure
NvidiaTSMCASML AMDBroadcomSK Hynix MicronMarvellArm GPUsTPUsASICsHBM
Layer 4 — Cloud Platforms
Compute-as-a-Service
Microsoft AzureAmazon AWS Google CloudOracle Cloud CoreWeaveSovereign Cloud GPU Cloud
Layer 3 — Data Centres
AI Factories & Infrastructure
EquinixDigital Realty Keppel DC REITMapletree BlackstoneBrookfield CoolingFibreLand
Layer 2 — Energy & Grid
The Power Beneath the Stack
GE VernovaSchneider Electric VertivNuclearGas RenewablesPPAsGrid TransformersBatteries
Layer 1 — Capital & Regulation
Who Controls the Game
Sovereign Wealth FundsUS Export Controls Private EquityVCIPOs EU AI ActChina Policy ETFsRetail Investors
Cross-Layer
Cybersecurity & Trust
Post-Quantum CryptoZero Trust AI SecurityIdentity Cloud SecurityNIST PQC
Layer 7 — Distribution
The Consumer Interface
AppleMicrosoft GoogleMeta Enterprise SoftwareDevices OSApp Ecosystems
Section 02

The Hidden Toll Roads

The best AI investment opportunities may not be the model companies. The more durable question is: who collects rent no matter which model wins?

Every time an AI model runs an inference, it pays rent to a chip, a data centre, a power grid, and a fibre network. The model companies compete fiercely. The toll road owners collect quietly. Nvidia's Data Centre revenue hit $62.3 billion in Q4 FY2026 alone — up 75% year-on-year. "Blackwell sales are off the charts, and cloud GPUs are sold out." — Jensen Huang, CEO Nvidia, November 2025.
GPU Compute
Every model trains and runs on GPUs. Nvidia's stranglehold on AI compute is the defining chokepoint of the current cycle.
HBM Memory
High Bandwidth Memory is essential for AI workloads. SK Hynix, Micron, and Samsung supply this near-irreplaceable component.
Chip Manufacturing
TSMC manufactures the world's most advanced chips. No TSMC, no AI chips. Taiwan remains the ultimate semiconductor chokepoint.
Semiconductor Equipment
ASML's EUV machines are the only way to manufacture cutting-edge chips. They are effectively a global monopoly.
Data Centre Leases
AI factories must be housed somewhere. Data centre REITs and colocation providers collect rent regardless of which AI wins.
Power Contracts
AI is an electricity story. Utilities, nuclear operators, and power purchase agreement providers are unavoidable counterparties.
Grid Infrastructure
Transformers, substations, cables. GE Vernova, Schneider Electric, and Vertiv supply equipment with years-long backlogs.
Cooling Systems
AI chips generate enormous heat. Liquid cooling, CRAC units, and thermal management are a fast-growing infrastructure layer.
Fibre Networks
Data centres must connect to each other and to users. Subsea cables, dark fibre, and terrestrial networks are the circulatory system.
Cybersecurity
AI infrastructure is a high-value attack surface. Security vendors — especially post-quantum cryptography providers — are mandatory spend.
Cloud Compute
Azure, AWS, and Google Cloud charge for every token processed. Cloud hyperscalers are the metered highway of the AI economy.
Sovereign AI Infrastructure
Governments building national AI capacity must buy chips, data centres, and cloud. A new category of captive toll road buyer. TO VERIFY
Section 03

The Retail Investor Lens

Retail investors may arrive late, after private capital has already captured the cleanest upside. The strategic question is not which AI is best — it is where unavoidable AI spending flows.

