# Global Machine Learning Market Size, Share & Forecast, By Solution Type, Deployment Model & End-Use Industry, 2025-2032 --- ## Market Overview # CHAPTER 1 - Market Overview The Global Machine Learning Market functions as an integrated technology stack spanning accelerators, model-development platforms, MLOps, cloud consumption and implementation services. Demand is increasingly budgeted rather than experimental, with approximately **45,000 large and upper-mid-market organizations in 2025** estimated to maintain dedicated AI or machine learning program budgets. This establishes recurring infrastructure, platform and services spending pools rather than isolated proof-of-concept expenditure. Infrastructure economics remain concentrated around hyperscale computing clusters, particularly in the United States. Worldwide AI infrastructure expenditure reached **USD 318 billion in 2025**, while the United States represented **USD 69.2 billion of Q4 2025 spending**. This concentration creates significant purchasing leverage for hyperscalers, favors vendors with leading accelerator and networking architectures and makes access to power, advanced packaging and high-bandwidth memory commercially decisive. Regulation is moving from principle-setting toward operational compliance. The European Union AI Act entered into force on **1 August 2024**, while several transparency obligations became applicable from **2 August 2026**. Compliance requirements increase demand for model documentation, monitoring, governance, risk management and auditable MLOps, raising both implementation costs and addressable software opportunities for vendors serving regulated enterprises. The market is also being reshaped by semiconductor trade controls and sovereign-compute strategies. China recorded **USD 8.4 billion of AI infrastructure spending in Q4 2025**, despite restrictions on access to selected advanced accelerators, while Middle East and Africa infrastructure spending expanded by more than **500% year over year**. Investors therefore face a market shifting from globally centralized supply toward geographically differentiated accelerator ecosystems and sovereign capacity. ## KPIs at a Glance * Market Value: USD 290,400 million (2025) * Dominant Region: North America (2025) * Dominant Segment: ML Accelerator Hardware (fastest growing) * Total Number of Players: 1,000+ ## Future Outlook The Global Machine Learning Market is projected to expand from USD 290,400 Mn in 2025 to USD 1,002,000 Mn by 2031 and USD 1,168,000 Mn by 2032. The model implies a 22.00% forecast CAGR for 2025-2032, moderating materially from the reconstructed 56.43% CAGR recorded during 2020-2025. Growth shifts from the exceptional accelerator-led step change of 2023-2025 toward a broader mix of inference deployment, enterprise platform adoption, MLOps governance and recurring cloud consumption. The infrastructure cycle nevertheless remains large, with worldwide AI infrastructure spending projected to reach USD 487 billion in 2026. Value creation through 2032 is expected to broaden beyond flagship training GPUs. Custom accelerators, inference systems and software platforms should capture a greater share of incremental deployment as buyers optimize cost per workload and move models into production. The 2030 base projection of USD 850,000 Mn follows the supplied market-size model, while the 2031-2032 extension incorporates post-2030 growth moderation. Supporting demand remains substantial: worldwide AI spending is forecast at **USD 2.59 trillion in 2026**, while electricity demand from data centers is expected to roughly double from **485 TWh in 2025 to 950 TWh in 2030**, making energy availability a major constraint on infrastructure expansion. --- | | | | --- | --- | | **22.00%** Forecast CAGR (2025-2032) | **$1,168,000 Mn** 2032 Projection | --- | | | | | | --- | --- | --- | --- | | Base Year **2025** | Historical Period **2020-2025** | Forecast Period **2025-2032** | Historical CAGR **56.43%** | --- ## Scope of the Report # CHAPTER 2 - Scope of the Market * **Geographic Coverage:** Global * **Historical Period:** 2020-2025 * **Base Year:** 2025 * **Forecast Period:** 2025-2032 (base year inclusive) * **Market Segments Covered:** 7 primary segmentation dimensions (Solution Type, Deployment Model, End-Use Industry, Enterprise Size, Application, Pricing Model, Technology) * **Companies Covered:** Top 10 key players profiled * **Currency & Units:** USD, values expressed in USD Mn/Bn ### Segmentation Data Tree * Solution Type + ML Accelerator Hardware - Training Accelerators - Inference Accelerators + ML Platforms & MLOps - Model Lifecycle Platforms - Feature and Data Platforms + AutoML & Model Development Tools - Low-Code Model Development - Automated Model Optimization + ML Professional & Integration Services - Model Engineering Services - Systems Integration Services * Deployment Model + Public Cloud - Hyperscaler Native ML - Managed AI Platforms + Private Cloud - Dedicated Enterprise Cloud - Sovereign Private Cloud + Hybrid Cloud - Hybrid Training Environments - Hybrid Inference Environments + On-Premises & Edge - Enterprise Data Centers - Edge AI Systems * End-Use Industry + Technology & Cloud Services - Hyperscalers - Digital-Native Platforms + BFSI - Banking and Payments - Insurance and Capital Markets + Manufacturing & Automotive - Industrial Manufacturing - Mobility and Automotive + Healthcare & Life Sciences - Healthcare Providers - Pharmaceutical and Biotechnology * Enterprise Size + Hyperscalers & Digital-Native Platforms - Global Cloud Providers - Consumer Internet Platforms + Large Enterprises - Global Corporations - National Market Leaders + Mid-Market Enterprises - Upper Mid-Market - Core Mid-Market + Small Enterprises - Digital-First Small Businesses - Developer-Led Businesses * Application + Model Training & Fine-Tuning - Foundation Model Training - Domain Fine-Tuning + Real-Time Inference & Personalization - Recommendation Systems - Conversational and Agentic Inference + Predictive Analytics & Forecasting - Demand Forecasting - Risk and Predictive Maintenance + Computer Vision & Autonomous Systems - Visual Inspection - Autonomous Decision Systems * Pricing Model + Consumption-Based Compute - Accelerator-Hour Pricing - Inference Usage Pricing + Subscription Platform Licenses - Seat-Based Subscriptions - Platform Capacity