NVIDIA und AMD haben ein Unicorn in der KI-Branche mit einem geschätzten Wert von 50 Milliarden Yuan ins Leben gerufen.

According to a report from Zhidongxi on August 15th, today, Canadian AI unicorn Cohere officially announced two major events in one go: new financing and personnel changes.

The company officially announced that it has obtained US$500 million (approximately RMB 3.6 billion) in new financing, and its valuation has increased from US$5.5 billion (approximately RMB 35.9 billion) to US$6.8 billion (approximately RMB 48.8 billion). The venture capital firms Radical Ventures and Inovia Capital led the investment, and AMD, NVIDIA, Salesforce, etc. participated in the investment.

Cohere has also recruited Joelle Pineau, the former vice president of Meta's fundamental AI research, to serve as the Chief AI Officer to advance Cohere's cutting-edge research and product development, and Francois Chadwick, a former Uber executive, to serve as the Chief Financial Officer, responsible for overseeing financial and business operations.

▲Joelle Pineau (left), the former vice president of Meta's fundamental AI research, and Francois Chadwick (right), a former Uber executive

Cohere was founded in 2019 by Aidan Gomez, a co-author of the Transformer paper. It mainly focuses on providing AI applications that can be privately deployed and solve practical problems in vertical scenarios for enterprises. Last month, according to foreign media The Information, citing two people familiar with the matter, Cohere has informed investors that it expects its annualized revenue to exceed US$200 million by the end of this year.

It can be seen that the heat of the financing and talent war among AI startups in the current AI field remains undiminished. On the one hand, AI startups are still favored by investors. On the other hand, the talent war sparked by Meta is intensifying, and technology giants and star startups are all involved.

01. Founded by a co-author of the Transformer paper, with NVIDIA and AMD as investors

Cohere was founded by three genius teenagers.

The three founders are alumni of the University of Toronto. CTO Ivan Zhang is a classmate of Gomez in the same department. He founded the AI research institution FOR.ai in 2017, which is now renamed Cohere For AI. Chief Scientist Nick Frosst graduated from the Department of Computer Science at the University of Toronto in 2015 and was a member of the team of Geoffrey Hinton, one of the "Three Giants of Deep Learning," during his tenure at Google Brain.

▲Cohere founders Ivan Zhang, Aidan Gomez, Nick Frosst (from left to right)

Since its establishment in 2019, Cohere has raised approximately US$1.1 billion (approximately RMB 7.9 billion) in total. The most recent financing was a US$500 million Series D financing in July 2024, when its valuation reached US$5.5 billion (approximately RMB 35.9 billion).

In previous financing rounds, AI star scientist Fei-Fei Li, Geoffrey Hinton, one of the "Three Giants of Deep Learning," Pieter Abbeel, the first doctoral student of Andrew Ng, NVIDIA, Oracle, Salesforce, and AMD are all among its investors.

Pineau once participated in the development of Meta's early open-source Llama model together with Yann LeCun, the vice president of Meta and the chief AI scientist, and left Meta in May this year. During the official announcement of the personnel change, she revealed in an interview with foreign media TechCrunch that she believes joining Cohere is a good opportunity beyond research. At Meta, the projects she was responsible for leading the research team on could take anywhere from 18 months to 10 years to complete. Now she will work within a more compact time frame and be involved in customer and product-related matters.

When talking about her next plan at Cohere, Pineau said that she hopes to focus on the AI Agent platform North, explore methods for developing AI Agents in a private and secure environment, create benchmarks to evaluate these systems, and explore the interaction of AI Agent networks in the real world.

02. Can be deployed with just two GPUs, low-cost and private deployment to seize the market

Different from OpenAI and Anthropic, which target general scenarios, Cohere mainly builds AI applications that can solve practical problems in vertical scenarios for enterprises and government departments. Its core products include the generative models Command A and Command A Vision, the retrieval models Embed 4 and Rerank 3.5, and the flagship AI Agent platform North. Enterprises can build applications based on these models.

Among them, the large language model Command A and the multimodal model Command A Vision only require two or even fewer GPUs to achieve local or private deployment. And the benchmark tests of Command A show that its performance in instruction following, SQL, programming, etc. is comparable to that of GPT-4o (Nov) and DeepSeek-V3.

North is its newly launched product. The platform was launched on August 6th and integrates Cohere's generative and search models with workflow automation tools. It can be used for private deployment by enterprises and provides many functions in enterprise workflows such as basic Q&A, document creation, chart production, AI Agent deployment, and contract review. It also only requires two GPUs to run.

It is worth mentioning that in May this year, Cohere also acquired the Vancouver AI Agent startup Cognosys and integrated the startup's automated market research platform Ottogrid into North.

▲Using North to create a chart

The company provides enterprises with three deployment models: SaaS, cloud platform, and private deployment. At the end of last year, the startup announced that it would shift its business focus to providing private deployment for enterprises and building smaller, domain-specific models. Currently, there is no unified pricing on Cohere's official website, and enterprises need to contact it one-on-one.

▲Cohere's three deployment models

On this basis, Cohere's core advantage is that in response to the demand for customized AI tools from enterprises in industries such as finance, healthcare, and government, it can gain market share by providing more private deployment support and defeating open-source models in these fields.

Based on this, the startup has already obtained well-known enterprise customers such as Oracle, Dell, and Fujitsu, and has reached an agreement with its investor Oracle to promote its products to Oracle's customers in the finance and healthcare fields.

In July, according to foreign media The Information, citing two people familiar with the matter, Cohere has informed investors that it expects its annualized revenue (calculated by multiplying the revenue of the most recent month by 12) to exceed US$200 million by the end of this year, which is nearly three times the annualized revenue of US$70 million the company expected in February. And it expects its annualized revenue to reach US$4 billion to US$5 billion by 2029.

In the early stage of the boom in generative AI, that is, in the middle of 2023, Cohere's popularity was comparable to that of Anthropic. However, now, compared with top startups such as OpenAI and Anthropic, Cohere's development in terms of models and customer scale is slightly lagging behind. Previously, OpenAI expected its actual revenue to reach approximately US$54 billion in 2027, and Anthropic expected its revenue to reach a maximum of US$35 billion in the same year.

This time, raising more funds and introducing an experienced person from Meta may help it achieve more research breakthroughs.

03. Conclusion: The AI track is burning money at a white-hot level

Technology giants are accelerating their efforts in the AI field. Meta, OpenAI, and Anthropic have invested billions of dollars in the AI field, which makes startups with limited financial strength and talent resources urgently need to explore new ideas to achieve more technological progress and implementation with fewer resources.

In this context, we are waiting to see how AI startups represented by Cohere can quickly transform these large investments into attractive products and maintain the company's competitive position.

Source: Cohere Blog

This article is from the WeChat official account “Zhidongxi” (ID: zhidxcom). Author: Cheng Qian. Republished by 36Kr with authorization.