Power Laws
Courtesy of Gemini 3 Pro DeepResearch 2025-11-26T16:25But what about AI spending?
The Physics of Concentration: A Power Law Analysis of Global AI Capital Expenditure (2024–2030)
Executive Summary: The Gravitational Well of Intelligence Capital
The definitive analysis of global Artificial Intelligence (AI) capital expenditure (CapEx) trends for the 2024–2025 period confirms that the market is not merely consolidating; it is adhering to a rigorous, steep power law distribution. In this economic architecture, the distribution of investment, compute capacity, and infrastructure ownership is heavily skewed, with a "hyper-head" of four primary entities—Microsoft, Amazon, Alphabet, and Meta—commanding a disproportionate share of global resources. This concentration is driven by the underlying physics of Large Language Model (LLM) scaling, where the computational cost to train frontier models increases exponentially, creating a natural monopoly structure that excludes all but the most capitalized participants from the training layer of the AI stack.
Current data indicates that these four "Hyperscalers" are on a trajectory to deploy over $300 billion in combined CapEx in 2025 alone, a figure that exceeds the GDP of many mid-sized nations and dwarfs the collective investment of the entire Tier 2 cloud market.1 This report validates the power law hypothesis through a multi-dimensional analysis of financial outlays, hardware acquisition, and energy procurement. It finds that while the "long tail" of the distribution is lengthening via sovereign AI initiatives and enterprise inference, the "head" is becoming heavier, creating a gravitational well that distorts supply chains, pricing power, and geopolitical strategy.
This document serves as an exhaustive roadmap of this capital phenomenon, dissecting the mechanisms of concentration, the emerging "middle class" of neoclouds and sovereign funds, and the physical constraints—energy and silicon—that will ultimately dictate the slope of the power law curve through 2030.
1. The Macroeconomic Architecture of the AI Power Law
1.1 Defining the Power Law in AI Economics
A power law distribution, or Pareto distribution, describes a system where a small number of events or actors account for the vast majority of the impact. In the context of AI infrastructure, this manifests as a "winner-takes-most" dynamic in the accumulation of floating-point operations (FLOPs) and the capital required to sustain them. The analysis of 2024–2025 spending data suggests that the AI infrastructure market is characterized by a Gini coefficient approaching 0.85, indicating extreme inequality in compute ownership.3
The drivers of this distribution are rooted in the "scaling laws" of deep learning. Research from Epoch AI and other institutes highlights that the compute required to train frontier models has grown by a factor of 4x to 5x annually since 2010, with costs rising approximately 2.4x per year since 2016.5 This exponential cost curve acts as a filter. When training a single model costs $100 million (GPT-4 era), a few dozen companies can compete. When the cost rises to $1 billion (projected for 2025/2026) and effectively $10 billion for future "super-models," the market mathematically collapses into an oligopoly.6
1.2 The Aggregate Scale of the "Arms Race"
The global AI market is undergoing a re-rating of unprecedented velocity. The infrastructure market alone is projected to reach $638 billion in 2024, scaling to $3.68 trillion by 2034.7 However, the leading indicator—CapEx—reveals the immediate intensity of the buildup. In 2024, the top five hyperscalers (adding Oracle to the Big Four) deployed an estimated $197 billion into AI infrastructure.1
This spending is not distributed evenly. The gap between the "Tier 1" hyperscalers and the "Tier 2" service providers (CoreWeave, Lambda, IBM, etc.) is widening. For instance, while a Tier 2 provider like CoreWeave projects a massive $12–14 billion CapEx for 2025, this entire sum represents merely 15% of Microsoft’s projected $80 billion outlay.1 This order-of-magnitude difference confirms the power law: the leading entity spends nearly 6x the leading challenger, who in turn spends orders of magnitude more than the typical enterprise.
Table 1.1: The Hierarchy of AI Capital Deployment (2025 Forecasts)
1.3 The "Circular Economy" of AI Financing
A critical mechanism reinforcing this power law is the circular nature of capital flows in the generative AI economy. The hyperscalers are not merely infrastructure builders; they are the primary financiers of the ecosystem. Microsoft’s investment in OpenAI, Amazon’s $4 billion stake in Anthropic, and Google’s backing of Anthropic create a closed-loop system where investment capital is "round-tripped" back to the investor as cloud revenue.9
This dynamic artificially inflates the revenue figures of the cloud divisions (Azure, AWS, GCP) while ensuring high utilization rates for their infrastructure. It creates a barrier to entry for independent cloud providers who lack the balance sheet to make multi-billion dollar equity investments in model labs. Consequently, the "head" of the distribution reinforces its own mass, utilizing its free cash flow to capture the most promising "torso" companies (the model builders), further steepening the power law curve.
