Graphcore

Corporate brand

Graphcore is a British semiconductor company developing massively parallel processors and software for artificial intelligence and machine learning.

Last updated October 14, 2026

AI crawler reads

No AI crawler has read this profile since Oct 28, 2026.

Overview

Graphcore Limited is a British semiconductor company focused on processors and software for artificial intelligence and machine-learning workloads. Founded in Bristol in 2016 by semiconductor entrepreneurs Simon Knowles and Nigel Toon, the company was established around the idea that neural-network computation required a processor architecture designed specifically for graph-based machine intelligence rather than an adaptation of conventional central-processing or graphics-processing hardware. Its principal technology is the Intelligence Processing Unit, or IPU. Graphcore's IPUs use a highly parallel architecture in which many compute tiles operate with local memory and communicate through an on-chip interconnect. This differs from the cache-centered design common in CPUs and GPUs. The architecture is intended to keep model data and intermediate results close to the processing units, reducing memory movement and exposing parallelism in training and inference workloads. Graphcore's software environment, the Poplar Software Stack, provides compilers, libraries, development tools, and integrations with machine-learning frameworks including TensorFlow and PyTorch. The company announced the Poplar stack in 2016 and introduced its first processor, the Colossus GC2, in 2017. The GC2 became available in 2018 and was also offered through the C2 IPU PCI Express card, which combined two processors in one accelerator board. Graphcore subsequently developed the second-generation GC200 processor, manufactured using TSMC's 7-nanometre FinFET process. Announced in 2020, the GC200 contained 59 billion transistors, 1,472 computational cores, and approximately 900 megabytes of on-chip memory. Its design supported massively parallel, multiple-instruction, multiple-data execution and distributed local memory. Graphcore expanded its proposition beyond individual accelerator cards into multi-IPU systems, cloud access, and data-centre infrastructure. In 2019, Microsoft Azure made Graphcore C2 IPUs available for preview, giving developers and researchers access to Graphcore hardware through a major public-cloud platform. Graphcore also worked with system manufacturers and technology partners to bring IPU-based systems to enterprise and research users. In 2022, Graphcore and TSMC introduced the Bow IPU, a second-generation implementation using three-dimensional packaging. The design bonded a GC200 compute die to a dedicated power-delivery die, allowing higher operating frequencies at lower core voltage. Graphcore connected the Bow generation with its longer-term ambition to build very large AI systems, including a concept named the Good Machine after statistician I. J. Good. The company positioned such systems as a route to training models with parameter counts beyond the scale of the human brain's estimated synaptic connections. Graphcore attracted substantial venture and strategic investment during the AI-accelerator boom. Its backers included Robert Bosch Venture Capital, Atomico, Sequoia Capital, Microsoft, Samsung, Dell Technologies, Amadeus Capital Partners, C4 Ventures, Draper Esprit, Foundation Capital, and Pitango. A December 2018 financing valued the company at approximately $1.7 billion, making it one of the United Kingdom's prominent semiconductor unicorns. Despite this support, Graphcore operated in an intensely competitive market dominated by established GPU suppliers and large integrated technology companies. In early 2023, Meta Platforms acquired Graphcore's AI networking technology team. In July 2024, SoftBank Group agreed to acquire Graphcore for approximately $500 million. The transaction was subject to review under the United Kingdom's national-security investment-screening process. Following the acquisition, Graphcore became part of SoftBank's broader technology portfolio. Its continuing strategic identity is centered on specialized AI compute, IPU architecture, compiler and systems software, and the potential use of its tec…

