Graphcore IPU Enterprise AI Computing and MaaS Platform

📌 Background

In 2026, the commercialization of large AI models entered the implementation phase, causing a surge in demand for computing power. As a British AI chip rising star, Graphcore leverages its IPU architecture specifically designed for artificial intelligence, holding unique advantages in sparse computing and large-scale parallel inference. Following its acquisition by SoftBank in 2025, it secured an additional $450 million in funding in 2026 and was integrated into SoftBank's Stargate project deployment pipeline, entering the enterprise AI computing market through chip sales and cloud inference services.

👤 Target Customers

Targeting enterprise-level AI application developers, cloud service providers, research institutions, and various enterprise clients requiring high-performance AI training and inference computing power.

💰 Revenue Streams

1) Primary revenue comes from the sales of IPU chips and system products (such as IPU-M2000, IPU-POD), followed by subscription fees for inference services provided via the Graphcore Cloud, charging enterprises for computing resource usage after attracting developers for free via the developer cloud; 2) Hardware maintenance: Annual subscription fees for maintenance, inspection, firmware iteration, and online support of IPU equipment; 3) Dedicated deployment: Single-contract deployment, commissioning, and integration fees charged to clients requiring local computing cluster deployment and setup or integration with existing training platforms.

🧮 Cost Structure

Chip design and manufacturing tape-out costs, data center infrastructure construction and operation/maintenance, R&D personnel salaries, market expansion, and ecosystem community operation expenses.

🛡️ Moat

Technical barriers formed by the IPU's unique architecture patents and the Poplar software stack, combined with large-scale computing order binding brought by SoftBank's Stargate project, as well as years of accumulated developer community and early cooperation cases.

🔑 Keys to Success

• Continuously iterate the IPU architecture and optimize the Poplar software stack
• Deeply bind with SoftBank's Stargate project to rapidly implement large-scale deployment
• Expand the developer community to attract more model adaptations

⚠️ Risks

• Incomplete ecosystem leads to low customer loyalty and high churn rates
• Uncertainties in chip mass production yields and capacity ramp-up
• Over-reliance on SoftBank investment and orders, lacking independent self-sustainability capabilities

🏢 Cases

• IPU applications within SoftBank's Stargate project deployment pipeline
• Microsoft participated in IPU development as an early partner of Graphcore
• BMW previously used Graphcore technology as an early customer

📊 SWOT Analysis