2. The Hyperscaler Singularity: Analyzing the "Head" of the Distribution
The analysis of individual corporate strategies reveals that the "Big Four" are operating with a distinct economic logic, treating compute not as a commodity to be sold at a margin, but as a strategic resource essential for survival. This has led to a decoupling of CapEx from immediate revenue recognition, a phenomenon described by analysts as "spending ahead of the curve" to avoid an "iPhone moment" of obsolescence.11
2.1 Microsoft: The Architect of the Supercluster
Microsoft’s behavior in 2024–2025 defines the upper bound of the power law. With a fiscal 2024 CapEx of $55.7 billion and a 2025 projection of $80 billion, Microsoft is outspending its nearest competitors by significant margins in absolute terms dedicated to AI.1
Infrastructure Strategy: Microsoft is pioneering the transition from "data centers" to "AI factories." The reported "Stargate" project, a potential $100 billion cluster planned with OpenAI for the latter half of the decade, represents a singularity in infrastructure investment—a single facility that would cost more than the total existing infrastructure of many cloud competitors.12
Silicon Diversity: While heavily dependent on Nvidia, purchasing hundreds of thousands of H100s, Microsoft is aggressively deploying its Azure Maia accelerators to reduce the "Nvidia tax" and improve margin structure for internal workloads like Copilot.13
Implication: Microsoft’s strategy assumes that the power law will hold indefinitely—that the largest model, trained on the largest cluster, will always yield superior economic returns, justifying infinite capital scaling.
2.2 Amazon (AWS): The Industrial Scale-Out
Amazon, the market leader in cloud infrastructure (31% share), has responded to the AI boom with a "brute force" industrial strategy. Its projected $75 billion CapEx for 2024 is largely directed toward maintaining its market share dominance against the Azure surge.2
Vertical Integration: Amazon’s differentiation lies in its custom silicon maturity. The Trainium and Inferentia chip lines allow AWS to offer compute at a lower cost basis than competitors reliant solely on Nvidia. This is a critical defensive moat. By controlling the chip design, Amazon attempts to flatten the power law of cost while maintaining the power law of scale.14
Energy Acquisition: Amazon has been the most aggressive acquirer of nuclear and renewable energy assets, effectively cornering the market on gigawatt-scale power interconnections. This secures the physical layer of the power law, ensuring that even if competitors have capital, they may lack the electricity to deploy it.15
2.3 Alphabet (Google): The Integrated AI Factory
Google faces the "Innovator's Dilemma," forced to cannibalize its high-margin search business with higher-cost generative AI results. Its $52 billion annual CapEx run rate is focused on defending its data monopoly.2
The TPU Advantage: Google is the only hyperscaler that has operated a successful custom AI chip (TPU) at scale for a decade. This gives Google a theoretical efficiency advantage. While Microsoft and Meta scramble to build custom silicon teams, Google is already on its 6th generation of TPUs.
DeepMind Integration: The merger of Google Brain and DeepMind into a single unit was a structural reorganization designed to concentrate compute resources. Previously fragmented clusters were unified to train Gemini, reflecting an internal power law: concentrating resources into a single massive model rather than many smaller research projects.10
2.4 Meta: The Open Source Spoiler
Meta acts as a unique force in the market. With a CapEx forecast of $35–40 billion (rising to >$45B in some estimates for 2025), Meta is spending like a cloud provider but has no public cloud revenue stream.2
The Llama Strategy: By releasing Llama (3, 4, etc.) as open weights, Meta commoditizes the software layer of AI. This destroys the margins of proprietary model providers (like OpenAI/Anthropic) while driving massive demand for hardware (GPUs), which benefits the infrastructure providers.
Personal Superintelligence: CEO Mark Zuckerberg’s stated goal of building "personal superintelligence" and owning 600,000 H100-equivalents by the end of 2024 places Meta firmly in the "Head" of the distribution.11 This is arguably the most inefficient capital deployment in terms of direct ROI, but serves as a massive strategic hedge against platform dependency on Apple or Google.
