Processing-in-Memory (PIM) Market By Technology, Memory Type, Application, End User - Market Size, Industry Dynamics, Opportunity Analysis and Forecast For 2026-2035

The processing-in-memory (PIM) market is entering a period of rapid expansion as the semiconductor industry increasingly focuses on reducing the performance and energy limitations created by conventional processor-memory architectures. The market is estimated to be valued at approximately USD 300.4 million in 2025 and is projected to reach around USD 8,030.2 million by 2035. Based on these estimates, the market would expand at a compound annual growth rate (CAGR) of 38.9% between 2026 and 2035. This projected growth reflects increasing interest in architectures that can bring computation closer to stored data and reduce the amount of information that must move between memory and processing units.

The underlying demand for PIM is closely associated with the rapid growth of artificial intelligence, machine learning, high-performance computing, data analytics, and other data-intensive applications. Conventional computing architectures typically require processors to retrieve data from memory before performing calculations. As datasets and AI models become larger, this constant movement of information can create bottlenecks in memory bandwidth, latency, and energy consumption. PIM attempts to address these limitations by enabling selected computational operations within or near the memory subsystem, potentially improving the efficiency of workloads that depend heavily on memory access.

Noteworthy Market Developments

The processing-in-memory (PIM) market is developing around memory manufacturers, semiconductor foundries, and specialised PIM technology companies. The competitive landscape is closely connected to the broader growth of artificial intelligence, high-performance computing, data analytics, and high-bandwidth memory. Samsung Electronics has been one of the most prominent companies in the development of processing-in-memory technologies, particularly through its work on HBM-PIM and other memory-centric computing architectures.

SK hynix is another major participant in the PIM ecosystem, particularly because of its strong position in the high-bandwidth memory market. HBM has become a critical component of modern AI accelerators because advanced processors require extremely high memory bandwidth to prevent computational resources from being constrained by data availability. Micron Technology represents another major force in advanced memory and AI infrastructure. The company has extensive capabilities in DRAM and high-bandwidth memory and has increasingly focused on memory technologies designed to support the growing requirements of AI accelerators.

Taiwan Semiconductor Manufacturing Company (TSMC) occupies a different but equally important position in the PIM ecosystem because it operates primarily as a semiconductor foundry rather than as a conventional memory manufacturer. UPMEM represents a different category of competitor because it is focused specifically on commercial processing-in-memory technology rather than operating as a large-scale conventional memory manufacturer or semiconductor foundry.

Core Growth Drivers

Solving the memory wall and reducing the energy associated with data movement are emerging as major factors supporting the growth of the processing-in-memory (PIM) market. Traditional von Neumann computing architectures separate processing units from memory, requiring data to move repeatedly between memory arrays and CPUs, GPUs, or other accelerators before calculations can be completed. As modern applications process increasingly large datasets, this separation creates a growing performance and energy constraint. Processors may be capable of executing enormous numbers of operations, but their effective performance can be limited by how quickly data can be supplied from memory.

Emerging Opportunity Trends

The commercialization of DRAM-based processing-in-memory (PIM) technologies, particularly High Bandwidth Memory PIM (HBM-PIM) and AXDIMM, represents an emerging opportunity for growth in the PIM market. As artificial intelligence, high-performance computing, and data-intensive applications continue to expand, conventional processor-memory architectures are increasingly challenged by the volume of data that must be transferred between computing units and memory. DRAM-based PIM addresses this challenge by introducing computational capabilities within or closer to the memory subsystem, allowing selected operations to be performed near the location where data is stored. This approach can reduce unnecessary data movement and potentially improve overall system performance and energy efficiency.

Barriers to Optimization

The lack of standardised programming models and a mature software ecosystem represents a significant challenge to the growth of the processing-in-memory (PIM) market. While PIM architectures can potentially reduce data movement, improve memory bandwidth utilisation, and increase energy efficiency, these hardware advantages can be difficult to realise without software capable of identifying which workloads should be executed within memory. Traditional computing systems rely on well-established programming models, compilers, libraries, operating systems, and development tools. PIM introduces an additional layer of architectural complexity because developers must determine how computations are divided between conventional processors, accelerators, and processing resources located within or near memory.

Detailed Market Segmentation

By technology, DRAM-based processing-in-memory (PIM) architectures are positioned as an important segment of the PIM market in 2026, particularly as AI workloads place increasing pressure on conventional processor-memory architectures. Technologies such as HBM-PIM and AXDIMM are designed to bring selected computational operations closer to where data is stored, addressing one of the fundamental limitations of the traditional von Neumann computing model. In conventional systems, processors and memory operate as separate components, requiring data to move repeatedly between them during computation. As AI workloads become increasingly data-intensive, this movement can create bandwidth constraints, latency, and additional energy consumption.