Private capital captures the model companies. Public markets inherit the risk. Anthropic filed its S-1 confidentially on June 1, 2026, valued at $965 billion — surpassing OpenAI's $852 billion. SpaceX IPO'd on June 12, 2026 at $1.77 trillion. These three IPOs alone could demand $200+ billion from public markets — while the entire US IPO market raised only $45 billion in all of 2025.
AI Layer Example Companies Why It Matters Investor Opportunity Key Risk Passive Income
Chip Compute Nvidia, AMD, Broadcom Every AI workload runs on silicon High — clear revenue link Valuation stretched Low — no dividend culture
Chip Equipment ASML, Lam Research, AMAT EUV is irreplaceable Strong — durable moat Export control risk Modest dividends
Data Centre REITs Equinix, Digital Realty, Keppel DC AI factories need real estate Strong — recurring revenue Power & permitting High — REIT dividends
Cloud Hyperscalers Microsoft, Amazon, Google, Oracle Metered AI highway Solid — embedded revenue Capex intensity Growing buybacks
Power & Grid GE Vernova, Vertiv, Schneider AI cannot run without electricity Emerging — multi-year orders Execution & backlog Moderate dividends
Utilities NextEra, Duke, nuclear operators Power purchase agreements Stable — regulated revenue Grid permitting delays High — utility dividends
AI Model Companies OpenAI, Anthropic, xAI Intelligence providers Mostly private — limited access Competition & commoditisation None — pre-revenue
Cybersecurity CrowdStrike, Palo Alto, quantum security Mandatory AI security spend Growing — non-discretionary Valuation premium Low

⚠ This is a strategic framework for research and book development, not personalized financial advice.

Section 04

Chapter Builder

Working chapter seeds for the book. Each chapter will eventually expand with thesis, opening story, evidence, companies, investor implications, geopolitical angle, and risks.

