SoftBank Group Corp. acquired Graphcore in July 2024, making the British AI-chip designer a wholly owned subsidiary while keeping its name and Bristol headquarters. Graphcore did not disclose the purchase price; contemporary reports put it at roughly $400 million to $500 million, a sharp drop from its reported $2.8 billion valuation in late 2020. The deal gave Graphcore a better-funded parent and gave SoftBank an AI-computing business with chip technology, software, and specialist engineers—but it did not prove Graphcore could replace Nvidia.
What happened to Graphcore?
Graphcore’s acquisition announcement is dated July 11, 2024. The buyer was SoftBank Group Corp., not SoftBank Corp., the Japanese telecommunications operator. Graphcore became a wholly owned subsidiary, retained its name and Bristol headquarters, and continued operating as a company. At the time of the announcement, co-founder Nigel Toon remained CEO. The company described the transaction as a platform for building the next generation of AI compute. (Graphcore’s acquisition announcement)
The price was not disclosed by Graphcore. Contemporary reporting offered estimates rather than a confirmed figure: one report put the deal at about $400 million and another at about $500 million. Those estimates should not be treated as the official consideration.
Why did Graphcore need a buyer?
Graphcore had a technically ambitious product and substantial backing, but it had not built the scale, software ecosystem, or customer adoption needed to compete sustainably with Nvidia. It designed specialized processors for AI and machine-learning workloads, raised hundreds of millions of dollars, and won customers including Microsoft. Yet building an accelerator business requires more than a promising chip: customers also need mature software, model and framework support, dependable supply, systems integration, financing, and confidence that a platform will be supported over time.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →#1 Best Overall
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
Graphcore’s prospects had weakened before the sale. Contemporary coverage reported a workforce reduction of about 20%, leaving roughly 500 employees, and a retrenchment from several markets, including Norway, Japan, and South Korea. The same coverage described approximately $700 million in investment, including money from Microsoft and Sequoia Capital. Nigel Toon later referred to more than $600 million in equity funding in a 2026 retrospective. These figures reflect different descriptions and should not be combined as if they were one audited total. (Contemporary acquisition and technical coverage; Toon’s 2026 statement)
What did Graphcore build?
The IPU and Poplar software
Graphcore’s Intelligence Processing Unit, or IPU, was designed for parallel AI computation. Its architecture used many independent processor cores, substantial on-chip SRAM, high internal memory bandwidth, and a graph-oriented execution model. Graphcore’s Poplar software stack was built to program and manage this hardware. The company’s approach was not simply to make a conventional GPU with different branding: it paired a specialized processor design with its own tools and programming model.
That specialization brought a trade-off. A workload that maps well to the IPU’s parallelism and local memory could benefit from its design, but the software must map the workload effectively. Nvidia’s competing platform had a broader developer base, mature tools and libraries, extensive framework support, and wide availability through systems and cloud providers. Those ecosystem advantages can matter as much as hardware characteristics when a customer is choosing what to deploy.
Rank #2
- High-Performance AI Processing: The MX3 is designed to handle the most demanding AI computer vision workloads, delivering exceptional performance and efficiency.
- Flexible Integration: The MX3 can be easily integrated into your existing systems via its M.2 M-key form factor and support for Linux operating systems.
- Energy Efficient: The MX3 is designed to provide high performance while minimizing power consumption.
- Comprehensive Software Development Kit (SDK): The MX3 is supported by a comprehensive SDK that simplifies development and deployment.
- Hardware compatability: The MX3 is compatible with the PCI-SIG M.2 M-key 2280 Specification. It can be used with the Raspberry Pi 5 with a M-key 2280 HAT.
Colossus MK2 figures are specifications, not a contest result
Contemporary coverage reported the following figures for Graphcore’s Colossus MK2 family and MK2 C600. They describe published or reported hardware specifications; they are not, by themselves, benchmarks of production AI performance.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
| Item | Reported specification |
|---|---|
| Colossus MK2 transistors | 59.4 billion |
| Colossus MK2 cores | 1,472 independent cores |
| Parallel threads | Up to 8,832 with simultaneous multithreading |
| On-chip SRAM | 900 MB |
| Aggregate on-chip memory bandwidth | 47.5 TB/s |
| Processor links | Ten IPU links for scaling between processors |
| MK2 C600 throughput | 560 TFLOPS FP8; 280 TFLOPS FP16; 70 TFLOPS FP32 |
| MK2 C600 power | Approximately 185 W |
Peak arithmetic throughput does not establish how quickly a real model trains or serves users. Precision, memory movement, sparsity, batch size, compiler maturity, interconnect, model support, and utilization can all change end-to-end performance. A meaningful comparison needs relevant workloads and complete systems, not just peak TFLOPS. Graphcore’s MK2 C600 product page provides a product reference.
Why was the reported sale price far below Graphcore’s earlier valuation?
Graphcore’s reported late-2020 valuation of approximately $2.8 billion came from a private-market funding period when investor appetite for AI-chip companies was high. A later sale, after Graphcore faced operating pressure and the challenge of competing with Nvidia’s established platform, reflected different conditions. A previous valuation is not a guaranteed sale price: it is an estimate made at a particular point, often tied to investment terms and expectations about future growth.
