AI Venture Capital's "First Wave of Downturn" is here: Where is the profit model? How to overcome the giants?

After six months of artificial intelligence prosperity triggered by ChatGPT, the market's enthusiasm for this concept is starting to wane, with some AI startups even beginning to lay off employees.

According to Business Insider, an AI and computer vision startup called Tractable laid off a large number of employees in May, citing a shift in market and venture capital focus from growth to profitability, stating that they "must be prepared for a challenging economic environment."

Jasper, a company developing AI writing tools, announced layoffs in July and reduced revenue expectations. Data shows that user growth for their tool has been continuously declining for four consecutive months as of July.

Even "star companies" in the generative AI field are not faring well. According to data from analytics platform Similarweb, ChatGPT experienced a 10% decline in monthly online visits in both June and July after months of growth. The monthly visits for Midjourney, an image generative AI platform, have been continuously decreasing for the past three months as of July.

Mark Goldberg, a partner at Index Ventures, stated that the emergence of commercial AI applications was once expected to achieve "instant success," but now there is a "superficial sense of disillusionment":

The initial surge in user growth for ChatGPT led to investors overestimating the speed at which consumers would adopt tools driven by generative AI.

Investors were eager to support early-stage companies building these products before they had customers or revenue, which raised concerns about the market potentially overheating.

However, despite some early-stage AI startups exiting the scene, overall, there is still an influx of hot money pouring into this race.

According to PitchBook data, investments in AI startups focusing on generating human-like text, images, and computer code have increased by 65% to reach $3.3 billion this year.

Unclear Business Models

Venture capitalists express that they are still uncertain about the winning business models of startups building new products around this technology—many startups have yet to prove their ability to retain users and develop products that existing companies cannot easily imitate.

John Luttig, an investor at Founders Fund, states that the investment potential of artificial intelligence was previously influenced by "extensive venture capital hype," to the point where sentiment about this field overshadowed all analysis of business models:

It was a clear optimism without any questions about the challenges of products, user interfaces, distribution, or end markets.

Due to the uncertainty of profit models, some companies have expressed doubts about investing in generative AI. Frank Slootman, CEO of data company Snowflake, stated during an earnings conference call in August:

We cannot venture into artificial intelligence without a business model.

Many executives describe their attempts with language models as experimental and exploratory, wondering how big of a "bread box" it is.

Sunil Dhaliwal, Managing Partner of Amplify Partners, an early-stage venture capital firm, says the focus on AI startups has shifted:

We have moved from the moment of "how big is the potential" to the moment of "how do we make it work."

Challenges for startups

Big companies are still heavily investing in AI, with Microsoft, Google, and others leading the way, and NVIDIA's business thriving.

For startups, developing large models requires access to and analysis of massive amounts of data, which can cost billions of dollars. Considering the uncertainty of profitability and competition from large companies, investors are starting to hesitate about continuing to invest in these small companies.

As the generative AI industry matures, barriers are gradually emerging, and data is one of them.

As stated in a recent report by technology investment firm Andreessen Horowitz, when AI products rely on proprietary data or data scale as key elements and differentiators, data provides a lasting competitive advantage:

For example, precious metals exploration company KoBold Metals has signed commercial agreements with major mining companies, granting exclusive access to the historical records of their various mining sites, thus creating a competitive moat.

As a defense startup, Anduril must secure appropriate federal partnerships to access sensitive data.

So far, most of the largest AI startups have received support from tech giants, and this assistance goes beyond money, including access to data and cloud server usage rights, among other invisible resources.

As regulations on training data for large models become more stringent, data accessibility becomes a challenge for startups without the support of big companies.

However, a16z also mentions that while the "data moat" often receives the most attention in defensive discussions for AI companies, the latest cycle of generative AI has introduced other potential defensive vectors.

For example, Character.AI has product network effects, where user interactions with the product become training data, improving the product experience. Midjourney focuses on developing the best-performing proprietary base models to create optimal application layer use cases.

Based on these cases, a16z believes that startups are more likely to capture market share from big companies at the application layer rather than at the infrastructure layer.

Existing companies can launch impactful generative AI products, even if these features are not actually new features, but are presented to users in new ways.

We believe that AI will completely redefine the workflow and user interface of software, as more and more AI software is not just a recording system, but a predictive and executive system.

We have seen this transformation happen very quickly, which means that flexible companies that can attract AI talent and rapidly deploy it have a competitive advantage.

Of course, regardless of where "defensive power" comes from, the ultimate winner in capturing market value will be the consumers.

Given the many interesting consumer applications that have already emerged, the scale of consumer surplus generated from AI is sure to spark an incredible wave of innovation.

If the history of software is any indication of innovation, it is that great entrepreneurs always find a way to build important and enduring companies in each new technological era.