History

Graphcore was founded in Bristol in 2016 by Simon Knowles and Nigel Toon. Both founders had prior experience in semiconductor entrepreneurship, and they created Graphcore to pursue a processor architecture tailored to machine intelligence. The company's central proposition was that neural-network workloads could benefit from a large array of simple, closely connected processing elements with local memory, rather than relying primarily on conventional CPU or GPU memory hierarchies. During its first year, Graphcore developed and announced the Poplar Software Stack, a graph-oriented toolchain intended to compile and manage machine-learning workloads on its processors. In autumn 2016, the company raised its first funding round, led by Robert Bosch Venture Capital. Samsung, Amadeus Capital Partners, C4 Ventures, Draper Esprit, Foundation Capital, and Pitango were among the other investors associated with the early financing. In July 2017, Graphcore announced its first silicon product, the Colossus GC2. Built using a 16-nanometre process, the processor was described as a massively parallel, mixed-precision floating-point device for machine-learning applications. The GC2 became available in 2018. Graphcore packaged two GC2 processors on the C2 IPU PCI Express card, enabling the card to function as an accelerator alongside conventional host systems and machine-learning frameworks. The company's fundraising accelerated as interest in dedicated AI hardware increased. A Series B round in July 2017 was led by Atomico, followed later that year by a $50 million investment from Sequoia Capital. In December 2018, Graphcore completed a $200 million Series D round at a reported valuation of approximately $1.7 billion. Microsoft, Samsung, and Dell Technologies participated alongside financial investors. The valuation made Graphcore a prominent British semiconductor unicorn. Graphcore's go-to-market strategy included partnerships with cloud and server companies. On 13 November 2019, it announced that C2 IPUs were available for preview on Microsoft Azure. This arrangement allowed users to experiment with Graphcore hardware without purchasing and operating their own accelerator servers, while also giving Graphcore a route into cloud-based AI development. In July 2020, Graphcore presented the GC200, its second-generation Colossus processor. Manufactured by TSMC using a 7-nanometre FinFET process, the GC200 contained 59 billion transistors, 1,472 computational cores, and around 900 megabytes of local memory. Its architecture organized processing and memory into tiles and islands, with message passing used to coordinate computation across the device and between devices. The approach supported multiple-instruction, multiple-data parallelism and was designed to keep data movement predictable and local. Graphcore and TSMC presented the Bow IPU in 2022. Bow used three-dimensional packaging to bond a GC200 compute die face-to-face with a separate power-delivery die. The arrangement was intended to raise clock speed while reducing the voltage required by the compute die. Graphcore presented Bow as part of a broader roadmap toward extremely large AI machines, including the proposed Good Machine, named for mathematician and statistician I. J. Good. The company operated in a difficult competitive environment. Its IPU architecture and software stack differentiated it from GPU suppliers, but customers also expected mature frameworks, extensive software support, cloud availability, and favorable total cost of ownership. Graphcore therefore continued to invest in compiler technology, framework integrations, systems, and networking in addition to chip design. In early 2023, Meta Platforms acquired Graphcore's AI networking technology team. In July 2024, SoftBank Group agreed to acquire Graphcore for approximately $500 million. The acquisition attracted regulatory attention because Graphcore was a significant British semiconductor and AI-computing company; the deal was reviewed under the United Kingdom's investment-security process. The transaction placed Graphcore within SoftBank's technology portfolio and represented a major ownership change after years of venture-backed development. Graphcore's established product identity remains associated with IPUs, the Colossus and Bow processor families, Poplar software, and specialized infrastructure for AI workloads.

2024
SoftBank agrees to acquire Graphcore
[1]Graphcoreen.wikipedia.org
SoftBank Group agrees to acquire Graphcore in a transaction reported at approximately $500 million, subject to United Kingdom investment-security review.
2023
Meta acquires an AI networking technology team
[1]Graphcoreen.wikipedia.org
Meta Platforms acquires Graphcore's AI networking technology team.
2022
Bow IPU is presented
[1]Graphcoreen.wikipedia.org
Graphcore and TSMC introduce a 3D-packaged IPU using a compute die bonded to a power-delivery die.
2020
GC200 is presented
[1]Graphcoreen.wikipedia.org
Graphcore announces its second-generation 7-nanometre Colossus processor with 1,472 cores and approximately 900 megabytes of local memory.
2019
C2 IPU enters Microsoft Azure preview
[1]Graphcoreen.wikipedia.org
Graphcore accelerator hardware becomes accessible through a preview offering on Microsoft Azure.
2018
C2 IPU becomes available
[1]Graphcoreen.wikipedia.org
Graphcore makes the C2 accelerator card, containing two GC2 processors, available for machine-learning systems.
2018
Graphcore reaches unicorn valuation
[1]Graphcoreen.wikipedia.org
A $200 million Series D financing values Graphcore at approximately $1.7 billion.
2017
Colossus GC2 is announced
[1]Graphcoreen.wikipedia.org
The company reveals its first IPU processor, a 16-nanometre massively parallel mixed-precision device.
2016
Graphcore is founded
[1]Graphcoreen.wikipedia.org
Simon Knowles and Nigel Toon establish Graphcore in Bristol to develop specialized processors and software for machine intelligence.
2016
Poplar Software Stack is announced
[1]Graphcoreen.wikipedia.org
Graphcore introduces a graph-oriented software toolchain for compiling and running machine-learning workloads on its processors.

Products and positioning

A specialist AI-chip and systems company positioned as an alternative to GPU-centric machine-learning infrastructure, emphasizing massively parallel computation, distributed local memory, compiler technology, and tightly integrated hardware-software systems.