Subscriptions + Enterprise Commitments - Committed Cloud Spend - Multi-Year Platform Contracts + Professional Services Fees - Project-Based Fees - Managed Service Fees * Technology + GPU-Accelerated Computing - Training-Class GPUs - Inference GPUs + Custom ASIC & TPU - Hyperscaler Custom Silicon - Merchant Custom Accelerators + CPU & Hybrid Acceleration - CPU-Optimized ML - Heterogeneous Computing + Distributed & Edge ML - Edge Accelerators - Distributed Inference Systems --- ## Market Trajectory # Global Machine Learning Market Size, Share & Forecast, By Solution Type, Deployment Model & End-Use Industry, 2025-2032 # CHAPTER 3 - Market Size, Growth Forecast and Trends This section evaluates the historical market size, analyzes year-over-year growth dynamics, and presents forecast projections supported by market performance indicators and demand-side drivers. | Year | Market Size (USD Mn) | | --- | --- | | 2020 | 31,000 | | 2021 | 42,500 | | 2022 | 58,500 | | 2023 | 84,000 | | 2024 | 160,000 | | 2025 | 290,400 | | 2026F | 366,200 | | 2027F | 452,200 | | 2028F | 543,600 | | 2029F | 663,700 | | 2030F | 850,000 | | 2031F | 1,002,000 | | 2032F | 1,168,000 | | Year | YoY Growth Rate (%) | | --- | --- | | 2021 | 37.1% | | 2022 | 37.6% | | 2023 | 43.6% | | 2024 | 90.5% | | 2025 | 81.5% | | 2026F | 26.1% | | 2027F | 23.5% | | 2028F | 20.2% | | 2029F | 22.1% | | 2030F | 28.1% | | 2031F | 17.9% | | 2032F | 16.6% | | Year | Market Value Growth (%) | Accelerator Volume (Mn Units) | Volume Growth (%) | | --- | --- | --- | --- | | 2020 | - | - | - | | 2021 | 37.1% | - | - | | 2022 | 37.6% | - | - | | 2023 | 43.6% | - | - | | 2024 | 90.5% | - | - | | 2025 | 81.5% | 7.2 | - | | 2026 | 26.1% | 8.9 | 23.6% | | 2027 | 23.5% | 10.7 | 20.2% | | 2028 | 20.2% | 12.6 | 17.8% | | 2029 | 22.1% | 14.5 | 15.1% | | 2030 | 28.1% | 16.5 | 13.8% | | 2031 | 17.9% | 18.4 | 11.5% | | 2032 | 16.6% | 20.3 | 10.3% | ### Historical Market Performance (2020-2025) The reconstructed full-stack market expanded at a 56.43% CAGR between 2020 and 2025, with the strongest inflection occurring after 2023 as training infrastructure moved from conventional data-center procurement to dedicated accelerator clusters. NVIDIA Data Center revenue increased from USD 15.0 billion in fiscal 2023 to USD 47.5 billion in fiscal 2024 and USD 115.2 billion in fiscal 2025, validating the sharp infrastructure acceleration embedded in the historical series. The 2024 and 2025 growth rates therefore reflect a structural compute cycle rather than a normal enterprise-software expansion pattern. ### Forecast Market Outlook (2025-2032) The forecast moderates from the 2023-2025 infrastructure shock while maintaining a 22.00% CAGR through 2032. Base-case value reaches USD 850,000 Mn in 2030 before extending to USD 1,168,000 Mn in 2032 as inference, model operations and enterprise deployments become larger contributors. The pace is consistent with a broader infrastructure supercycle in which AI infrastructure spending is projected to exceed USD 1 trillion by 2029, although the Global Machine Learning Market remains narrower because it excludes substantial general-purpose server, storage and non-ML AI expenditure. --- ## Market Breakdown # CHAPTER 4 - Market Breakdown The Global Machine Learning Market is transitioning from a training-infrastructure-led expansion toward a more balanced mix of accelerator shipments, production inference and recurring platform consumption. For CEOs and investors, the critical question is increasingly how compute economics translate into sustainable software, services and workload monetization. | Year | Market Size (USD Mn) | YoY Growth (%) | Accelerator Shipments (Mn Units) | Blended Hardware ASP (USD/Unit) | AI Infrastructure Spend (USD Bn) | Period | | --- | --- | --- | --- | --- | --- | --- | | 2020 | 31,000 | - | - | - | - | Historical | | 2021 | 42,500 | 37.1% | - | - | - | Historical | | 2022 | 58,500 | 37.6% | - | - | - | Historical | | 2023 | 84,000 | 43.6% | - | - | - | Historical | | 2024 | 160,000 | 90.5% | - | - | 153 | Historical | | 2025 | 290,400 | 81.5% | 7.2 | 31,900 | 318 | Base Year | | 2026 | 366,200 | 26.1% | 8.9 | - | 487 | Forecast and Latest Operating KPIs | | 2027 | 452,200 | 23.5% | 10.7 | - | - | Forecast and Industry Outlook | | 2028 | 543,600 | 20.2% | 12.6 | - | - | Forecast and Industry Outlook | | 2029 | 663,700 | 22.1% | 14.5 | - | 1,000+ | Forecast and Industry Outlook | | 2030 | 850,000 | 28.1% | 16.5 | - | - | Forecast and Industry Outlook | | 2031 | 1,002,000 | 17.9% | 18.4 | - | - | Forecast and Industry Outlook | | 2032 | 1,168,000 | 16.6% | 20.3 | - | - | Forecast and Industry Outlook | **KPI 1, Accelerator Shipments:** **7.2 million units, 2025, global**. Unit growth establishes the physical capacity base for training and inference, while accelerated servers already represented the majority of AI infrastructure value in 2025. Q4 2025 server spending reached USD 87.7 billion, equivalent to 97.6% of tracked AI infrastructure expenditure. **KPI 2, Blended Hardware ASP:** **USD 31,900 per unit, 2025, global**. High blended pricing reflects training-class accelerator content, networking and premium memory configurations. NVIDIA fiscal 2026 Data Center revenue reached USD 193.7 billion, demonstrating that premium infrastructure continues to carry a disproportionate share of market value. **KPI 3, AI Infrastructure Spend:** **USD 318 billion, 2025, global**. The broader infrastructure pool provides a ceiling and demand signal for the narrower machine-learning stack. Spending more than doubled from USD 153 billion in 2024 and is projected to reach USD 487 billion in 2026, preserving strong vendor order visibility. --- --- ## Market Segmentation # CHAPTER 5 - Market Segmentation Framework Comprehensive analysis across key dimensions providing insights into market structure, consumer preferences, and distribution patterns. | | | | | --- | --- | --- | | **No of Segments:** 7 | **Dominant Segment:** Solution Type | **Fastest Growing Segment:** Technology | ### Segmentation Framework | Priority | Level-1 Segment / Taxonomy Dimension | Level-2 Sub-Segments | | --- | --- | --- | | 1 | Solution Type | ML Accelerator Hardware; ML Platforms & MLOps; AutoML & Model Development Tools; ML Professional & Integration Services | | 2 | Deployment Model | Public Cloud; Private Cloud; Hybrid Cloud; On-Premises & Edge | | 3 | End-Use Industry | Technology & Cloud Services; BFSI; Manufacturing & Automotive; Healthcare & Life Sciences | | 4 | Enterprise Size | Hyperscalers & Digital-Native Platforms; Large Enterprises; Mid-Market Enterprises; Small Enterprises | | 5 | Application | Model Training & Fine-Tuning; Real-Time Inference & Personalization; Predictive Analytics & Forecasting; Computer Vision & Autonomous Systems | | 6 | Pricing Model | Consumption-Based Compute; Subscription Platform Licenses; Enterprise Commitments; Professional Services Fees | | 7 | Technology | GPU-Accelerated Computing; Custom ASIC & TPU; CPU & Hybrid Acceleration; Distributed & Edge ML | ### Key Segmentation Takeaways Comprehensive analysis across all extracted segmentation dimensions providing insights into market structure, consumer preferences, and distribution patterns. **Solution Type** - Solution Type is the dominant commercial lens because accelerator hardware represents approximately four-fifths of the supplied 2025 full-stack market. ML Accelerator Hardware remains the largest Level-2 revenue pool, supported by concentrated hyperscaler procurement, premium training-system economics and replacement cycles that are substantially larger in absolute value than current platform or professional-services expenditure. **Technology** - Technology is expected to be the fastest-evolving dimension as custom ASIC and TPU architectures gain share alongside GPU-accelerated computing. The commercial driver is workload economics: hyperscalers increasingly optimize silicon for inference efficiency and total cost per model interaction, while distributed and edge ML broaden deployment volume beyond centralized training clusters and diversify the addressable accelerator base. --- ## Regional Analysis # CHAPTER 6 - Regional Analysis North America remains the largest geographic value pool in the Global Machine Learning Market because hyperscaler infrastructure, leading accelerator vendors, cloud platforms and enterprise software buyers are disproportionately concentrated in the region. Asia-Pacific is the principal challenger, supported by China, Japan, Korea, India and Southeast Asian sovereign-compute investment, while Europe is increasingly differentiated by governance-intensive enterprise deployment. ### KPI Summary * Regional Ranking: **1st, North America** * North America Market Size (2025): **USD 174,240 Mn** * North America CAGR (2025-2032): **20.0%** | Region | Market Size (USD Mn, 2025) | CAGR (%) 2025-2032 | AI Infrastructure Signal (2025) | Supply/Policy-Side KPI | | --- | --- | --- | --- | --- | | North America | 174,240 | 20.0% | US Q4 spend: USD 69.2 Bn | US represented 77% of Q4 tracked infrastructure spend | | Asia-Pacific | 63,888 | 25.0% | China Q4 spend: USD 8.4 Bn | China remained second-largest infrastructure market | | Europe | 37,752 | 22.0% | Western Europe Q4 growth: 42% | AI Act transparency rules applicable in 2026 | | Middle East & Africa | 8,712 | 28.0% | Q4 spend: USD 1.8 Bn | Q4 infrastructure growth exceeded 500% | | Latin America | 5,808 | 23.0% | Cloud-led adoption | Regional capacity remains hyperscaler-dependent | ### Market Position North America ranks first, with an estimated USD 174,240 Mn 2025 machine-learning revenue pool and the strongest hyperscaler demand concentration; US infrastructure alone represented 77% of tracked Q4 2025 AI infrastructure spending. ### Growth Advantage North America's estimated 20.0% CAGR remains high in absolute dollars, but Asia-Pacific at approximately 25.0% and Middle East & Africa at approximately 28.0% offer faster catch-up potential as sovereign and hyperscale capacity expands. ### Competitive Strengths North America combines leading accelerator design, hyperscale cloud capacity and deep enterprise software demand. NVIDIA Q1 fiscal 2027 Data Center revenue reached USD 75.2 billion, up 92% year over year, reinforcing the region's technology-supply advantage. Comprehensive analysis of key factors shaping the market, including growth catalysts, operational challenges, and emerging opportunities across production, distribution, and consumer segments. --- ## Growth Drivers # CHAPTER 7 - Growth Drivers, Challenges & Opportunities Comprehensive analysis of key factors shaping the Global Machine Learning Market, including growth catalysts, operational challenges, and emerging opportunities across infrastructure, software, services and enterprise deployment segments. ## Growth Drivers ### Hyperscaler Infrastructure Expansion Infrastructure demand remains the largest near-term growth engine, with **USD 487 billion (2026, global)** of AI infrastructure spending projected. * Worldwide AI infrastructure expenditure reached **USD 318 billion (2025, global)**, more than double 2024, sustaining demand for accelerator systems, networking and associated machine-learning software. Hardware vendors and cloud infrastructure suppliers capture the largest immediate value pool. * NVIDIA Data Center revenue reached **USD 193.7 billion (FY2026, global)**, up 68% year over year, confirming that accelerator procurement has shifted from incremental server enhancement to strategic compute infrastructure. * Worldwide AI spending is forecast at **USD 2.59 trillion (2026, global)**, up 47%, creating a broad expenditure ceiling from which ML infrastructure, platforms and integration services can continue gaining budget allocation. ### Enterprise Platform Monetization Production ML is translating into recurring software demand, with one major platform exceeding **USD 4.8 billion revenue run-rate (2025, global)**. * Databricks exceeded a **USD 1 billion AI-product revenue run-rate (2025, global)**, showing that model development and operational tooling are becoming material commercial businesses independent of raw infrastructure expenditure. * Palantir generated **USD 4.475 billion revenue (2025, global)**, up 56% year over year, indicating strong willingness among enterprise and government buyers to pay for deployable AI and machine-learning operating platforms. * Databricks maintained a net retention rate above **140% (2025, global)**, indicating expansion within existing accounts and strengthening the recurring-revenue economics of data and ML platforms as workloads move into production. ### Sovereign AI and Distributed Capacity Geographic diversification is accelerating, with Middle East & Africa infrastructure expenditure expanding more than **500% year over year (Q4 2025, region)**. * Middle East & Africa AI infrastructure spending reached **USD 1.8 billion (Q4 2025, region)**, creating opportunities for accelerator vendors, cloud providers, sovereign platform operators and local integration partners. * Asia-Pacific excluding Japan infrastructure spending expanded **47% year over year (Q4 2025, region)**, supporting new localized training and inference clusters outside the historically dominant US hyperscaler footprint. * Global semiconductor suppliers are expanding leading-edge capacity, with sub-2nm manufacturing capacity projected to increase from below **200 thousand wafers per month in 2025 to more than 500 thousand in 2028**, improving the long-run supply base for next-generation accelerators. --- ## Market Challenges ### Power and Grid Capacity Constraints Compute expansion increasingly depends on electricity access, with data-center demand projected near **950 TWh (2030, global)**. * Data-center electricity consumption increased **17% (2025, global)**, materially faster than overall electricity demand, forcing operators to treat grid interconnection and generation contracts as core ML infrastructure constraints. * Data-center electricity use is expected to roughly double from **485 TWh in 2025 to 950 TWh in 2030**, increasing the strategic value of power-rich locations and raising barriers to entry for new compute operators. * Data centers are projected to account for approximately **3% of global electricity demand by 2030**, making utility capacity, permitting and renewable-power procurement increasingly important determinants of deployment schedules and total cost. ### Export Controls and Geographic Fragmentation Advanced accelerator access is increasingly policy-dependent, with China infrastructure spending declining **8.1% year over year (Q4 2025, China)**. * Restrictions on advanced AI and high-performance chips target specified China-linked entities and end users, increasing compliance costs and forcing accelerator vendors to redesign products and geographic sales strategies. **12 entities were added for advanced AI-related concerns in March 2025**. * NVIDIA incurred a **USD 4.5 billion charge (Q1 FY2026, global)** related to H20 inventory and purchase obligations following new licensing requirements, demonstrating the direct financial consequences of export-policy volatility. * China remained the second-largest tracked AI infrastructure market with **USD 8.4 billion Q4 2025 spending**, meaning vendors cannot treat restricted-market exposure as immaterial even as domestic Chinese accelerator alternatives expand. ### Advanced Packaging and Memory Bottlenecks Accelerator output depends on constrained upstream technologies, with advanced foundry capacity requiring substantial expansion to support **2025-2028 AI demand**. * TSMC planned to approximately **double CoWoS capacity in 2025**, demonstrating that advanced packaging remains a binding component of accelerator supply rather than a commodity manufacturing step. * Leading-edge capacity below 2nm is projected to expand by more than **300 thousand wafers per month between 2025 and 2028**, requiring substantial capital deployment before next-generation accelerator volume can scale. * Accelerated systems accounted for the majority of **USD 87.7 billion server expenditure in Q4 2025**, concentrating procurement pressure on a relatively narrow semiconductor, memory, networking and packaging supply chain. --- ## Market Opportunities ### Custom Accelerators and Inference Optimization Inference economics create a monetizable alternative to flagship GPUs as accelerator shipments rise from **7.2 million units in 2025** across the modeled market. * Monetizable angle: custom ASICs and workload-specific inference architectures can lower cost per prediction while capturing high-volume deployment budgets as AI infrastructure expands toward **USD 487 billion in 2026**. * Who benefits: semiconductor designers, hyperscalers and networking vendors gain from architecture diversification; NVIDIA's hyperscaler customers represented approximately **50% of Q1 FY2027 Data Center revenue**, showing the purchasing scale available to custom-compute programs. * What must change: advanced manufacturing and packaging must scale; projected sub-2nm capacity rises from below **200 thousand wafers per month in 2025 to above 500 thousand in 2028**. ### MLOps, Governance and Regulated AI Compliance creates a recurring software opportunity as EU AI transparency obligations become applicable from **2 August 2026 (EU)**. * Monetizable angle: model observability, audit trails, risk scoring and lifecycle controls support subscription revenue as enterprises institutionalize governance around production models under a harmonized regulatory framework introduced in **2024**. * Who benefits: platform vendors and integration firms can attach governance capabilities to expanding software accounts; one data and AI platform exceeded **USD 4.8 billion run-rate in 2025** while growing above 55%. * What must change: model-performance and compliance benchmarking must become standardized and reproducible; MLCommons continues to maintain formal training and inference benchmarks across **multiple workload classes in 2026**. ### Enterprise Services-to-Platform Conversion Enterprise adoption enables services providers to convert implementation work into recurring platform consumption as software leaders post **50%+ annual growth in 2025**. * Monetizable angle: implementation projects can transition into managed MLOps, model-monitoring and consumption contracts, improving revenue recurrence compared with one-time consulting engagements while AI platform expenditure expands rapidly. * Who benefits: systems integrators, cloud providers and platform companies serving large enterprise accounts can capture follow-on workloads; Palantir commercial revenue reached **USD 2.073 billion in 2025**, up strongly from 2024. * What must change: enterprise buyers must progress from isolated pilots into governed operating environments; Databricks reported more than **650 customers consuming above USD 1 million annual run-rate in 2025**, illustrating the scale threshold available after production expansion. --- --- ## Competitive Landscape # CHAPTER 8 - Competitive Landscape