DeepSeek's cut-price generative AI offering makes it a direct rival to OpenAI and its US peers, but it's far from alone. Here are some alternative startups from outside the US that have received corporate backing.

Non-US artificial intelligence startups

The emergence of China’s DeepSeek, an AI chatbot that reportedly cost less than $6m to develop, has sent shockwaves through an industry where the established leader, OpenAI, raised over $10bn in financing this past October alone.

Chinese startup DeepSeek released its large language model last week and it has already become the most downloaded free app on Apple’s US App Store.

The firm’s claims it developed the chatbot at a fraction of the cost of US counterparts suggests Silicon Valley AI startups are overvalued.

PitchBook data below shows that funding for North American artificial intelligence and machine learning startups dwarfs those in other regions, but perhaps the fundamental lesson from DeepSeek is that AI innovation doesn’t need to be led by Silicon Valley.

Foundational artificial intelligence is being worked on by a range of startups outside the US west coast. They may not have the same level of funding as their US counterparts, but they are pursuing approaches that place them as genuine alternatives.

Here are some of the AI startups outside of the US that have received backing from corporate investors.

Moonshot AI (China)

DeepSeek may be grabbing the headlines right now but there are several other Chinese startups building their own foundational AI models, most notably Moonshot AI, which is developing a large language model (LLM) designed to exceed the individual query character length limit in ChatGPT.

Moonshot’s platform is intended to work with lengthy inputs of text or data, giving similarly long results. It raised $1bn nearly a year ago, in a series B round led by ecommerce group Alibaba and backed by fellow corporates Meituan and Xiaohongshu. It doubled its valuation to $3bn six months later.

Not that China’s AI innovation stops there – Baichuan launched an open-source model yesterday that can understand text, images, audio and video to generate text and audio, on the back of nearly $700m from the likes of Alibaba, Tencent and Xiaomi.

Zhipu is focusing on the enterprise side and revealed earlier this month that it plans to compete on price with a model that costs less than 10% of current industry rates per prompt.

Mistral (France)

Europe’s most prominent LLM developer is probably France-based Mistral, the biggest name to emerge from a thriving Parisian AI scene. It’s certainly Europe’s best funded competitor in the AI race, having lured Nvidia, Samsung, Salesforce, Bertelsmann and ServiceNow for a $640m series B round last June valuing it at $6bn.

The startup is also consciously countering some US rivals through a deal that will integrate thousands of articles from press agency AFP in a bid to position itself more as a fact-based model compared with less reliable options. CEO Arthur Mensch said recently that he expects AI to expand from models to systems that combine LLMs and contextual business data this year.

Mensch was previously part of Google’s DeepMind AI programme, as were most of the founders of H Company, which emerged from stealth in May with $220m in seed capital from investors including Amazon, UiPath and Samsung.

Cohere (Canada)

While many of the best known generative AI developers are working on consumer-facing chatbot models, Canada-based Cohere has created an LLM called Command R+ designed for enterprise customers and which it claimed outperformed rivals from OpenAI and Anthropic (not to mention Mistral).

The company is among the most well-funded AI startups outside the US, having secured $500m from backers such as Salesforce, Nvidia and Saudi Aramco’s Prosperity7 Ventures fund in June. And earlier this month, it made a pitch for the entirety of the enterprise market with the launch of North, a workplace AI platform that’s intended to compete against not only other LLMs but Microsoft’s CoPilot and Google’s Vertex.

A Deloitte report last year suggested Canada was behind only the US, UK and Israel for the number of generative AI startups it hosts, but the question has always been whether they can compete while staying on their own patch – unlike, say, AI chipmaker TensTorrent, which raised nearly $700m last month after upping sticks to Silicon Valley. The emergence of DeepSeek could be a big point in its favour.

AI21 Labs (Israel)

Nowhere on earth, however, has more generative AI startups per capita than Israel, which began assembling a national expert forum on AI in September. While most of the new wave of AI model developers were formed in the wake of OpenAI’s GPT-4 model early last year, AI21 Labs has been building its own generative AI model since 2017.

Israel-based AI21’s model uses an alternative architecture that allows it take in longer sequences of data, meaning it can process that data more effectively in situations where more context is needed. Intel and Comcast helped it close a $208m series C round in late 2023 that valued it at over $1.4bn.

Israel’s startups also benefit from the highest concentration of AI talent in the world as well as a very strong local cybersecurity sector that can help secure their systems. Many of its startups focus on applying AI to specific areas like healthcare or farming, though the country is also home to Hailo, which is backed by ABB and NEC and which claims to build the world’s best edge AI processors.

Aleph Alpha (Germany)

Most of the new wave of AI model developers were formed in the wake of OpenAI’s GPT-4 model early last year, but Aleph Alpha has been building its own generative AI model since 2019, and it bagged $500m in a Bosch Ventures-backed round just over a year ago.

Aleph Alpha is pursuing a different path to many of its rivals, subsequently pivoting from its own LLM to a tool allowing users to customise third-party models so they can take in industry-specific data and languages that aren’t ordinarily supported. It may not be a DeepSeek-level game changer but that move potentially gives it access to a far wider market than any single LLM provider can claim.

That approach can also be seen elsewhere in Germany, where the other most highly funded AI startups – AI military tech developer Helsing, customer support chatbot creator Cognigy – are working on ways to use the technology rather than innovating their own models.

Moreh (South Korea)

The startup that perhaps comes closest to DeepSeek’s approach is South Korea’s Moreh, which has created a software tool that allows users to build and optimise their own AI models using a more flexible, modular approach. And it can run on a far wider range of GPU chips than the ones produced by market leader Nvidia.

In other words, using Moreh’s software means an organisation can seek more affordable chips to build out its AI infrastructure, but its applications will still be compatible with Nvidia’s toolkit as well as machine learning frameworks from the likes of Facebook owner Meta and Google.

The company, which counts AMD and local telecoms firm KT as investors, released an open-source version of its Korean language LLM last month. Compatriot Upstage AI has raised over $100m for its enterprise generative AI platform, and the country looks well placed to grow following the National Assembly’s announcement of an AI regulatory framework last month.

Robert Lavine

Robert Lavine is special features editor for Global Venturing.