3. The "Middle Class" and the High-Beta Torso
Beneath the hyperscalers lies a volatile "Tier 2" ecosystem. These companies are growing faster than the hyperscalers in percentage terms but remain significantly smaller in absolute CapEx, confirming the "long tail" structure of the power law.
3.1 Oracle: The Bridge to Sovereignty
Oracle has successfully positioned itself as the "Hyperscaler for the Rest." By aggressively courting Tier 2 model builders (Cohere, xAI) and sovereign nations, Oracle has justified a doubling of its CapEx.
Forecast: Oracle’s FY2026 CapEx is projected to hit $35 billion, a massive jump from $21 billion in FY2025.12
Strategy: Oracle focuses on "Superclusters"—networking massive numbers of Nvidia GPUs (up to 131,072 in a single cluster) better than some larger competitors. This technical differentiation has allowed them to capture xAI’s business, although the durability of this advantage is questionable as Microsoft and AWS close the networking gap.19
3.2 The Neoclouds: CoreWeave and Lambda
Specialized GPU cloud providers like CoreWeave and Lambda Labs represent the "speculative" layer of the power law.
CoreWeave: With a 2025 CapEx guidance of $12–14 billion (slashed from an initial $20B+ due to shell delays), CoreWeave is spending more than many sovereign nations.8 Its business model relies on "circularity" with a twist: it often serves as overflow capacity for Microsoft. This makes CoreWeave less of a competitor and more of a satellite to the "Head".20
Lambda Labs: Having raised $1.5 billion in late 2025 to build "gigawatt-scale" factories, Lambda illustrates the capital intensity required just to stay relevant. Their "one person, one GPU" vision drives a populist narrative, but their scale remains an order of magnitude below the top tier.21
3.3 xAI: The Wildcard
Elon Musk’s xAI is attempting to break the power law by brute force.
Colossus Cluster: xAI brought a 100,000 GPU cluster online in Memphis in record time (19 days), with plans to scale to 200,000.22
Financials: With a projected spend of $18 billion on data centers and a burn rate of $1 billion per month, xAI is operating with the capital intensity of a hyperscaler despite having a fraction of the revenue.23 This anomaly—a startup spending like a sovereign—creates a localized spike in the distribution curve.
4. The Sovereign Counter-Weight: Geopolitics as a Market Force
A significant new variable in the 2024–2025 data is the entry of Nation-States as direct participants in the AI CapEx ecosystem. Motivated by "Digital Sovereignty" and the fear of intelligence dependency on the U.S., nations are allocating public funds to build domestic compute capacity.
Table 4.1: Sovereign AI Investment Commitments (2024-2030)
4.1 Analysis of Sovereign Impact
While the headline numbers for sovereign initiatives appear large, a comparative analysis reveals they ultimately reinforce the power law of U.S. dominance.
Scale Disparity: The UK’s £2 billion investment over five years is equivalent to roughly two weeks of Microsoft’s 2025 CapEx. Canada’s $2 billion CAD is similarly negligible in global terms.
Hardware Dependence: Virtually all sovereign initiatives rely on Nvidia hardware. Japan’s subsidies to Sakura Internet or the UK’s AIRR expansion ultimately flow back to Nvidia, strengthening the "Head" of the hardware power law.
The Middle East Exception: Saudi Arabia and the UAE are the only sovereign actors with capital pools (via PIF and Mubadala) capable of matching hyperscaler spending. However, they are constrained by U.S. export controls. The U.S. government restricts the sale of H100/Blackwell chips to these regions to prevent leakage to China, effectively capping their ability to disrupt the market structure.37
5. The China Bifurcation: A Constrained Power Law
The global power law is not a single curve; it is a bifurcated system due to the "Silicon Curtain." China represents a separate, parallel distribution heavily distorted by U.S. sanctions.
5.1 The Efficiency Penalty
Chinese hyperscalers (Alibaba, Tencent, Baidu, ByteDance) are investing aggressively but inefficiently.
