소규모 언어 모델(SLMs) : 시장 점유율 분석, 업계 동향과 통계, 성장 예측(2026-2031년)

According to Mordor Intelligence, the small language models (SLMs) market size is projected to expand from USD 1.05 billion in 2025 and USD 1.37 billion in 2026 to USD 4.84 billion by 2031, registering a CAGR of 28.71% from 2026 to 2031.

Global Small Language Models (SLMs) Market Trends and Insights

Rising Demand For Low-Latency On-Device AI

Low-latency inference has moved from a technical preference to a commercial requirement across the small-language-models market. Enterprises now expect AI functions to deliver faster response times, lower power consumption, and fewer network dependencies when embedded in software and devices. Microsoft made Phi-4-mini-reasoning available on Snapdragon-powered Copilot+ PCs in May 2025, demonstrating that reasoning workloads could run with NPU offload rather than relying on CPU-heavy execution. Apple also released an on-device foundation model at WWDC 2025 that reduced KV cache memory use by 37.5% while maintaining multilingual performance across 15 languages, reinforcing the practicality of fully offline inference. Google extended that direction in April 2026 with Gemma 4 on Google Cloud, which supported more than 140 languages and included edge-oriented variants built for mobile and device deployment. As a result, the small language models market is being pulled forward by hardware and software categories that now treat local AI capability as a baseline feature rather than an optional enhancement.

Cost Pressure Versus Frontier Model Inference Economics

Inference cost has become one of the clearest adoption drivers in the small language models market because it changes how enterprises compare production architectures. Buyers are no longer choosing models only on capability benchmarks, because recurring API costs can quickly outweigh any performance advantage in high-volume workflows. The small language models market is therefore gaining traction not only because compact models are cheaper today, but also because their economics are pushing even frontier model providers to release smaller, more efficient alternatives.

Limited Context Depth Versus Larger Foundation Models

Context depth remains the clearest technical limit on how far the small language model market can go in complex analytical tasks. Many compact models perform well in bounded workflows, but long-document review, extended agent memory, and layered reasoning still expose the gap with larger systems. Apple stated that its on-device foundation model supports a 4,000-token context window, while its Private Cloud Compute pathway supports 32,000 tokens, highlighting the trade-off between privacy-first deployment and broader contextual capacity. Mistral addressed part of that gap in July 2026 with Magistral Small 2507, which supported a 128,000-token context window under an Apache 2.0 license while remaining deployable on-premises. Even with that progress, the small-language-models market still faces a design trade-off between edge efficiency and the broader contextual memory needed for the most demanding enterprise use cases.

Other drivers and restraints analyzed in the detailed report include:

1. Data Privacy, Residency, and Sovereign AI Requirements
2. Edge-Optimized Enterprise Automation Use Cases
3. Benchmark Fragmentation and Weak Cross-Model Comparability

For complete list of drivers and restraints, kindly check the Table Of Contents.

Segment Analysis

Frameworks captured 68.45% of revenue in 2025, placing them at the center of the small language models market during a period when enterprises were still standardizing their deployment approach. That position reflected the need for inference runtimes, orchestration layers, fine-tuning pipelines, and hardware-aware backends before a business application could move into production. Enterprises generally treated this layer as a long-term infrastructure choice because once a framework is embedded in internal workflows, it tends to support many subsequent model releases. The small language models market, therefore, showed stronger early spending on reusable model plumbing than on narrow point solutions. This pattern also suggested that buyers were building environments that could host successive proprietary and open-weight models rather than committing to one fixed algorithm.

