The Cake is Almost Eaten by OpenAI, AI Startups Face Financing Difficulties

[baoshi.rao]

According to overseas media reports, the investment train for AI startups is slowing down. Some young startups are finding it increasingly difficult to achieve their new round of financing goals, whereas a year ago, investors would have been eager to jump in.

Take the startup Liquid AI as an example. The company is developing a new type of AI model that can continuously learn during use, rather than just undergoing pre-training. This approach is fundamentally different from the popular OpenAI model's working method.

In July of this year, Ramin Hasani, the CEO of Liquid AI, negotiated with investors to raise up to $100 million for the startup he co-founded a few months earlier with three other researchers from MIT.

However, some investors were already cautious about Liquid AI's unproven technology at the time. Earlier this month, the company only managed to raise $37.6 million from the family office of Bain Capital senior advisor Stephen Pagliuca and OSS Capital.

"Nine months ago, a simple demo on Twitter could secure funding," said Joanne Chen, General Partner at Foundation Capital. "Now, founders need to be more thoughtful, and their companies must have long-term value propositions." Foundation Capital has supported AI companies including model monitoring startup Arize and chip developer Cerebras.

Although investors have become more cautious about AI companies, AI deals remain a highlight in the current financing market. PitchBook data shows that while overall U.S. startup investments in Q3 decreased by 21% year-over-year, investments in AI startups surged by 150%. More established companies like Anthropic and OpenAI received investments worth over 60 times their annualized revenue. However, it's important to note that the AI financing market is vastly different from the first half of this year when deals could be completed within weeks.

Since startups like Stability AI and Jasper faced difficulties after raising large sums, investors have grown more cautious about writing big checks for emerging generative AI companies. Tech giants like OpenAI and Microsoft, along with open-source AI providers, have become major threats. Many startups also face the harsh reality of enterprise sales, where clients quickly drop suppliers at the slightest sign of trouble.

Some ongoing investment talks have fallen through. London-based large language model developer Conjecture began discussions with investors in August to raise around $100 million. However, after months of talks, the startup decided to pause fundraising as investors grew increasingly concerned about competition from rivals like Meta Platforms and Mistral, as well as the high costs of model development. Many AI startups are under pressure, as they rely on foundational models from companies like OpenAI—which are now planning to sell directly to the startups' target customers. This would allow those customers to develop their own tools in-house.

Jenny Lefcourt, a partner at venture capital firm Freestyle Capital, said: "One issue is the growing belief that most of the value in AI will go to companies that already have distribution channels, making it harder for AI startups." Lefcourt noted that startups securing funding are those able to demonstrate more mature products.

AI 融资市场的另一个晴雨表是Ideogram，这家公司的产品可以根据文字描述生成图像。Ideogram 于2022年由四名前谷歌人工智能研究人员创立，今年早些时候以超过1亿美元的估值进行了融资， 当时他们甚至不清楚自己想要推出什么产品。而到今年10月份，Andreessen Horowitz 等顶级公司提出以高达5亿美元的投前估值为这家初创公司提供资金，尽管它没有收入。

而一些风险投资者对此表示反对。新企业联合公司（New Enterprise Associates）也曾为 Rewind 和搜索引擎 Perplexity 等人工智能初创公司提供过支持，该公司担心 Ideogram 的人工智能模型运行成本过高，也担心它将如何与 Midjourney 和 OpenAI 的 Dall-E 等竞争对手区分开来。除了估值过高之外，这些顾虑也导致该公司放弃了本轮融资。目前还不清楚是否有其他公司参与，而 Ideogram 也并没有透露，公司当前的估值究竟是多少。

生成式人工智能热潮中的一些早期赢家所遭遇的麻烦也导致了新创始人的冷淡现状。Jasper公司销售基于OpenAI 的GPT模型的写作工具订阅服务，该公司去年获得了15亿美元的估值。但最近，该公司将其内部估值下调了20%，并将其2023年年度经常性收入预测下调了至少30%。

Voice-to-text startup Deepgram, which raised $86 million from venture capital firms Madrona, Tiger Global Management, and Y Combinator, laid off about 20% of its employees in October this year, marking its second round of layoffs this year. Both companies are facing market pressure from competitors like OpenAI. Before its recent CEO crisis, OpenAI's business had been thriving.

Of course, not all companies are struggling to secure funding. Some smaller startups continue to attract investor interest with high valuations. Adaptive, an Amsterdam-based startup focused on helping large language model developers improve model quality, is in talks to raise $20 million at a post-money valuation of $100 million. Index Ventures and Iconiq Capital are negotiating funding with this months-old company, though Iconiq was not previously involved in this round.

Another issue many investors face is determining whether companies purchasing AI products from startups are merely experimenting or genuinely committed to long-term development in the field. A recent report by app development startup Retool revealed that about half of the surveyed companies said they purchased AI applications either for temporary purposes or because their needs were still in the early stages. Nearly half of the respondents also stated they did not customize models created by companies like OpenAI, indicating weak market demand for these startup products.

Even OpenAI saw a sharp decline in ChatGPT website traffic this summer, and although Microsoft's enterprise AI sales appear to be taking off, usage rates have not returned to peak levels.

"First come the innovators, then the imitators, and then the idiots—that's the order in which these things typically develop," said Venky Ganesan, a partner at Menlo Ventures. "We're still in the innovator phase of the cycle, but we're slowly starting to see imitators emerge. Of course, we haven't seen any idiots yet."