CapEx Trends: Alibaba and Tencent are increasing CapEx, but absolute figures are suppressed compared to U.S. peers. For instance, Alibaba’s cloud CapEx is focused on maximizing the utility of older chips (H20, A800) and domestic alternatives (Huawei Ascend).38
ByteDance: As the most aggressive player, ByteDance has sought to acquire hundreds of thousands of chips through various channels, spending billions to maintain a cluster capable of training Doubao and other models. Their CapEx efficiency is lower because they must pay premiums for hardware and engineered solutions to interconnect lower-bandwidth chips.39
5.2 The Divergence
Analysts forecast that while U.S. tech giants will increase CapEx by ~35% in 2025 to over $300 billion, China’s "Big 4" will increase spending to only ~$50 billion. This 6:1 ratio indicates that the power law is geographically exclusive; the U.S. ecosystem is pulling away exponentially from the Chinese ecosystem in terms of raw raw compute power availability.40
6. The Long Tail: Inference, Edge, and the Apple Anomaly
The "tail" of the power law is populated by enterprise inference, edge devices, and Apple. This section investigates whether the shift to inference will flatten the curve.
6.1 Apple’s Hybrid Strategy: The Anti-Hyperscaler?
Apple presents a stark contrast to the "brute force" strategy of Microsoft or Meta.
Spending: Apple’s 2025 CapEx was reported at $12.7 billion—a fraction of its peers.41
Private Cloud Compute (PCC): Instead of building massive general-purpose training clusters, Apple is building specialized "Private Cloud Compute" servers using its own Apple Silicon (M-series chips). This creates a highly efficient, privacy-centric inference layer that offloads the heavy lifting to the device (iPhone/Mac).43
Implication: Apple is betting that the power law applies to training but not to inference. By distributing inference across 2 billion active devices, Apple creates a decentralized "fog" of compute that rivals the hyperscalers in aggregate FLOPs but costs Apple very little in direct CapEx.
6.2 The Inference Market Split
Data suggests the ratio of training to inference is shifting. While training is centralized (Power Law), inference is fragmenting.
Edge AI Growth: The Edge AI hardware market is projected to reach $143 billion by 2034, growing at 21% CAGR.45
Enterprise On-Prem: Companies are increasingly buying their own inference hardware (e.g., Dell/HPE servers with smaller GPUs) to avoid cloud margins and data privacy risks. This "repatriation" of compute creates a fatter tail, distributing CapEx across thousands of enterprises rather than just four hyperscalers.46
7. The Physical Constraints: Energy and Supply Chain
The power law of capital is ultimately enforced by physical bottlenecks. The ability to spend money is not the same as the ability to deploy infrastructure.
7.1 The Energy Wall
The most binding constraint is power. AI data centers are projected to increase power demand by 165% by 2030.15
Winner-Takes-Power: Hyperscalers are pre-buying power capacity years in advance. Amazon’s acquisition of the Cumulus nuclear data center campus and Microsoft’s nuclear power deals demonstrate a "land grab" for electrons.47
Implication: Smaller players cannot sign 20-year PPAs (Power Purchase Agreements) for gigawatt-scale power. This physical constraint solidifies the power law; only the largest balance sheets can secure the energy required to run the largest clusters.
7.2 The Hardware Monoculture
Nvidia’s dominance (90% market share) means that CapEx dollars funnel into a single choke point.
Distribution of GPUs: Snippets indicate that the distribution of GPU clusters themselves follows a power law. A few dozen clusters globally hold the majority of H100s.48
Supply Allocation: Nvidia allocates chips based on strategic partnerships, not just price. This "allocation economy" favors the Hyperscalers and key strategic partners (like CoreWeave/sovereigns), starving the long tail of necessary hardware.13
8. Risks to the Power Law Thesis
While the current data strongly supports a power law, several economic risks could disrupt or break this trend.
8.1 The "Revenue Gap" and Depreciation
A major concern is the discrepancy between infrastructure spend and AI revenue. Analysts note a "$600 billion hole" where CapEx depreciation exceeds the aggregate revenue of AI software.50
Risk: If Generative AI fails to deliver productivity gains commensurate with its cost, the Hyperscalers will face a "CapEx Wall," forcing a rapid contraction in spending. This would likely hurt Tier 2 providers (CoreWeave) hardest, as Hyperscalers would retreat to their own owned infrastructure, consolidating the market further in a downturn.51
8.2 Obsolescence Risk
The rapid improvement of hardware (Blackwell replacing Hopper) means that billions of dollars of 2023/2024 investments could become technologically obsolete before they pay back their cost. This high depreciation rate favors the largest players with the most diversified cash flows, as they can absorb the write-downs that would bankrupt a pure-play Tier 2 provider.51
Conclusion: The Immutable Laws of AI Capital
The comprehensive analysis of 2024–2025 research data confirms that AI capital expenditure follows a rigorous power law distribution. This is not a temporary market anomaly but a structural inevitability driven by the physics of deep learning scaling and the economics of infrastructure scarcity.