Rank #3
- ✅Powered by 26 Tera-Operations Per Second (TOPS) Hailo-8 AI Processor. 2.5W typical power consumption
- ✅Scalable, enabling simultaneous processing of multi-streams & multi-models
- ✅Enabling real-time, low latency and high-efficiency AI inferencing on the edge devices
- ✅Supports TensorFlow, TensorFlow Lite, ONNX, Keras, Pytorch frameworks
- ✅Supports Linux and Windows. Supports the temperature range of -40°C to 85°C
A strategic buyer can also value a company differently from financial investors. SoftBank may have valued Graphcore’s processor designs, software, engineers, and future development potential even if the independent business had not demonstrated a durable path to scale. But the exact transaction value remains unconfirmed: Graphcore did not disclose it, and the $400 million and $500 million figures were differing contemporary estimates. (Contemporary coverage)
What did SoftBank gain, and what might it do with Graphcore?
The publicly confirmed structure is straightforward: SoftBank Group acquired the company. The likely strategic value lies in what that company contained—Graphcore’s IPU architecture and related silicon designs, Poplar software, semiconductor engineering and verification talent, AI-systems experience, customer relationships, and knowledge from deployments. These assets could give SoftBank a foothold in AI compute beyond investments in other parts of the technology stack.
SoftBank’s announcement framed next-generation semiconductors and compute systems as relevant to its ambitions around artificial general intelligence (AGI). A relationship with Arm, which SoftBank controls, is a possible source of strategic coordination, but Graphcore did not announce a formal Arm integration or merger as part of the deal. Nor did the announcement establish a detailed product roadmap. Access to a larger parent may give Graphcore more time and resources to develop products; it does not guarantee a commercial breakthrough. (Graphcore’s announcement)
Rank #4
- 48GB AI graphics accelerator
Has Graphcore displaced Nvidia?
No evidence in the acquisition announcement or the company’s subsequent expansion statements shows that Graphcore has become a mainstream replacement for Nvidia’s GPUs. Graphcore entered the deal with a distinctive architecture, but competing in accelerators requires a reliable combination of hardware, software, customers, manufacturing and supply, developer adoption, and support. Nvidia’s ecosystem creates switching costs that a chip specification alone cannot overcome.
- Software and portability: Developers need workable tools, framework support, and a practical path to port models. A specialized architecture can be difficult to adopt if its software stack is unfamiliar or incomplete for a customer’s needs.
- Workload fit: On-chip SRAM and parallelism may suit some tasks, while memory capacity, model partitioning, or software mapping may limit others.
- Production evidence: Buyers need to know whether systems are available at useful scale, supported over time, and effective on their actual workloads.
- Economics: The relevant comparison is total cost and useful output—such as training time or inference throughput—not just theoretical arithmetic rates.
That is why Graphcore’s acquisition should be read as a second chance under strategic ownership, not as proof that its chips beat Nvidia or that the company had already solved its adoption challenge.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What has happened under SoftBank?
Graphcore’s own announcements through August 2026 describe expansion rather than a shutdown or absorption into another brand. In a July 31, 2026 statement, the company said it was approaching 1,000 employees, had opened development centers in Austin, Texas, and Bengaluru, India, and had expanded activity in Taiwan, Poland, Cambridge, and London. It also said it planned to move into a purpose-built Bristol headquarters in September 2026. “Approaching 1,000” is the company’s description, not a confirmed count of exactly 1,000. (Graphcore’s July 31, 2026 announcement)
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallBest Value
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
Toon stepped down as executive chair effective July 31, 2026, and the company said Marcus McElroy took the helm. On August 3, Graphcore announced a Taipei office and engineering lab, describing the move as continued investment in Taiwan and its semiconductor supply-chain relationships. These are meaningful signs that SoftBank has continued to fund and organize the business. They do not establish revenue growth, profitability, product volume, or market share. (Graphcore’s Taipei announcement)
What would show that the acquisition is working?
Hiring and new offices are inputs, not the final test. The stronger evidence will be commercial and technical outcomes that customers can verify:
- New chip generations, products, production availability, and dependable supply.
- Named customer deployments at meaningful scale, with evidence of sustained use.
- Support for current AI frameworks and models, alongside developer adoption of Poplar.
- Relevant independent or customer benchmarks on end-to-end workloads, rather than peak arithmetic specifications alone.
- Revenue, order volume, and evidence that the business can support ongoing development.
- Any concrete integration with Arm or other SoftBank companies, or data-center partnerships with measurable deployment plans.
- Clarity on whether Graphcore remains a merchant product business serving external customers or becomes primarily an internal technology group.
SoftBank bought time, capital, talent, and strategic optionality in AI hardware. Graphcore’s continued operation and expansion show that the buyer preserved and invested in the company. Whether that investment produces a commercially durable accelerator platform—and a credible alternative for any of Nvidia’s workloads—remains a separate question.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Free tools Windows power users keep installed
One-click scans. No signup required.
