AI is nothing new. For decades it has been in use in data-driven and analytical workflows, with increasing sophistication. But creativity and ideation were considered human skills, far from the capabilities of artificial intelligence.
The emergence of Generative AI (GenAI) and programs such as StableDiffusion and ChatGPT has turned this assumption on its head. GenAI is an emerging frontier of AI, which uses Large Language Models (LLMs) trained on large data sets of content media (text, images, audio, video) to create new text, audio, images and more.
During a rapid emergence, Generative AI startups have attracted huge funding from investors, with over $26B in funding in the last five years.
The US is the clear leader in Generative AI VC funding, even when looking beyond OpenAI.
GenAI immediately sent shockwaves in adjacent markets. NVIDIA shares ramped more than 100% in H1 2023 (NVIDIA is the leader in AI chips), while companies such as Chegg (education tutoring) lost over 50% due to their business model being disrupted by GenAI.
Median round sizes
Generative AI startups are showing a significant premium in their Seed and Series A rounds.
In particular, the median Series A round size is 2x that of other startups. This holds also going back in time to 2018.
VC by stage
Early-stage investment into Generative AI startups has steadily increased since 2016, with acceleration from 2020 onward. We are now seeing a slowdown in Q2 and Q3 2023.
However, figures will likely rise slightly due to some reporting lag on smaller rounds.
Breakout-stage GenAI funding grew 5x from 2020 to 2022 but slightly slowed down in the last few quarters.
Late-stage funding spiked in 2023 with over $14B in funding, led by OpenAI's $10B funding, but also by several other megarounds.
Model makers have raised over 70% of GenAI funding, followed by applications and infrastructure.
See our mapping of Generative AI startups on Dealroom.
Model maker OpenAI leads in terms of funding raised by GenAI companies, but Anthropic, Adept AI, Inflection AI, Aleph Alpha and a handful of other players have also raised significant sums. In general, considerable funding is required to sustain the high training and deployment costs of LLMs' general models.
Verticalized model makers are starting to emerge, such as Hippocratic.ai, which came out of stealth with a $50M seed round for its health-focused LLM. Industries poised for specifically developed LLMs include health, fintech, and legal tech.
Applications are the second most funded segment of Generative AI after model makers. Use cases range across all media types (text, image, video, speech/audio/music, code, and 3d assets).
Most applications have been built around text, such as copywriting, customer relations assistants/chatbots and knowledge & search. Other notable segments include code generation, image generation, speech generation and game design.
Applications are split between those built on proprietary models and applications built on third-party models.
Most of the applications are built on third-party models, such as Jasper and Typeface. However, several startups are building applications based on their proprietary GenAI models. Examples include Character.ai, Runaway and Descript.
Building on proprietary GenAI models can provide a hedge against competition as applications will likely take advantage of gathered data and user interaction to fine-tune proprietary models. Others may build layers of model fine-tuning on top of third-party models.
The huge increase in GenAI usage across multiple use cases, both in consumer and enterprise adoption, has created the need for dedicated infrastructure from prompt engineering to MLops (training, deployment, optimization and monitoring) to data and embedding.
Some of these solutions are being added as add-ons on previous MLops offerings, such as in the case of Scale AI.
Others respond to new GenAI needs, such as Vector databases, which already raised an all-time-high $191M in 2023, led by Pinecone and Weaviate.
Many top-tier investors have been building up their Generative AI portfolios. Andreessen Horowitz and Sequoia have made nearly 50% more Generative AI investments than anyone else so far.
Ycombinator is by a large margin the most active accelerator for GenAI startups, with over 100 startups supported, including OpenAI, Jasper and Replit.
Top countries and cities
The top country globally for Generative AI funding is the US, with a large lead, followed by Israel and Canada. The UK and Germany follow.
The Bay Area has been the main hub for Generative AI, attracting over $20B in less than 4 years.
Even without OpenAI's $12.3B funding, the Bay Area still attracted almost 10x the funding of the next hub - New York.Toronto (Cohere), Tel Aviv (AI21) and London (Stability.ai) follow as the three leading global hubs outside the US.
AI chips: the pillars of GenAI
The GenAI wave is increasing demand for AI chips and processors for training and deploying LLMs at scale. This has sent NVIDIA shares ramping up more than 100% in the first half of 2023 (NVIDIA is the leader in AI chips). However, even Nvidia is two to three months behind on new order fulfilment for cloud server chips. Training costs and computing power availability are becoming a constraint for startups and companies wanting to train and deploy LLMs.
Globally, AI chips funding started ramping up in 2017-2018 and peaked in 2021.
However, 2022 was the most active year ever by number of rounds, showing some new early-stage innovation coming up.
China has been by far the leading geography for AI Chips investments.
However, despite the clear unmet and growing demand for AI chips and the limitations in semiconductor computation capacity, AI chip startups have, in some cases, not yet lived up to their promise.
Once heralded to great excitement, UK AI chip startup Graphcore had their valuation written down to zero by Sequoia Capital in April 2023, after having lost a major deal with Microsoft in late 2022. Graphcore has raised $682M in total VC funding and was valued at $2.8B in Dec 2020.