The "Head" of the distribution—Microsoft, Amazon, Google, and Meta—has achieved escape velocity, creating a self-reinforcing cycle of capital deployment, energy acquisition, and talent hoarding. Their combined ~$300 billion annual spend creates a barrier to entry that is mathematically insurmountable for traditional competitors.
While the "Long Tail" is active—populated by sovereign nations seeking autonomy, specialized neoclouds handling overflow, and a vast array of edge devices—it remains economically subservient to the Head. Sovereign initiatives, while politically significant, are financially diminutive compared to the hyperscalers. Tier 2 providers operate largely as satellites within the hyperscaler orbit.
Final Verdict: The power law holds. The future of AI infrastructure is one of extreme concentration, where a handful of "AI Factories" in the US (and potentially China) will serve as the engines of global intelligence, while the rest of the world competes for access to their output.
Appendix: Data & Forecast Aggregation
Table A1: Global CapEx Projections by Segment (2025)
Table A2: The Cost of Intelligence (Training vs. Inference)
Works cited
Charted: The Rise of AI Hyperscaler Spending - Visual Capitalist, accessed November 26, 2025, https://www.visualcapitalist.com/the-rise-of-ai-hyperscaler-spending/
Where Big Tech AI Spending Goes: Cloud Platforms GPUs and Data ..., accessed November 26, 2025, https://www.softwareseni.com/where-big-tech-ai-spending-goes-cloud-platforms-gpus-and-data-centre-investment-breakdown
Scientific Reports Title to see here - arXiv, accessed November 26, 2025, https://arxiv.org/html/2509.10109v1
Anthropic Economic Index report: Uneven geographic and enterprise AI adoption, accessed November 26, 2025, https://www.anthropic.com/research/anthropic-economic-index-september-2025-report
Machine Learning Trends - Epoch AI, accessed November 26, 2025, https://epoch.ai/trends
How much does it cost to train frontier AI models? - Epoch AI, accessed November 26, 2025, https://epoch.ai/blog/how-much-does-it-cost-to-train-frontier-ai-models
Artificial Intelligence (AI) Market Size and Growth 2025 to 2034 - Precedence Research, accessed November 26, 2025, https://www.precedenceresearch.com/artificial-intelligence-market
CoreWeave Earnings Call: 40% Reduction in This Year's Capital Expenditure Guidance Mainly Attributed to Delivery Delays; Prices for Older-Generation GPUs Remain Firm, accessed November 26, 2025, https://news.futunn.com/en/post/64703361/coreweave-earnings-call-40-reduction-in-this-year-s-capital
AI: Real Revenues, Real Cash Flow, Real Infrastructure – And This Is Only Phase One, accessed November 26, 2025, https://www.financialsense.com/blog/21477/ai-real-revenues-real-cash-flow-real-infrastructure-and-only-phase-one
Computational Power and AI - AI Now Institute, accessed November 26, 2025, https://ainowinstitute.org/publications/compute-and-ai
Chart: Tech's AI-Fueled Spending Surge | Statista, accessed November 26, 2025, https://www.statista.com/chart/35046/capital-expenditure-of-meta-alphabet-amazon-and-microsoft/
Big Red borrows a lot of green, hopes AI will put it in the black, accessed November 26, 2025, https://www.theregister.com/2025/11/21/oracle_ai_adventures/
Nvidia Competitors: Top AI Chip Stocks to Watch Now, accessed November 26, 2025, https://www.ebc.com/forex/nvidia-competitors-top-ai-chip-stocks-to-watch-now
Chart: AWS Stays Ahead as Cloud Market Accelerates - Statista, accessed November 26, 2025, https://www.statista.com/chart/18819/worldwide-market-share-of-leading-cloud-infrastructure-service-providers/
AI to drive 165% increase in data center power demand by 2030 | Goldman Sachs, accessed November 26, 2025, https://www.goldmansachs.com/insights/articles/ai-to-drive-165-increase-in-data-center-power-demand-by-2030