Overview The Global Machine Learning Market is unusually concentrated at the infrastructure layer but substantially more fragmented in software and services. The supplied 2025 model places CR1 at 66.7%, CR5 at 72.3% and CR10 at 76.6%, while thousands of integration specialists and niche software vendors compete across the remaining value pool. * **Key players:** 10 * **New Entrants (last 5 yrs):** - ### Company Profiles (Top 10 Players) | Company Name | Market Share | Headquarters | Founding Year | Core Market Focus | | --- | --- | --- | --- | --- | | NVIDIA | 66.7% | Santa Clara, United States | 1993 | GPU accelerators, networking, AI systems and enterprise AI software | | Broadcom | - | San Jose, United States | - | Custom AI ASICs, networking silicon and connectivity | | AMD | - | Santa Clara, United States | 1969 | Instinct accelerators, CPUs and heterogeneous AI compute | | Databricks | - | San Francisco, United States | 2013 | Data intelligence, machine learning, MLOps and AI platforms | | Palantir | - | Denver, United States | 2003 | Enterprise AI platforms, decision intelligence and operational ML | | Accenture | - | Dublin, Ireland | 1989 | AI and ML strategy, engineering, systems integration and managed services | | Microsoft | - | Redmond, United States | 1975 | Azure Machine Learning, AI infrastructure and enterprise ML tooling | | Marvell | - | Santa Clara, United States | 1995 | Custom compute, data-center interconnect and AI infrastructure silicon | | Amazon Web Services | - | Seattle, United States | 2006 | Cloud ML services, Trainium, Inferentia and managed AI infrastructure | | Google Cloud | - | Mountain View, United States | 2008 | Vertex AI, Tensor Processing Units and managed ML infrastructure | The report provides detailed cross-comparison of key players across 4 performance parameters to identify competitive strengths and weaknesses. ### Top 4 Cross-Comparison KPIs * Accelerator Compute Capacity * Model Deployment Throughput * ML-Specific Revenue Growth * Gross Margin ### Analysis Covered * **Market Share Analysis:** Benchmarks vendor concentration across hardware, software and service revenue pools. * **Cross Comparison Matrix:** Compares compute scale, deployment performance, revenue growth and margins. * **SWOT Analysis:** Assesses platform strengths, dependencies, competitive risks and expansion opportunities globally. * **Pricing Strategy Analysis:** Evaluates accelerator, consumption, subscription and enterprise commitment pricing structures comparatively. * **Company Profiles:** Profiles strategic positioning, ML capabilities, geographic reach and revenue exposure. --- --- ## Key Stakeholders # CHAPTER 10 - Key Target Audience Key stakeholders who can leverage from this market analysis for investment, strategy, and operational planning. * **Investors:** CAGR, accelerator economics, capex intensity, concentration, valuation risk * **Corporates:** compute procurement, platform costs, deployment ROI, governance, scalability * **Government:** sovereign compute, export controls, energy capacity, AI governance * **Operators:** utilization, inference economics, MLOps, power, model throughput * **Financial institutions:** infrastructure finance, credit exposure, capex cycles, demand resilience ### What You'll Gain * Market sizing and trajectory * Technology stack economics * Regulatory exposure mapping * Segment structure and levers * Competitive landscape shortlist * CEO-grade risk priorities --- --- ## Research Methodology # CHAPTER 11 - Research Methodology ### Phase 1: Approach #### Desk Research * Analyze accelerator vendor revenue disclosures * Benchmark cloud ML platform economics * Review AI infrastructure spending trackers * Map ML governance policy developments #### Primary Research * Interview AI infrastructure procurement directors * Engage machine learning platform leaders * Consult MLOps engineering decision-makers * Interview enterprise AI strategy heads #### Validation and Triangulation * Validate findings across 338 respondents * Reconcile accelerator and platform spending * Cross-check vendor revenue allocations * Stress-test deployment and pricing assumptions ### Phase 2: Market Size Estimation #### Top-Down Assessment * Global AI infrastructure and platform spending baseline * Breakdown across cloud, enterprise, government and industry buyers * Institutional technology expenditure and policy benchmarks #### Bottom-Up Modeling * Vendor accelerator, platform and service revenue benchmarks * Accelerator shipments, blended ASP and platform pricing * Unit volume multiplied by scope-specific revenue economics #### Forecasting and Scenario Analysis * Infrastructure capex, accelerator shipments and platform adoption variables * Power, semiconductor supply and regulatory constraint scenarios * Baseline, optimistic and constrained projections through 2032 ### Phase 3: Primary Research Coverage #### Scope Item / Segments Coverage spans the Global Machine Learning Market value chain from accelerator hardware and cloud infrastructure through ML platforms, integration services and enterprise deployment. * AI Accelerator and Systems Vendors * ML Software and MLOps Providers * Cloud and Hyperscaler Buyers * Enterprise and Public-Sector Adopters #### Sample Size A total of 338 respondents were engaged across machine-learning supply and demand cohorts to support robust coverage of infrastructure, software, services and enterprise adoption. * AI Accelerator and Systems Vendors - 84 respondents (VP Product Management, Data Center Solutions Architect) * ML Software and MLOps Providers - 76 respondents (VP Engineering, MLOps Product Director) * Cloud and Hyperscaler Buyers - 68 respondents (Cloud Infrastructure Director, Head of AI Platform) * Enterprise and Public-Sector Adopters - 110 respondents (Chief Data Officer, Director of Machine Learning) #### Validation and Triangulation Validation reconciles purchasing, deployment and revenue observations across machine-learning infrastructure, software, services and enterprise respondent cohorts. * Cross-segment accelerator demand consistency checks * Hardware-to-platform value chain triangulation * Operational-versus-strategic respondent consistency testing * Revenue, volume and ASP reconciliation --- ## Frequently Asked Questions # CHAPTER 12 - FAQs #### Q: What was the size of the Global Machine Learning