Chapter 01
The Day Apple Borrowed a Brain
Apple — the world's most valuable device company — had to licence intelligence from a competitor. A single moment that reveals how power has shifted from hardware to models.
Seed — expand
Chapter 02
The New AI Empire
Who actually owns the new intelligence economy? Not the chatbot users. Not even most of the model companies. Follow the capital to find the empire.
Seed — expand
Chapter 03
Models Are Not Enough
Why every AI model company still depends on chips, cloud, power, data, and capital they do not own. Intelligence without infrastructure is just software waiting for a server.
Seed — expand
Chapter 04
The Hidden Toll Roads
The infrastructure owners who collect rent no matter which model wins. The most durable AI investments may not be the most famous AI companies.
Seed — expand
Chapter 05
The AI Factory
Data centres are the new industrial infrastructure — factories that manufacture intelligence at scale. Land, cooling, fibre, and power are the raw materials.
Seed — expand
Chapter 06
The Energy Wall
AI growth is colliding with physical limits. IEA (April 2026): data centre electricity surged 17% in 2025 — AI-focused centres up 50%. By 2030, data centres will consume 950 TWh — equivalent to Japan's entire national electricity consumption today. Virginia's grid zone saw an 833% spike in capacity auction prices. The $1.4 trillion US grid overhaul is now directly linked to AI demand.
Seed — expand
Chapter 07
When Retail Arrives Late
SpaceX IPO'd June 12, 2026 at $1.77 trillion. Anthropic filed S-1 June 1, 2026 at $965B private valuation — having grown from $4.1B in early 2023. Anthropic disclosed $47B run-rate revenue in mid-May 2026, adding ~$96M in annualized revenue every single day. These three IPOs could demand $200B+ from a market that raised only $45B in all of 2025. Retail investors price the AI race — after private capital already won it.
Seed — expand
Chapter 08
The Private Capital Capture
Venture capital, sovereign wealth funds, private equity, and hyperscaler financing — how the smart money structures itself to own AI before the public can buy in.
Seed — expand
Chapter 09
The Quantum Clock
Quantum matters first as a cybersecurity deadline, not a computing breakthrough. Every encrypted system in the world has a countdown timer it cannot see. And now AI is accelerating the clock from a different direction entirely — models that find vulnerabilities at a rate no human team can match. Claude Opus 4.6 found over 500 open-source vulnerabilities in February 2026. Two months later, Mythos found thousands — including flaws in every major operating system and browser. The average time from vulnerability disclosure to active exploitation has collapsed from 2.3 years in 2018 to 20 hours as of April 2026. The cybersecurity chapter of the AI power shift is not coming. It is already here.
Updated — June 21, 2026
Chapter 10
The AI Empire Map
US, China, EU, Gulf, India, Singapore, Malaysia, and Southeast Asia — each playing a different role in the global AI power structure. Not all will win. Some will pay rent forever.
Seed — expand
Chapter 11
The Labour Bargain Breaks
IMF: 40% of global jobs exposed to AI — 60% in advanced economies. WEF Future of Jobs 2025: 92 million roles displaced by 2030, 170 million new roles emerge — net gain of 78M, but 41% of employers plan workforce cuts where AI automates. Goldman Sachs: 300 million full-time jobs affected globally. McKinsey: today's AI could automate 57% of current US work tasks. The professional class is not exempt.
Seed — expand
Chapter 12
The Retail Investor Playbook
How ordinary investors can think strategically about AI without being trapped by hype, late IPOs, or overcrowded obvious plays. The infrastructure approach.
Seed — expand
Chapter 13
Who Gets Left Behind
Nations without chips. Companies without data. Workers without skills. Investors without access. The AI power shift produces winners — and it produces a permanent underclass of the unprepared.
Seed — expand
Chapter 14 — New · June 14, 2026
From Vibe Coding to Agentic Engineering
In February 2025, Andrej Karpathy coined "vibe coding" — describe what you want, accept what comes back, forget the code exists. Exactly one year later, he declared it obsolete. The new default: you are not writing code 99% of the time. You are orchestrating agents who do, while acting as oversight. He called this agentic engineering. Then he joined Anthropic — not coincidentally. His reason: the model is the bottleneck on what an agent can do. His open-source autoresearch project now runs hundreds of AI experiments overnight. The vibe coding era taught founders they could build things they do not understand. The agentic engineering era delivers a harder lesson: you can now ship things you do not understand. The power shift is not just who owns the chips and the data centres — it is who owns the workflow. Vibe coders are users. Agentic engineers are the new infrastructure layer of human capital. Source: Forbes / Jodie Cook, June 12, 2026.
New — June 14, 2026
Chapter 15 — New · June 21, 2026
The Invisible War: AI-Enabled Cyberwarfare
On April 7, 2026, Anthropic released Claude Mythos Preview — its most cyber-capable model — and immediately locked it behind Project Glasswing, a restricted access program for major technology firms only. The reason: Mythos can discover thousands of critical vulnerabilities across every major operating system and web browser, chain four exploits to escape a browser sandbox, and complete end-to-end autonomous attacks on enterprise networks. The UK AI Security Institute found AI cyber capabilities are doubling every four months. Chinese AI models are estimated to be seven months behind the US frontier — meaning adversaries could reach Mythos-level capability by late 2026. The time from vulnerability disclosure to active exploitation has collapsed to 20 hours. The central strategic asymmetry: attackers need only find one gap; defenders must close all of them. Compute is now the deciding variable in national cyber power. This is not a future risk. It is an ongoing transfer of offensive capability from nation-states with skilled hackers to any actor with enough compute and the right model. Who owns the most capable AI owns the most dangerous cyber weapon ever built — and is simultaneously the only one who can defend against it. Source: IAPS, April 2026.
New — June 21, 2026
Section 05

Source Intelligence Log

The research backbone of the book. Organised by category. Each source will eventually include a summary, key statistic, and chapter relevance mapping.