Microsoft's ONNX-based Phi optimizations for Snapdragon-powered NPUs, available from May 2025, and Apple's Core AI framework introduced at WWDC26 both showed how platform vendors are trying to own the model-to-hardware path from compilation to runtime management. That strategy matters because open-weight model releases are lowering the scarcity value of access to basic frameworks across the small language model market. Mistral Small 3 in January 2025 and Mistral Small 4 in March 2026 were both released under the Apache 2.0 license, which increased pressure on vendors that rely solely on proprietary framework differentiation. Services are projected to grow at a 26.34% CAGR through 2031, and that shift points to a later monetization layer where enterprises want managed fine-tuning, domain adaptation, and monitoring after the base stack is in place. In that sense, the small language models industry is moving from framework selection toward operational support, where service depth becomes a more durable source of revenue than access to the framework itself.

Cloud held 57.38% of revenue in 2025, giving it the largest share of the small language models market at the start of enterprise adoption. Many organizations chose cloud-first because it reduced upfront hardware commitments, integrated with existing identity systems, and enabled usage-based testing before larger infrastructure decisions were made. That early dominance did not mean cloud was the long-term answer for every workload, because regulated sectors still needed stronger control over where data moved and how inference was handled. On-premises deployments continued to serve healthcare, BFSI, and other governance-heavy users that could not easily route sensitive tasks through third-party cloud environments. The small language models market has therefore kept a multi-path deployment structure, with each mode matching a different combination of cost, speed, and compliance needs.

Edge devices are projected to expand at a 28.98% CAGR through 2031, which makes them the fastest-growing deployment mode in the small language models market. That growth reflects hardware pull more than software supply, because enterprises are now seeing local inference as a practical production option rather than a lab exercise. Microsoft's Phi releases for Snapdragon-powered Copilot+ PCs and Apple's developer tooling around on-device inference both helped turn compact local execution into a more standardized path. Springer Nature reported in June 2026 that NPUs can deliver 40 to 60 times the energy efficiency of GPUs for edge inference within their operating range, thereby changing the economics of distributed AI deployment. Shenzhen's "AI + Advanced Manufacturing 2026-2027" action plan also backed pruning, quantization, and distillation for production-line deployment, indicating that public policy was reinforcing device-side demand in the small language models market.

Complete Report Scope:

• By Offering
  • Frameworks
  • Services
• By Deployment Mode
  • Cloud
  • On-Premises
  • Edge Devices
• By Application
  • Content Generation
  • Sentiment Analysis
  • Semantic Search and Information Retrieval
  • Conversational AI
  • Other Applications (Translation and Localization, Data Extraction and Document Analysis, etc.)
• By End-User
  • BFSI
  • Healthcare and Life Sciences
  • Retail and E-Commerce
  • Technology and Software Providers
  • Media and Entertainment
  • Telecommunications
  • Other End-Users
• By Geography
  • North America
    • United States
    • Canada
  • South America
    • Brazil
    • Rest of South America
  • Europe
    • Germany
    • United Kingdom
    • France
    • Italy
    • Spain
    • Nordics
    • Rest of Europe
  • Asia-Pacific
    • China
    • India
    • Japan
    • South Korea
    • Australia
    • Southeast Asia
    • Rest of Asia-Pacific
  • Middle East
    • Turkey
    • Saudi Arabia
    • United Arab Emirates
    • Rest of Middle East
  • Africa
    • South Africa
    • Rest of Africa

Geography Analysis

North America accounted for 37.89% of global revenue in 2025, making it the largest region in the small language models market. The region's lead rested on the concentration of early-adopting enterprises, mature cloud infrastructure, and the presence of major platform companies such as Microsoft, Google LLC, Amazon Web Services, and Apple. It also remained the clearest center for framework consolidation, because regulated enterprises tended to narrow procurement around a smaller set of production-validated platforms. New York's 2026 financial sector guidance on AI-related cybersecurity risk added to that pattern by reinforcing the need for auditable and locally governed deployments in compliance-sensitive environments. As a result, the small language models market in North America continued to face strong demand amid tighter enterprise scrutiny of deployment architecture.