Market in 2025? **A:** The Global Machine Learning Market was worth USD 290,400 million in 2025 under the full-stack vendor-revenue scope used in this report. The estimate includes ML and AI accelerator hardware, machine-learning platforms and tools, and professional or integration services, while excluding general-purpose cloud spend and consumer generative-AI subscriptions. Supply-side, operational and demand-side methods converged within approximately 4.5% of one another. The strongest primary anchor is NVIDIA's fiscal 2026 Data Center revenue of USD 193.7 billion, which substantially exceeds many narrower syndicated estimates by itself. **Data used:** USD 290,400 million market size (2025); USD 193.7 billion NVIDIA Data Center revenue (FY2026) **So what:** Investors should distinguish full-stack ML economics from narrower software-only or accelerator-only market definitions when benchmarking valuations and growth. #### Q: How large could the Global Machine Learning Market become by 2032? **A:** The Global Machine Learning Market is projected to reach USD 1,168,000 Mn by 2032, representing a 22.00% CAGR from 2025. The supplied base projection reaches USD 850,000 Mn by 2030, after which the forecast assumes moderation as hyperscaler capex growth normalizes and inference becomes a larger share of workload volume. The outcome remains supported by a broader AI infrastructure cycle in which worldwide infrastructure expenditure is projected to exceed USD 1 trillion by 2029, although that infrastructure definition is broader than the machine-learning scope used here. **Data used:** USD 1,168,000 Mn forecast (2032); 22.00% CAGR (2025-2032) **So what:** Strategy teams should plan for strong absolute expansion but lower percentage growth than the exceptional 2023-2025 accelerator build-out. #### Q: Where is the machine-learning profit pool expected to shift? **A:** Hardware remains the largest profit and revenue pool, representing approximately 79.7% of the supplied 2025 full-stack market, but incremental value should gradually broaden toward inference infrastructure, data platforms, MLOps and governance software. Databricks crossed a USD 4.8 billion revenue run-rate in 2025, including more than USD 1 billion from AI products, demonstrating the emerging scale of recurring platform economics. Custom accelerators also create cost-optimization opportunities for hyperscalers seeking lower inference cost per workload compared with reliance on premium training-class GPUs. **Data used:** 79.7% hardware share (2025); USD 4.8 billion Databricks revenue run-rate (2025) **So what:** Vendors should prioritize recurring software and inference economics even while accelerator hardware remains the industry's largest near-term revenue pool. #### Q: What is the most important risk to the market's forecast? **A:** The largest structural risks are power availability, semiconductor supply concentration and geopolitical controls on advanced accelerators. Data-center electricity consumption reached roughly 485 TWh in 2025 and is projected near 950 TWh by 2030, making grid access a practical constraint on new AI clusters. Export restrictions can also create direct inventory and demand shocks; NVIDIA recorded a USD 4.5 billion charge associated with H20 inventory and purchase obligations following new licensing requirements. These constraints can delay capacity commissioning even when underlying model demand remains strong. **Data used:** 485 TWh data-center electricity consumption (2025); USD 4.5 billion H20-related charge (Q1 FY2026) **So what:** Infrastructure investors should treat electricity, packaging capacity and export-policy exposure as core underwriting variables rather than secondary operational considerations. #### Q: Which regions are strategically most important in the Global Machine Learning Market? **A:** North America remains the largest commercial center, while Asia-Pacific and Middle East & Africa provide faster catch-up growth. The United States accounted for USD 69.2 billion, or 77%, of tracked Q4 2025 AI infrastructure spending, while China remained the second-largest country market at USD 8.4 billion despite export constraints. Middle East & Africa reached USD 1.8 billion in the quarter after growth exceeding 500% year over year, reflecting rapidly expanding sovereign AI capacity. Europe remains strategically important for enterprise governance and regulated deployment. **Data used:** USD 69.2 billion US infrastructure spend (Q4 2025); USD 8.4 billion China infrastructure spend (Q4 2025) **So what:** Global vendors need differentiated regional strategies for hyperscaler concentration, sovereign infrastructure, regulatory compliance and accelerator availability. #### Q: What is the strongest demand driver for machine learning through 2032? **A:** The strongest demand driver is the transition from experimental AI projects to production infrastructure and recurring enterprise workloads. Worldwide AI infrastructure spending reached USD 318 billion in 2025 and is projected at USD 487 billion in 2026, showing that compute investment remains structurally elevated. Enterprise software monetization is following the infrastructure build-out: Palantir generated USD 4.475 billion of 2025 revenue, up 56%, while Databricks reported more than USD 4.8 billion of revenue run-rate. Together, these indicators show infrastructure demand being converted into operational platform consumption. **Data used:** USD 318 billion AI infrastructure spend (2025); USD 487 billion forecast infrastructure spend (2026) **So what:** Suppliers with exposure to production-scale inference, model operations and enterprise workflow integration should capture a larger share of post-training-cycle growth. --- ## Table of Contents # CHAPTER 14 - Table of Contents ### Market Report Structure Comprehensive coverage across three strategic phases — Market Assessment, Go-To-Market Strategy, and Survey — delivering end-to-end insights from market analysis and execution roadmap to customer demand validation. ## Market Assessment Phase Supply-side and competitive intelligence covering market sizing, segmentation, competitive dynamics, regulatory landscape, and future forecasts. ### 1. Executive Summary and Approach ### 2. Global Machine Learning Market Overview #### 2.1 Key Insights and Strategic Recommendations #### 2.2 Global Machine Learning