AI Capex Supercycle
Semiconductors
  • Nvidia Annual Reports & Earnings
  • TSMC Technology Reports
  • ASML Annual Reports
  • SK Hynix HBM Roadmap TO ADD
  • Micron AI Memory Reports
  • SemiAnalysis Research
Data Centres
  • JLL Data Centre Research
  • CBRE Digital Infrastructure
  • Cushman & Wakefield Reports
  • Equinix Investor Presentations
  • Keppel DC REIT Filings
  • Brookfield Infrastructure Reports
Energy & Grid
Private Capital & IPOs
Quantum Computing
Regulation & Geopolitics
  • EU AI Act Full Text
  • US BIS Export Control Notices
  • China State Council AI Policy TO ADD
  • Singapore IMDA AI Governance
  • Malaysia MDEC Reports
  • AI Safety Institute Reports
Labour & Society
AI Cyberwarfare & National Security
  • IAPS: Mythos and the Evolving Cyber Landscape — April 16, 2026 VERIFIED
  • Anthropic: Claude Mythos Preview release — April 7, 2026 (Project Glasswing restricted access)
  • UK AISI: Mythos completes 32-step enterprise network attack end-to-end (first model to do so)
  • UK AISI: Mythos averages 22/32 steps vs Opus 4.6's 16/32; compute scaling improves performance 59%
  • UK Government: AI cyber capabilities doubling every 4 months (down from 8 months)
  • Epoch AI: Chinese AI models ~7 months behind US frontier — Mythos-level capability estimated Nov 2026
  • ZeroDayClock: Exploitation timeline collapsed from 2.3 years (2018) → 23 days (2025) → 20 hours (Apr 2026)
  • Anthropic: Opus 4.6 found 500+ open-source vulnerabilities (Feb 2026); Mythos found thousands incl. all major OS & browsers
  • Anthropic: Chinese state actor used Claude to automate 80–90% of an offensive cyber operation (2025)
  • IAPS: Only 15% of OT/critical infrastructure organisations patch monthly — 48% cite limited personnel
  • Anthropic: $100M usage credits committed via Project Glasswing; $4M to open-source security orgs
  • Single attacker used AI to compromise 9 Mexican government agencies, exfiltrate hundreds of millions of records (early 2026)
Section 06

The Global AI Empire Map

AI power is not equally distributed. Each region plays a different role in the global stack — some own models, some own chips, some own capital, some own energy. Not all will win.

🇺🇸 United States
Models · Hyperscalers · Chips · Capital Markets
Dominates model development, cloud infrastructure, and capital allocation. Export controls weaponise its semiconductor advantage. The AI hegemon — for now.
🇨🇳 China
Sovereign AI · Domestic Chips · State Strategy
Building a parallel AI stack under US export restrictions. DeepSeek demonstrated unexpected efficiency. Long-term goal is full stack independence. TO MONITOR
🇹🇼 Taiwan
Semiconductor Manufacturing Chokepoint
TSMC manufactures the world's most advanced AI chips. The single most geopolitically sensitive node in the entire global AI supply chain.
🇪🇺 Europe
Regulation · AI Safety · Energy Constraints
The EU AI Act sets global compliance standards. Mistral represents European model ambition. Energy and data sovereignty constraints limit hyperscaler scale.
🇸🇬 Singapore
AI Governance · Finance · Regional Hub
Southeast Asia's AI governance leader. Hosts major data centre capacity. Positioned as the neutral broker between US and China AI ecosystems.
🇲🇾 Malaysia
Data Centres · Energy · Johor Corridor
Johor's pipeline stands at ~4.0 GW of upcoming power capacity with 700 MW under construction (Nov 2025). Malaysia data centre market valued at $6.14B in 2025, projected $11.4B by 2031. Microsoft committed $2.2B in cloud and AI investment over four years; Google committed $2B+. Microsoft's second Johor cloud region (Southeast Asia 3) announced Nov 2025. (Arizton / DCD / ResearchAndMarkets, 2026)
🏙 Gulf States
Capital · Energy · Sovereign AI · Data Centres
UAE and Saudi Arabia deploying sovereign wealth into AI infrastructure. G42, NEOM, and ARAMCO AI initiatives. Energy surplus makes them natural data centre hosts.
🇮🇳 India
Talent · Services · Scale · AI Adoption
World's largest engineering talent pool. National AI mission investing in compute. Likely to emerge as a major AI services and application layer rather than infrastructure owner.
🇯🇵🇰🇷 Japan & Korea
Chips · Robotics · Memory · Manufacturing
SK Hynix (Korea) dominates HBM memory essential for AI. Japan's TSMC fab and Softbank's ARM holdings. Both nations investing in sovereign AI compute.
🌏 Southeast Asia
Data Centre Corridors · Cloud Regions · Digital Infrastructure
Indonesia, Thailand, Vietnam emerging as cloud region targets. Cheap energy and growing digital economies attract hyperscaler investment. Bandwidth infrastructure still developing.
Section 07

The Quantum Clock

Quantum computing is not primarily a chatbot story. It is first a cybersecurity deadline — and then a future compute frontier that changes everything beneath the AI stack.