AI accelerator software · 2016
Poplar Software Stack
Graphcore's compiler and software environment for programming IPU systems. Poplar translates machine-learning and other graph-based workloads into operations suitable for Graphcore's tiled, distributed-memory architecture. It includes development tools and integrations intended to support established machine-learning frameworks, including TensorFlow and PyTorch.
AI processor · 2017
Colossus GC2
Graphcore's first announced IPU, the GC2 is a 16-nanometre massively parallel mixed-precision processor. It uses many processing tiles with local scratchpad memory and message passing rather than depending on a conventional cache hierarchy. The chip became available in 2018 and formed the basis of Graphcore's early accelerator products.
PCI Express AI accelerator card · 2018
Graphcore C2 IPU
The C2 IPU card packages two Colossus GC2 processors on a single PCI Express board. It is designed to work with standard host systems and machine-learning frameworks, and it provided an initial hardware platform for Graphcore's IPU proposition. C2 access was later offered through a Microsoft Azure preview.
AI processor · 2020
Colossus GC200
The GC200 is Graphcore's second-generation Colossus processor, manufactured using TSMC's 7-nanometre FinFET process. It contains 59 billion transistors, 1,472 computational cores, and about 900 megabytes of local memory. Its architecture is optimized for highly parallel execution, distributed memory, and communication among processing tiles.
AI processor · 2022
Bow IPU
Bow is a later IPU generation based on the GC200 compute design and advanced three-dimensional packaging. Graphcore and TSMC bonded a compute die face-to-face with a dedicated power-delivery die. The packaging approach is intended to support higher clock speeds while lowering compute-die operating voltage.

Brand decisions

2024
Agree to SoftBank acquisitionM&A
Background
Graphcore operated in a highly competitive AI-accelerator market and had developed a substantial technology portfolio but remained privately held.
Action taken
Graphcore agreed to be acquired by SoftBank Group in a transaction reported at approximately $500 million.
Aftermath
The transaction was subject to review under the United Kingdom's investment-security regime and placed Graphcore within SoftBank's technology group.
reported acquisition value. approximately $500 million (2024 transaction announcement)
Source
2022
Adopt 3D packaging for the Bow IPUGeneration change
Background
Graphcore and TSMC pursued higher performance and improved power characteristics for the company's second-generation IPU technology.
Action taken
They introduced a package that bonds a GC200 compute die to a separate power-delivery die.
Aftermath
The Bow design extended Graphcore's processor roadmap and supported its strategy of scaling specialized AI systems.
Source
2019
Bring C2 IPUs to Microsoft AzureProduct launch
Background
Graphcore sought to make its specialized accelerator technology accessible to developers without requiring them to buy dedicated hardware.
Action taken
The company announced preview availability of C2 IPUs through Microsoft Azure.
Aftermath
The arrangement gave Graphcore a public-cloud distribution channel and allowed users to evaluate IPU-based workloads remotely.
Source

Recent events

2024
SoftBank agrees to acquire Graphcore[1]Graphcoreen.wikipedia.org
M&A
SoftBank Group agreed to acquire Graphcore for approximately $500 million. The transaction was subject to review by the United Kingdom's investment-security authorities.
2023
Meta acquires Graphcore's AI networking technology team[1]Graphcoreen.wikipedia.org
M&A
Meta Platforms acquired Graphcore's AI networking technology team, while Graphcore continued as an independent AI-chip company at that time.
2020
Graphcore presents the GC200 second-generation processor[1]Graphcoreen.wikipedia.org
Product generation
Graphcore announced the GC200, a 7-nanometre processor with 1,472 computational cores and approximately 900 megabytes of local memory.
2019
Graphcore C2 IPUs become available for preview on Microsoft Azure[1]Graphcoreen.wikipedia.org
Product launch
Microsoft Azure began offering preview access to Graphcore C2 IPU technology, extending the company's hardware to a major public-cloud environment.
2018
Graphcore closes a $200 million financing round[1]Graphcoreen.wikipedia.org
Other
Graphcore completed a Series D financing that valued the company at approximately $1.7 billion and included investment from Microsoft, Samsung, and Dell Technologies.
2017
Graphcore announces its first Colossus processor[1]Graphcoreen.wikipedia.org
Product launch
The company introduced the Colossus GC2, a massively parallel processor designed for machine-intelligence workloads.
2016
Graphcore raises its first venture funding round[1]Graphcoreen.wikipedia.org
Other
Graphcore secured an initial financing round led by Robert Bosch Venture Capital, with participation from several technology and venture investors.

Sources10 sources

Cite this profile: Cite the canonical profile. /brand-wiki/graphcore · Editorial policy · How profiles are compiled