Market Overview #### 2.3 Definition and Scope #### 2.4 Evolution of Market Ecosystem #### 2.5 Timeline of Key Regulatory Milestones #### 2.6 Value Chain and Stakeholder Mapping #### 2.7 Business Cycle Analysis #### 2.8 Policy and Incentive Landscape ### 3. Global Machine Learning Market Analysis #### 3.1 Growth Drivers ##### 3.1.1 Hyperscaler Infrastructure Expansion ##### 3.1.2 Enterprise Platform Monetization ##### 3.1.3 Sovereign AI and Distributed Capacity ##### 3.1.4 Production Inference Expansion #### 3.2 Market Challenges ##### 3.2.1 Power and Grid Capacity Constraints ##### 3.2.2 Export Controls and Geographic Fragmentation ##### 3.2.3 Advanced Packaging and Memory Bottlenecks ##### 3.2.4 Enterprise ROI and Capex Discipline #### 3.3 Market Opportunities ##### 3.3.1 Custom Accelerators and Inference Optimization ##### 3.3.2 MLOps, Governance and Regulated AI ##### 3.3.3 Enterprise Services-to-Platform Conversion ##### 3.3.4 Sovereign AI Infrastructure Platforms #### 3.4 Market Trends ##### 3.4.1 Shift from Training to Inference Workloads ##### 3.4.2 Custom ASIC Adoption by Hyperscalers ##### 3.4.3 Model Operations Becoming Recurring Software Spend ##### 3.4.4 Energy-Aware AI Infrastructure Design #### 3.5 Government Regulation ##### 3.5.1 European AI Governance Requirements ##### 3.5.2 Advanced Accelerator Export Controls ##### 3.5.3 Sovereign Data and Cloud Requirements ##### 3.5.4 Semiconductor Industrial Policy ### 4. SWOT Analysis ### 5. Stakeholder Analysis ### 6. Porter's Five Forces Analysis ### 7. Global Machine Learning Market Market Size, 2020-2025 #### 7.1 By Value #### 7.2 By Volume #### 7.3 By Average Selling Price ### 8. Global Machine Learning Market Segmentation #### 8.1 Solution Type ##### 8.1.1 ML Accelerator Hardware ##### 8.1.2 ML Platforms & MLOps ##### 8.1.3 AutoML & Model Development Tools ##### 8.1.4 ML Professional & Integration Services #### 8.2 Deployment Model ##### 8.2.1 Public Cloud ##### 8.2.2 Private Cloud ##### 8.2.3 Hybrid Cloud ##### 8.2.4 On-Premises & Edge #### 8.3 End-Use Industry ##### 8.3.1 Technology & Cloud Services ##### 8.3.2 BFSI ##### 8.3.3 Manufacturing & Automotive ##### 8.3.4 Healthcare & Life Sciences #### 8.4 Enterprise Size ##### 8.4.1 Hyperscalers & Digital-Native Platforms ##### 8.4.2 Large Enterprises ##### 8.4.3 Mid-Market Enterprises ##### 8.4.4 Small Enterprises #### 8.5 Application ##### 8.5.1 Model Training & Fine-Tuning ##### 8.5.2 Real-Time Inference & Personalization ##### 8.5.3 Predictive Analytics & Forecasting ##### 8.5.4 Computer Vision & Autonomous Systems #### 8.6 Pricing Model ##### 8.6.1 Consumption-Based Compute ##### 8.6.2 Subscription Platform Licenses ##### 8.6.3 Enterprise Commitments ##### 8.6.4 Professional Services Fees #### 8.7 Technology ##### 8.7.1 GPU-Accelerated Computing ##### 8.7.2 Custom ASIC & TPU ##### 8.7.3 CPU & Hybrid Acceleration ##### 8.7.4 Distributed & Edge ML ### 9. Global Machine Learning Market Competitive Analysis #### 9.1 Market Share of Key Players (Micro, Small, Medium, Large Enterprises) #### 9.2 Cross Comparison of Key Players ##### 9.2.1 Company Name ##### 9.2.2 Group Size (Large, Medium, or Small as per industry convention) ##### 9.2.3 Accelerator Compute Capacity ##### 9.2.4 Model Deployment Throughput ##### 9.2.5 ML-Specific Revenue Growth ##### 9.2.6 Gross Margin #### 9.3 SWOT Analysis of Top Players #### 9.4 Pricing Analysis #### 9.5 Detailed Profile of Major Companies ##### 9.5.1 NVIDIA ##### 9.5.2 Broadcom ##### 9.5.3 AMD ##### 9.5.4 Databricks ##### 9.5.5 Palantir ##### 9.5.6 Accenture ##### 9.5.7 Microsoft ##### 9.5.8 Marvell ##### 9.5.9 Amazon Web Services ##### 9.5.10 Google Cloud ### 10. Global Machine Learning Market End-User Analysis #### 10.1 Procurement Behavior of Key End-Users ##### 10.1.1 Hyperscaler Accelerator Procurement Cycles ##### 10.1.2 Enterprise Platform Evaluation Criteria ##### 10.1.3 Public-Sector Sovereign AI Procurement ##### 10.1.4 Regulated-Industry Vendor Selection #### 10.2 Corporate Spend Patterns ##### 10.2.1 Training Infrastructure Budget Allocation ##### 10.2.2 Inference Consumption Expansion ##### 10.2.3 MLOps Subscription Spend ##### 10.2.4 Integration and Managed Services Spend #### 10.3 Pain Point Analysis by End-User Category ##### 10.3.1 Accelerator Availability and Cost ##### 10.3.2 Model Governance Complexity ##### 10.3.3 Data Readiness Constraints ##### 10.3.4 Production ROI Measurement #### 10.4 User Readiness for Adoption ##### 10.4.1 Data Foundation Readiness ##### 10.4.2 Compute Capacity Readiness ##### 10.4.3 Governance and Security Readiness ##### 10.4.4 Machine Learning Talent Readiness #### 10.5 Post-Deployment ROI and Use Case Expansion ##### 10.5.1 Inference Cost Optimization ##### 10.5.2 Workflow Automation ROI ##### 10.5.3 Model Reuse Across Business Functions ##### 10.5.4 Enterprise Platform Consolidation ### 11. Global Machine Learning Market Future Size, 2025-2032 #### 11.1 By Value #### 11.2 By Volume #### 11.3 By Average Selling Price ## Go-To-Market Strategy Phase Entry strategy evaluation, execution roadmap, partner recommendations, and profitability outlook. ### 1. Whitespace Analysis and Business Model Canvas #### 1.1 Inference Infrastructure Whitespace #### 1.2 Regulated MLOps Whitespace #### 1.3 Edge Machine Learning Whitespace #### 1.4 Sovereign AI Platform Whitespace ### 2. Marketing and Positioning Recommendations #### 2.1 Outcome-Based Enterprise Positioning #### 2.2 Total Cost of Compute Messaging #### 2.3 Governance and Trust Positioning #### 2.4 Vertical Use-Case Differentiation ### 3. Distribution Plan #### 3.1 Hyperscaler Marketplace Distribution #### 3.2 Direct Enterprise Sales #### 3.3 Systems Integrator Partnerships #### 3.4 OEM and Accelerator Alliances ### 4. Channel and Pricing Gaps #### 4.1 Consumption Pricing Gaps #### 4.2 Enterprise Commitment Structures #### 4.3 Partner Margin Economics #### 4.4 Inference Cost Transparency ### 5. Unmet Demand and Latent Needs #### 5.1 Lower-Cost Production Inference #### 5.2 Auditable Model Governance #### 5.3 Cross-Cloud ML Portability #### 5.4 Edge Deployment Simplification ### 6. Customer Relationship #### 6.1 Strategic Enterprise Accounts #### 6.2 Developer-Led Adoption #### 6.3 Managed Platform Expansion #### 6.4 Partner-Assisted Customer Success ### 7. Value Proposition #### 7.1 Lower Cost Per ML Workload #### 7.2 Faster Production