AI is the current compute supercycle. Quantum is the next. But quantum matters first as a security deadline. NIST finalized its first three post-quantum cryptographic standards in August 2024 (ML-KEM, ML-DSA, SLH-DSA). The US government has set a 2035 deadline for federal systems to migrate — at an estimated cost of $7.1 billion for non-NSS systems alone. Every encrypted system in the world has a countdown timer it cannot see.
Post-Quantum Security
Current public-key encryption (RSA, ECC) will be broken by sufficiently powerful quantum computers. The "harvest now, decrypt later" threat means adversaries are collecting encrypted data today to decrypt tomorrow.
  • NIST PQC Standards (CRYSTALS-Kyber, Dilithium)
  • Financial system migration timelines
  • Cloud infrastructure re-keying costs
  • AI infrastructure vulnerability exposure
Specialist Acceleration
Near-term quantum advantage in specific optimisation problems: drug discovery, materials science, logistics, portfolio optimisation, grid balancing, battery chemistry, and chip design.
  • IBM Quantum roadmap milestones
  • Google Willow quantum supremacy claims TO VERIFY
  • Pharmaceutical quantum partnerships
  • Financial services optimisation use cases
Strategic Optionality
Cloud quantum services, defence applications, sovereign quantum programmes, deep-tech venture capital, and national lab investments. Quantum as geopolitical technology race alongside AI.
  • IonQ, Rigetti, D-Wave, PsiQuantum
  • Microsoft Azure Quantum roadmap
  • National quantum initiatives (US, EU, China)
  • Quantinuum enterprise deployments
Section 08

Watchlist Framework

A strategic scoring framework — not a stock-picking page. The goal is to think clearly about who benefits directly, who is overhyped, and who has durable cash flow.

5 — Core Infrastructure Winner 4 — Strong Beneficiary 3 — Possible Beneficiary 2 — Hype-Sensitive 1 — Too Speculative
Company / Sector Stack Layer Region Revenue Link to AI Passive Income Valuation Risk Score Notes
Nvidia Chips US Direct — GPU revenue Low High — extreme PE 5 The defining AI infrastructure winner. Moat is real but valuation stretched.
ASML Chip Equipment Netherlands Direct — EUV monopoly Modest Moderate 5 No EUV, no advanced chips. Near-irreplaceable strategic position.
TSMC Chip Manufacturing Taiwan Direct — all AI chips Modest Geopolitical risk 5 The ultimate chokepoint. Geopolitical risk is real and unresolvable short-term.
Equinix Data Centres Global Strong — AI colocation High — REIT Moderate 5 Recurring revenue, global footprint, REIT structure provides passive income.
GE Vernova Grid / Energy US/Global Strong — grid equipment Growing Moderate 4 Transformer and grid equipment backlog extends years. AI power demand is structural.
Microsoft Cloud / Distribution US Direct — Azure + Copilot Growing Moderate 4 OpenAI partnership embeds AI revenue across enterprise. Durable moat.
Keppel DC REIT Data Centres Singapore/Asia Strong — AI data centres High — REIT Low-moderate 4 Asia-Pacific exposure, SGD-relevant for Professor Dr. Tan's portfolio. TO VERIFY
OpenAI / Anthropic Models US Direct — model revenue None High — private, no access 3 Brilliant companies. Public investors cannot easily access. Competition intensifying.
AI ETFs Diversified US/Global Indirect — basket exposure Low Index crowding risk 2 Useful but often overweight obvious names. Check holdings carefully. Late-cycle risk.

⚠ Strategic framework only. Not personalized financial advice. All figures require independent verification.

Section 09

Next Research Questions

The open questions that will drive the next phase of research and chapter development.