Big Tech AI Stocks to Showcase AI Gains, Capex in Q4 Reports | by Beth Kindig - Medium, accessed November 26, 2025, https://beth-kindig.medium.com/big-tech-ai-stocks-to-showcase-ai-gains-capex-in-q4-reports-d25f69b7f904
Oracle: OK cloud gluttons, we're doubling CapEx again - The Stack, accessed November 26, 2025, https://www.thestack.technology/oracle-ok-cloud-gluttons-were-doubling-capex-again/
Oracle Announces Fiscal Year 2026 First Quarter Financial Results, accessed November 26, 2025, https://investor.oracle.com/investor-news/news-details/2025/Oracle-Announces-Fiscal-Year-2026-First-Quarter-Financial-Results/default.aspx
Oracle's Record-Breaking Cloud Growth and $25B Capex Investment: The Comeback Nobody Saw Coming - Datacenters.com, accessed November 26, 2025, https://www.datacenters.com/news/oracle-s-record-breaking-cloud-growth-and-25b-capex-investment-the-comeback-nobody-saw-coming
CoreWeave Aims for a $35 Billion IPO as AI Industry Matures - Inscope, accessed November 26, 2025, https://www.inscopehq.com/post/coreweave-ipo-tech-investing
Lambda Raises Over $1.5B from TWG Global, USIT to Build Superintelligence Cloud Infrastructure, accessed November 26, 2025, https://lambda.ai/blog/lambda-raises-over-1.5b-from-twg-global-usit-to-build-superintelligence-cloud-infrastructure
xAI Rises Nearly 20% This Year, Plans Major AI Chip Investment Over Next 5 Years, accessed November 26, 2025, https://www.pminsights.com/insights/xai-rises-nearly-20-this-year-plans-major-ai-chip-investment-over-next-5-years
Musk's xAI to hit $13b by 2029: Morgan Stanley - Tech in Asia, accessed November 26, 2025, https://www.techinasia.com/news/musks-xai-to-hit-13b-by-2029-morgan-stanley
Elon Musk's xAI is projected to lose $13 billion in 2025 — AI project burns $1 billion a month in expenditures | Tom's Hardware, accessed November 26, 2025, https://www.tomshardware.com/tech-industry/artificial-intelligence/elon-musks-xai-is-projected-to-lose-usd13-billion-in-2025-ai-project-burns-usd1-billion-a-month-in-expenditures
Project Transcendence and Abu Dhabi's Digital Strategy: Fuelling AI and Data Centre Growth in the GCC, accessed November 26, 2025, https://gulfdca.com/en/project-transcendence-and-abu-dhabis-digital-strategy-fuelling-ai-and-data-centre-growth-in-the-gcc/
Project Transcendence: Saudi Arabia's $100 Billion Gambit to Dominate AI - Medium, accessed November 26, 2025, https://medium.com/@cognidownunder/project-transcendence-saudi-arabias-100-billion-gambit-to-dominate-ai-a397c7d04891
AI Servers à la Carte: Major Investment by France and UAE in French Data Center, accessed November 26, 2025, https://www.morganlewis.com/blogs/datacenterbytes/2025/02/ai-servers-a-la-carte-major-investment-by-france-and-uae-in-french-data-center
France Bolsters National AI Strategy With NVIDIA Infrastructure, accessed November 26, 2025, https://blogs.nvidia.com/blog/france-sovereign-ai-infrastructure/
Japan's $135B AI Revolution: Quantum Computing Meets GPUs - Introl, accessed November 26, 2025, https://introl.com/blog/japan-ai-infrastructure-135-billion-investment-2025
Cabinet Decision on the Bill for the Act for Partially Amending the Act on Facilitation of Information Processing and the Act on Special Accounts, accessed November 26, 2025, https://www.meti.go.jp/english/press/2025/0207_001.html
Budget 2025: compute, capital and skills at heart of UK automation push, accessed November 26, 2025, https://www.roboticsandautomationmagazine.co.uk/news/ai/budget-2025-compute-capital-and-skills-at-heart-of-uk-automation-push.html
UK Compute Roadmap - GOV.UK, accessed November 26, 2025, https://www.gov.uk/government/publications/uk-compute-roadmap/uk-compute-roadmap
Canadian Sovereign AI Compute Strategy - Innovation, Science and Economic Development Canada, accessed November 26, 2025, https://ised-isde.canada.ca/site/ised/en/canadian-sovereign-ai-compute-strategy
