AI Industry Booms, But Startups Find Financing Harder

[baoshi.rao]

On December 13, news emerged that venture capital returns for AI startups may be slowing down. While these companies once easily secured funding, it's becoming increasingly difficult now.

Take Liquid AI as an example. They're developing a novel AI model that can learn not just through training but also during use - a different approach from OpenAI's models. According to reports, Liquid AI's CEO Ramin Hasani discussed raising up to $100 million with investors in July this year.

However, some investors remain cautious about Liquid AI's unproven technology. Earlier this month, the company announced it had raised only $37.6 million from Bain Capital senior advisor Stephen Pagliuca and OSS Capital's family office, far below its $100 million target.

Joanne Chen, General Partner at Foundation Capital, noted: "Nine months ago, a simple demo or 'packaging' on X could secure funding. Now founders need more thoughtful approaches and long-term value propositions." Foundation Capital has invested in AI companies including model monitoring startup Arize and chip developer Cerebras.

Despite the overall gloomy funding market, AI remains a bright spot. PitchBook data shows that while Q3 investments in US startups dropped 21% year-over-year, AI startup funding grew by 150%. Established companies like Anthropic and OpenAI secured funding at over 60 times their annualized revenue. However, the AI funding landscape has changed significantly from early 2023 when deals closed within weeks.

Since companies like Stability AI and Jasper struggled after raising large sums, investors have grown more cautious about writing big checks for emerging generative AI startups. Meanwhile, major players like OpenAI and Microsoft, along with open-source AI providers, have become significant threats. Many startups also face challenges in enterprise sales, with customers quick to abandon vendors at the first sign of trouble.

Some investment negotiations often fall through. According to two informed sources, Conjecture, a large language model developer based in London, began discussions with investors in August this year, hoping to raise approximately $100 million. However, after several months of negotiations, the startup decided to pause its fundraising efforts as investors grew increasingly concerned about competition from rivals like Meta and Mistral, as well as the high costs of model development.

Newly established AI startups face even greater pressure, with many purchasing technology such as large language models from OpenAI and other major companies. These larger firms are also attempting to sell their technology to the same target customers as the startups. These customers might include software developers in need of AI coding assistants or chief information officers seeking better technology for searching enterprise data or summarizing meeting notes.

Jenny Lefcourt, a general partner at venture capital firm Freestyle Capital, noted that one issue is the growing belief that "most AI value will flow to existing enterprises that already have distribution channels, making it harder for AI startups." Lefcourt believes that AI startups capable of securing funding are those that can demonstrate more fully developed products.

One standout in the generative AI funding space is Ideogram, which generates images from text descriptions. Founded in 2022 by four former Google AI researchers, Ideogram secured funding earlier this year at a valuation exceeding $100 million, even though it was unclear what product it would launch. By October, top venture capital firms like Andreessen Horowitz had proposed funding the startup at a high valuation of $500 million, despite the company having no revenue.

However, some venture capitalists remain hesitant about investing in Ideogram. Two informed sources revealed that New Enterprise Associates was concerned about the high operational costs of Ideogram's AI model and how it would differentiate itself from competitors like Midjourney and OpenAI's Dall-E. New Enterprise Associates, which has also invested in AI startups like note-taking software Rewind and search engine Perplexity, ultimately walked away from the funding round due to these concerns and the perceived overvaluation. It remains unclear whether other firms participated or what their valuations were. Ideogram did not respond to requests for comment.

The issues troubling some early winners in the AI boom are now creating a harsher environment for new founders. For example, Jasper, a subscription service provider that develops writing tools based on OpenAI's GPT model, received a pre-investment valuation of $1.5 billion last year. Today, the company has internally lowered its valuation by 20% and reduced its 2023 annual recurring revenue expectations by at least 30%.

Speech-to-text startup Deepgram has secured $86 million in funding from venture capital firms Madrona, Tiger Global Management, and Y Combinator. However, the company laid off approximately 20% of its workforce in October, marking its second round of layoffs this year. Both companies face increasingly fierce competition from larger rivals like OpenAI.

Despite this, some smaller startups continue to attract investor interest with high valuations. Amsterdam-based startup Adaptive aims to help large language model developers improve model quality. According to two sources familiar with the matter, the company is in talks to raise $20 million at a post-money valuation of $100 million. Sources indicate that Index Ventures and Iconiq Capital are negotiating funding for this months-old company.

One challenge investors face is determining whether companies purchasing AI products from startups are merely experimenting with the technology or planning long-term investments. In a recent report by application development startup Retool, about half of the surveyed companies stated they were just beginning to adopt AI applications or using them for temporary purposes. Nearly half also reported that their companies had not customized models created by OpenAI, suggesting lukewarm demand for tools that fine-tune these models from VC-backed startups.

Even OpenAI witnessed a sharp decline in ChatGPT website traffic this summer, and its usage rates have yet to return to peak levels, despite Microsoft's enterprise AI sales seemingly taking off.

Venky Ganesan, a partner at Menlo Ventures, said, "In the hype cycle, there are innovators, then imitators, and then idiots—that's the order of things. We're still in the innovation phase, but we're slowly starting to see imitators, just not idiots yet."