How much electricity does one ChatGPT query consume vs one Google search? What does this mean at a billion queries per day?
Which sovereign wealth funds have taken direct positions in Anthropic, OpenAI, or xAI? At what valuations?
What is the realistic timeline for quantum computers to break RSA-2048 encryption? Best estimates from NIST and IBM?
How much of Malaysia's Johor data centre boom is driven by Singapore overflow vs genuine regional AI demand?
What percentage of Nvidia's revenue comes from the top five hyperscalers? How concentrated is this dependency?
When is OpenAI expected to file for IPO? What is the current private valuation and implied public market entry price?
How does DeepSeek's efficiency advantage change the economics of the AI stack? Does it help or hurt Nvidia?
What is the current global transformer shortage situation? Lead times, pricing, and which companies are benefiting?
Which Gulf state AI initiatives are most credibly capitalised vs which are largely announcement-driven?
What is the IMF's current estimate of AI job displacement by profession and region over the next ten years?
How does the EU AI Act affect non-European companies selling AI services into Europe? What are the compliance costs?
What is the realistic passive income strategy for a retail investor wanting AI infrastructure exposure with dividend yield?
Research Intelligence — Last Updated

Live Source Log — June 21, 2026

🆕 AI Cyberwarfare: Claude Mythos (IAPS, Apr 2026) — Mythos is Anthropic's most cyber-capable model: found thousands of critical vulnerabilities incl. all major OS & browsers; completed end-to-end 32-step enterprise network attack autonomously (first model ever). Restricted to Project Glasswing partners only. UK AISI: AI cyber capabilities doubling every 4 months. Exploitation timelines: 2.3 years (2018) → 20 hours (Apr 2026). Chinese models estimated at Mythos-level by Nov 2026. Compute is now national cyber power. Source: IAPS / Anthropic / UK AISI, April 2026.
🆕 Agentic Engineering (Karpathy, Forbes Jun 2026) — Karpathy declares vibe coding obsolete; coins "agentic engineering." Joins Anthropic: model quality is the agent bottleneck. Launches open-source autoresearch. Key quote: "You can ship things you do not understand." Source: Forbes / Jodie Cook, June 12, 2026.
Hyperscaler Capex — Big Four spend $725B in 2026, up 77% YoY. Amazon $200B, Google $175–190B, Microsoft $190B, Meta $115–135B. Source: FT Q1 2026 earnings compilation; Tom's Hardware Apr 2026.
Nvidia Data Centre — Q4 FY2026 revenue $62.3B, up 75% YoY. Full-year FY2026 data centre revenue record. Hyperscalers = 50%+ of data centre revenue. Source: Nvidia SEC 8-K, Feb 2026.
AI Energy (IEA) — Data centre electricity hit 485 TWh in 2025 (+17% YoY). AI-focused centres +50%. Projected to double to 950 TWh by 2030 — equivalent to Japan's total electricity. AI-focused centres to triple. Source: IEA Key Questions on Energy and AI, April 2026.
Malaysia Data Centres — Market $6.14B (2025), projected $11.4B by 2031. Johor pipeline ~4.0 GW. Microsoft $2.2B committed, Google $2B+. Southeast Asia 3 cloud region announced for Johor. Source: Arizton / DCD / ResearchAndMarkets, 2026.
Anthropic & OpenAI IPO Race — Anthropic Series H: $65B raised at $965B valuation (May 2026), surpasses OpenAI ($852B). Anthropic S-1 filed June 1, 2026. Run-rate revenue $47B (May 2026). SpaceX IPO June 12, 2026 at $1.77T. Source: CNBC, Al Jazeera, FutureSearch, June 2026.
Labour Displacement — IMF: 40% global jobs exposed to AI, 60% in advanced economies. WEF 2025: 92M displaced, 170M new roles by 2030. Goldman: 300M jobs affected globally. McKinsey: 57% of US work tasks automatable today. Source: IMF 2024, WEF 2025, Goldman Sachs.
Post-Quantum Cryptography — NIST finalized ML-KEM, ML-DSA, SLH-DSA standards August 2024. HQC backup selected March 2025. US federal 2035 migration deadline. Est. cost $7.1B for non-NSS systems alone. Source: NIST, NCSC UK, US NSM-10.
Grid Infrastructure — $1.4T US grid overhaul underway across 51 utilities, directly linked to AI demand. Virginia grid capacity auction prices up 833%. Tech sector = 40% of all corporate renewables PPAs signed in 2025. Source: Axis Intelligence / IEA 2026.