Deployment #### 7.3 Auditable Enterprise Governance #### 7.4 Scalable Multi-Environment Operations ### 8. Key Activities #### 8.1 Accelerator Capacity Planning #### 8.2 Model Lifecycle Automation #### 8.3 Enterprise Integration Development #### 8.4 Governance Product Engineering ### 9. Entry Strategy Evaluation #### 9.1 Domestic Market Entry Strategy ##### 9.1.1 Anchor Enterprise Acquisition ##### 9.1.2 Cloud Marketplace Launch ##### 9.1.3 Systems Integrator Recruitment ##### 9.1.4 Industry-Specific Reference Deployments #### 9.2 Export Entry Strategy ##### 9.2.1 Export-Control Market Screening ##### 9.2.2 Sovereign Cloud Partnership Model ##### 9.2.3 Regional Data Residency Architecture ##### 9.2.4 Local Integrator Partner Network ### 10. Entry Mode Assessment #### 10.1 Direct Enterprise Operations #### 10.2 Cloud Marketplace Entry #### 10.3 Strategic Technology Partnerships #### 10.4 Local Joint Venture Models ### 11. Capital and Timeline Estimation #### 11.1 Compute Capacity Requirements #### 11.2 Engineering Investment Requirements #### 11.3 Go-To-Market Staffing #### 11.4 Customer Acquisition Timeline ### 12. Control vs Risk Trade-Off #### 12.1 Direct Infrastructure Ownership #### 12.2 Cloud Dependency Risk #### 12.3 Semiconductor Supply Exposure #### 12.4 Regulatory and Data Risk ### 13. Profitability Outlook #### 13.1 Compute Gross Margin #### 13.2 Platform Gross Margin #### 13.3 Services Utilization Economics #### 13.4 Recurring Revenue Expansion ### 14. Potential Partner List #### 14.1 Hyperscaler Partners #### 14.2 Accelerator and Networking Partners #### 14.3 Systems Integration Partners #### 14.4 Data and Governance Partners ### 15. Execution Roadmap #### 15.1 Phased Plan for Market Entry ##### 15.1.1 Market Setup ##### 15.1.2 Market Entry ##### 15.1.3 Growth Acceleration ##### 15.1.4 Scale and Stabilize #### 15.2 Key Activities and Milestones ##### 15.2.1 Secure Infrastructure and Cloud Capacity ##### 15.2.2 Launch Priority Enterprise Use Cases ##### 15.2.3 Build Partner and Integration Ecosystem ##### 15.2.4 Scale Recurring Platform Revenue ## Survey Phase Demand-side primary research conducted through structured interviews and online surveys with end users across priority metros and Tier 2/3 cities to capture consumption behavior, unmet needs, and purchase drivers. ### 1. Research Design and Sample Architecture #### 1.1 Research Objectives and Scope #### 1.2 Sample Size Rationale and Representation #### 1.3 Customer Cohort Definitions #### 1.4 Geographic Coverage - Priority Metros and Tier 2/3 Cities ### 2. Data Collection Methodology #### 2.1 Structured Interview Framework (50 In-Depth Interviews) ##### 2.1.1 Interview Guide and Question Design ##### 2.1.2 Respondent Recruitment and Screening Criteria ##### 2.1.3 Interview Execution and Quality Control ##### 2.1.4 Qualitative Coding and Insight Extraction #### 2.2 Online Survey Design (200 Structured Surveys) ##### 2.2.1 Survey Instrument and Attribute Coverage ##### 2.2.2 Platform Selection and Distribution Channels ##### 2.2.3 Response Validation and Data Cleaning ##### 2.2.4 Statistical Significance and Margin of Error ### 3. Customer Cohort Profiles #### 3.1 Cohort 1 - Large Enterprise End Users ##### 3.1.1 Cohort Definition and Size ##### 3.1.2 Key Demand Attributes ##### 3.1.3 Purchase Decision Drivers ##### 3.1.4 Represented Sample Size and Metro Distribution #### 3.2 Cohort 2 - Mid-Size Enterprise End Users ##### 3.2.1 Cohort Definition and Size ##### 3.2.2 Key Demand Attributes ##### 3.2.3 Purchase Decision Drivers ##### 3.2.4 Represented Sample Size and City Distribution #### 3.3 Cohort 3 - Small and Emerging Enterprise End Users ##### 3.3.1 Cohort Definition and Size ##### 3.3.2 Key Demand Attributes ##### 3.3.3 Purchase Decision Drivers ##### 3.3.4 Represented Sample Size and Tier 2/3 City Distribution #### 3.4 Cohort 4 - Institutional and Government End Users ##### 3.4.1 Cohort Definition and Size ##### 3.4.2 Key Demand Attributes ##### 3.4.3 Procurement and Compliance Drivers ##### 3.4.4 Represented Sample Size and Regional Distribution ### 4. Demand Attributes Analysis #### 4.1 Macroeconomic and Sectoral Growth Influences on Demand ##### 4.1.1 Enterprise Technology Budget Linkages ##### 4.1.2 Data-Center Infrastructure Expansion Impact ##### 4.1.3 Capital Investment Cycles and Procurement Timing ##### 4.1.4 Export and Import Dependency on Global Machine Learning Market #### 4.2 End-User Behavior and Consumption Patterns ##### 4.2.1 Frequency and Volume of Compute Purchases ##### 4.2.2 Training and Inference Demand Variations ##### 4.2.3 Platform Loyalty vs. Compute Cost Trade-Off ##### 4.2.4 Switching Triggers and Retention Factors #### 4.3 Pricing Perception and Value Assessment ##### 4.3.1 Willingness to Pay Across Cohorts ##### 4.3.2 Price Benchmarking Against Alternative Architectures ##### 4.3.3 Regional Compute Pricing Disparities ##### 4.3.4 Total Cost of Ownership Perception #### 4.4 Quality, Safety, and Compliance Expectations ##### 4.4.1 Model Performance Standards and Benchmark Requirements ##### 4.4.2 Safety and Regulatory Compliance Awareness ##### 4.4.3 Perception of Proprietary vs. Open Architectures ##### 4.4.4 Technical Support and Platform Reliability Expectations #### 4.5 Cultural, Regional, and Contextual Demand Factors ##### 4.5.1 Regional Compute Clusters and Demand Hotspots ##### 4.5.2 Data Sovereignty Norms Influencing Procurement ##### 4.5.3 Peer Influence and Industry Consortium Impact ##### 4.5.4 Cloud Adoption and Developer Readiness #### 4.6 Marketing, Awareness, and Channel Influence ##### 4.6.1 Impact of Developer Conferences and Industry Events ##### 4.6.2 Role of Technical Content and Cloud Marketplaces ##### 4.6.3 Systems Integrator Influence on Purchase ##### 4.6.4 Hyperscaler and Hardware Partnership Impact ### 5. Unmet Needs and Latent Demand Signals #### 5.1 Identified Gaps Between Current Supply and User Expectations #### 5.2 Latent Demand in Underpenetrated Enterprise Segments #### 5.3 Willingness to Adopt New Accelerator Architectures #### 5.4 Pain Points Surfaced Across Cohorts ### 6. Key Findings and Strategic Implications #### 6.1 Top Demand Drivers Ranked by Cohort #### 6.2 Barriers to Purchase and Adoption #### 6.3 High-Priority Customer Segments for Market Entry #### 6.4 Recommendations for Product, Pricing, and Channel Strategy ### Disclaimer ### Contact Us