By memory type, High Bandwidth Memory (HBM) is positioned as a central technology in the rapidly expanding AI memory market, supported by the growing deployment of GPUs, AI accelerators, and other high-performance computing platforms. The increasing size and complexity of artificial intelligence models are placing greater demands on memory bandwidth, capacity, and energy efficiency. Traditional memory architectures can struggle to supply data to high-performance processors quickly enough, creating a bottleneck between computational capability and memory access. HBM addresses this challenge by providing exceptionally high data-transfer bandwidth within a relatively compact package, making it particularly suitable for AI accelerators used in data centers and high-performance computing environments.

By application, AI inference is emerging as a leading segment of the processing-in-memory (PIM) market in 2026, reflecting the rapid transition of artificial intelligence from experimental model development to large-scale commercial deployment. While AI training remains highly resource-intensive and continues to require substantial computing capacity, inference workloads are generated continuously once models are deployed in real-world applications. Every interaction with an AI-powered search engine, virtual assistant, recommendation system, enterprise application, fraud-detection platform, or generative AI service can generate an inference workload. As organisations increasingly integrate AI into customer-facing and operational systems, the cumulative volume of inference activity is creating a significant requirement for efficient, high-throughput computing architectures.

By end user, data center operators are expected to represent the largest share of demand in the market, reflecting the rapidly increasing computational and infrastructure requirements associated with artificial intelligence, cloud computing, high-performance computing, and large-scale data processing. Hyperscale data center operators in particular are under growing pressure to accommodate increasingly intensive workloads while maintaining operational efficiency. As AI models become larger and inference workloads become more demanding, data centers must provide not only greater processing capacity but also sufficient memory bandwidth, power availability, cooling capacity, and high-speed interconnects. These requirements are making memory architecture an increasingly important consideration in overall data center design.

• Segment Breakdown

By Technology

• DRAM-Based PIM (HBM-PIM, AXDIMM)
• SRAM-Based Compute-in-Memory
• Non-Volatile/ReRAM-Based
• Near-Memory Compute

By Memory Type

• HBM
• DDR/LPDDR
• GDDR
• Emerging NVM

By Application

• AI Inference
• Recommendation Systems
• Genomics & Analytics
• Edge AI

By End User

• Data Center Operators
• AI Chip Vendors
• Edge Device OEMs
• Research

By Region

• North America
• The U.S.
• Canada
• Mexico
• Europe
• Western Europe
• The UK
• Germany
• France
• Italy
• Spain
• Rest of Western Europe
• Eastern Europe
• Poland
• Russia
• Rest of Eastern Europe
• Asia Pacific
• China
• India
• Japan
• Australia & New Zealand
• South Korea
• ASEAN
• Rest of Asia Pacific
• Middle East & Africa (MEA)
• Saudi Arabia
• South Africa
• UAE
• Rest of MEA
• South America
• Argentina
• Brazil
• Rest of South America
• Geography Analysis
• North America is expected to maintain a dominant position in the global market in 2026, supported by rapid infrastructure expansion and substantial investments by leading cloud service providers. The region's advantage is closely linked to the scale of its technology ecosystem, particularly the presence of major hyperscalers, semiconductor companies, AI developers, and data-centre operators.
• The United States represents the principal contributor to North America's market position and is estimated to account for more than 85% of regional revenue. This dominance reflects the country's concentration of major technology companies, semiconductor manufacturers, cloud platforms, and AI research organisations. Leading US-based silicon companies and hyperscale cloud providers are increasingly collaborating on specialised hardware architectures designed to address the computational and memory requirements of advanced AI applications.
• Canada provides an additional source of momentum for North America's AI ecosystem through public-sector support for artificial intelligence research and commercialisation. Government-backed funding and initiatives aimed at strengthening AI research, technology development, and commercial deployment have helped reinforce Canada's position within the regional innovation landscape. Investments directed toward AI research commercialisation in technology centres such as Toronto can support the development of applications and infrastructure spanning cloud computing, edge computing, and specialised AI hardware.

Leading Market Participants

• Samsung Electronics
• SK Hynix
• Micron Technology
• UPMEM
• Mythic
• Syntiant
• EnCharge AI
• d-Matrix
• Axelera AI
• Untether AI
• TSMC
• Rain AI
• Neuchips
• Alibaba (DAMO)
• Imec
• Other Prominent Players