Asia-Pacific is projected to expand at a 29.76% CAGR through 2031, making it the fastest-growing regional market for small language models. Growth is being driven by manufacturing-edge AI demand in Japan and South Korea, a highly active domestic model ecosystem in China, and cost-sensitive deployment demand across India. Mitsubishi Electric announced a manufacturing-domain language model for edge deployment in June 2025, and the company targeted product application in industrial equipment and robots within FY2026. Tokyo Electron Device launched its "Try it! SLM on Edge" support program in June 2026 to help manufacturers evaluate and productize compact models in edge environments. These developments showed that the small language models market in Asia-Pacific was moving from experimentation toward production integration, especially in industrial settings where local inference aligns with operational needs.

Europe remained the third-largest regional market, and its adoption pattern was defined by regulatory-first deployment choices in the small language models market. The EU AI Act transparency obligations under Article 50 were set to apply from August 2, 2026, which increased the importance of disclosure, labeling, and documentation in customer-facing AI systems. Germany and the United Kingdom led enterprise procurement in the region, while the Nordics were gaining relevance as design centers for sovereign AI architectures. South America was still in an early growth phase with banking, e-commerce, and media localization supporting initial adoption, while Middle East markets and parts of Africa were developing around sovereign AI infrastructure and local-language service demand.

1. Microsoft Corporation
2. Google LLC
3. Meta Platforms, Inc.
4. Nomic AI, Inc.
5. NVIDIA Corporation
6. Amazon Web Services, Inc.
7. IBM Corporation
8. Liquid AI, Inc.
9. Cohere Inc.
10. Mistral AI SAS
11. AI21 Labs Ltd.
12. 01.AI
13. Hugging Face, Inc.
14. Databricks, Inc.
15. Salesforce, Inc.
16. DeepSeek Artificial Intelligence Co., Ltd.
17. Qualcomm Incorporated
18. Apple Inc.
19. Ollama, Inc.
20. Alibaba Cloud (Alibaba Group Holding Limited)
21. Cerebras Systems, Inc.
22. Groq, Inc.
23. H2O.ai, Inc.
24. Upstage AI, Inc.

Additional Benefits:

• The market estimate (ME) sheet in Excel format
• 3 months of analyst support

TABLE OF CONTENTS

1 INTRODUCTION

• 1.1 Study Assumptions and Market Definition
• 1.2 Scope of the Study

2 RESEARCH METHODOLOGY

3 EXECUTIVE SUMMARY

4 MARKET LANDSCAPE

• 4.1 Market Overview
• 4.2 Market Drivers
  • 4.2.1 Rising Demand for Low-Latency On-Device AI
  • 4.2.2 Cost Pressure Versus Frontier Model Inference Economics
  • 4.2.3 Data Privacy, Residency, and Sovereign AI Requirements
  • 4.2.4 Edge-Optimized Enterprise Automation Use Cases
  • 4.2.5 Open-Source Model Ecosystems and Fast Fine-Tuning Cycles
  • 4.2.6 Hardware Efficiency Gains in NPUs, GPUs, and AI Accelerators
• 4.3 Market Restraints
  • 4.3.1 Limited Context Depth Versus Larger Foundation Models
  • 4.3.2 Benchmark Fragmentation and Weak Cross-Model Comparability
  • 4.3.3 Integration Complexity Across Legacy Enterprise Stacks
  • 4.3.4 Hardware Supply Chain Constraints and Accelerator Cost Sensitivity
• 4.4 Industry Value Chain Analysis
• 4.5 Regulatory Landscape
• 4.6 Technological Outlook
• 4.7 Porter's Five Forces Analysis
  • 4.7.1 Threat of New Entrants
  • 4.7.2 Bargaining Power of Suppliers
  • 4.7.3 Bargaining Power of Buyers
  • 4.7.4 Threat of Substitutes
  • 4.7.5 Competitive Rivalry
• 4.8 Impact of Macroeconomic Factors on the Market