Catalyzing AI infrastructure: opportunities for investment with the Canadian Sovereign AI Compute Strategy and the 2024 Fall Economic Statement | Insights | Torys LLP, accessed November 26, 2025, https://www.torys.com/our-latest-thinking/publications/2024/12/opportunities-with-canadian-sovereign-ai-compute-strategy-and-2024-fall-economic-statement
Transforming India with AI - Press Release:Press Information Bureau, accessed November 26, 2025, https://www.pib.gov.in/PressReleasePage.aspx?PRID=2178092
Microsoft and G42 partner to accelerate AI innovation in UAE and beyond, accessed November 26, 2025, https://blogs.microsoft.com/blog/2024/04/15/microsoft-and-g42-partner-to-accelerate-ai-innovation-in-uae-and-beyond/
U.S. Approves Advanced Chip Sales to Middle East A.I. Giants in Major Policy Reversal, accessed November 26, 2025, https://observer.com/2025/11/us-approves-ai-chip-sales-middle-east-humain-g42/
Alibaba outpaces ByteDance, Tencent in China's AI cloud: report - Tech in Asia, accessed November 26, 2025, https://www.techinasia.com/news/alibaba-outpaces-bytedance-tencent-in-chinas-ai-cloud-report
How large is the capital expenditure for AI? - Longbridge, accessed November 26, 2025, https://longbridge.com/en/news/229582105
Three reasons to favor US AI companies over China's - UBS, accessed November 26, 2025, https://www.ubs.com/us/en/wealth-management/insights/article/_jcr_content.0000023273.file/PS9jb250ZW50L2RhbS9pbXBvcnRlZC9jaW9yZXNlYXJjaC9wZGYvMjAvMzQvNjMvNS8yMDM0NjM1L2VuLzIwMzQ2MzUucGRm/2034635.pdf
Apple's $12.7B AI bet defies Big Tech's capex arms race | The Tech Buzz, accessed November 26, 2025, https://www.techbuzz.ai/articles/apple-s-12-7b-ai-bet-defies-big-tech-s-capex-arms-race
Apple's AI spending trails far behind other megacaps. It's not hurting sales - Mac Daily News, accessed November 26, 2025, https://macdailynews.com/2025/10/31/apples-ai-spending-trails-far-behind-other-megacaps-its-not-hurting-sales/
In the Loop: Shipping Apple's American-made advanced servers, accessed November 26, 2025, https://www.apple.com/newsroom/in-the-loop/2025/10/shipping-apples-american-made-advanced-servers/
Private Cloud Compute: A new frontier for AI privacy in the cloud - Apple Security Research, accessed November 26, 2025, https://security.apple.com/blog/private-cloud-compute/
Edge AI Market Size to Attain USD 143.06 Billion by 2034 - Precedence Research, accessed November 26, 2025, https://www.precedenceresearch.com/edge-ai-market
Artificial Intelligence (AI) in Hardware Market Size to Hit USD 210.50 Billion by 2034, accessed November 26, 2025, https://www.precedenceresearch.com/artificial-intelligence-in-hardware-market
Beyond the Bubble: Why AI Infrastructure Will Compound Long after the Hype | KKR, accessed November 26, 2025, https://www.kkr.com/insights/ai-infrastructure
The US hosts the majority of GPU cluster performance, followed by China - Epoch AI, accessed November 26, 2025, https://epoch.ai/data-insights/ai-supercomputers-performance-share-by-country
Estimates of GPU or equivalent resources of large AI players for 2024/5 - LessWrong, accessed November 26, 2025, https://www.lesswrong.com/posts/bdQhzQsHjNrQp7cNS/estimates-of-gpu-or-equivalent-resources-of-large-ai-players
AI's $600B Question - Sequoia Capital, accessed November 26, 2025, https://sequoiacap.com/article/ais-600b-question/
AI Investment Potential Accelerates as Demand Outpaces Capacity - the TCW Group, accessed November 26, 2025, https://www.tcw.com/Insights/2025/2025-11-25-AI-Investment-Potential
Chart: The Extreme Cost of Training AI Models | Statista, accessed November 26, 2025, https://www.statista.com/chart/33114/estimated-cost-of-training-selected-ai-models/
The 2025 AI Index Report | Stanford HAI, accessed November 26, 2025, https://hai.stanford.edu/ai-index/2025-ai-index-report
No comments:
Post a Comment