5 MARKET SIZE AND GROWTH FORECASTS (VALUE)

• 5.1 By Offering
  • 5.1.1 Frameworks
  • 5.1.2 Services
• 5.2 By Deployment Mode
  • 5.2.1 Cloud
  • 5.2.2 On-Premises
  • 5.2.3 Edge Devices
• 5.3 By Application
  • 5.3.1 Content Generation
  • 5.3.2 Sentiment Analysis
  • 5.3.3 Semantic Search and Information Retrieval
  • 5.3.4 Conversational AI
  • 5.3.5 Other Applications (Translation and Localization, Data Extraction and Document Analysis, etc.)
• 5.4 By End-User
  • 5.4.1 BFSI
  • 5.4.2 Healthcare and Life Sciences
  • 5.4.3 Retail and E-Commerce
  • 5.4.4 Technology and Software Providers
  • 5.4.5 Media and Entertainment
  • 5.4.6 Telecommunications
  • 5.4.7 Other End-Users
• 5.5 By Geography
  • 5.5.1 North America
    • 5.5.1.1 United States
    • 5.5.1.2 Canada
  • 5.5.2 South America
    • 5.5.2.1 Brazil
    • 5.5.2.2 Rest of South America
  • 5.5.3 Europe
    • 5.5.3.1 Germany
    • 5.5.3.2 United Kingdom
    • 5.5.3.3 France
    • 5.5.3.4 Italy
    • 5.5.3.5 Spain
    • 5.5.3.6 Nordics
    • 5.5.3.7 Rest of Europe
  • 5.5.4 Asia-Pacific
    • 5.5.4.1 China
    • 5.5.4.2 India
    • 5.5.4.3 Japan
    • 5.5.4.4 South Korea
    • 5.5.4.5 Australia
    • 5.5.4.6 Southeast Asia
    • 5.5.4.7 Rest of Asia-Pacific
  • 5.5.5 Middle East
    • 5.5.5.1 Turkey
    • 5.5.5.2 Saudi Arabia
    • 5.5.5.3 United Arab Emirates
    • 5.5.5.4 Rest of Middle East
  • 5.5.6 Africa
    • 5.5.6.1 South Africa
    • 5.5.6.2 Rest of Africa

6 COMPETITIVE LANDSCAPE

• 6.1 Market Concentration
• 6.2 Strategic Moves
• 6.3 Market Share Analysis
• 6.4 Company Profiles (includes Global Level Overview, Market Level Overview, Core Segments, Financials as available, Strategic Information, Market Rank/Share, Products and Services, Recent Developments)
  • 6.4.1 Microsoft Corporation
  • 6.4.2 Google LLC
  • 6.4.3 Meta Platforms, Inc.
  • 6.4.4 Nomic AI, Inc.
  • 6.4.5 NVIDIA Corporation
  • 6.4.6 Amazon Web Services, Inc.
  • 6.4.7 IBM Corporation
  • 6.4.8 Liquid AI, Inc.
  • 6.4.9 Cohere Inc.
  • 6.4.10 Mistral AI SAS
  • 6.4.11 AI21 Labs Ltd.
  • 6.4.12 01.AI
  • 6.4.13 Hugging Face, Inc.
  • 6.4.14 Databricks, Inc.
  • 6.4.15 Salesforce, Inc.
  • 6.4.16 DeepSeek Artificial Intelligence Co., Ltd.
  • 6.4.17 Qualcomm Incorporated
  • 6.4.18 Apple Inc.
  • 6.4.19 Ollama, Inc.
  • 6.4.20 Alibaba Cloud (Alibaba Group Holding Limited)
  • 6.4.21 Cerebras Systems, Inc.
  • 6.4.22 Groq, Inc.
  • 6.4.23 H2O.ai, Inc.
  • 6.4.24 Upstage AI, Inc.

7 MARKET OPPORTUNITIES AND FUTURE OUTLOOK

• 7.1 White-Space and Unmet Need Assessment