The AI Bubble

Peter Lynch writes in the foreword to his “One Up on Wallstreet” in the year 2000 that internet companies were overvalued, right before the dotcom bubble crashed during 2000 and 2001. He highlights Amazon and a few companies that have reasonable earnings to their valuation but criticizes the market as a whole due to the outsized valuations of companies with no revenue.

This is an important to understand as some AI investments are unreasonable and we want to avoid buying or creating bubbles in AI assurance that erode trust in safety industrialization.

Why is AI not a bubble?

There are two reasons why one might think AI is in a bubble:

§1.1 OpenAI has a valuation of $157B on the private market despite a net loss of $5B with $3.7B in revenue. According to traditional metrics for public market investing, even for SaaS growth companies, this is an overvaluation. Even if we assume they can reduce their non-compute operating expenses to $0, their compute costs still total upwards of $7B, meaning that they would remain unprofitable by about $2.3B, indicating much less efficient profit margins than e.g. Google’s ad business. This is emblematic of similar companies, such as Anthropic and Perplexity.

§1.2 Generally intelligent generative machine intelligence, such as GPT-4o, is a digital commodity, meaning that anyone can spin up an open weights model, such as DeepSeek R1, for their own use and compete at the same level of tech as OpenAI.

But, I will argue that there are other strong reasons pointing in the opposite direction:

§2.1 There are a few companies that will become dominant on the other side of the race to AI, assuming society continues to function. This is similar to how a few of the dotcom companies emerged as the dominant forces on the internet during the 2000s. Whereas dotcom companies disrupted industries such as print advertisement, retail, and media and entertainment¹, AI and robotics has the potential to displace all human labor at a value of 51.4% of total world GDP (~$68T). This also implies that people would be willing to purchase ChatGPT for the $200+ / month if it could replace a personal assistant.

§2.2 The consequences of the §2.1 imply a series of philosophical and ideological goals that reach beyond mere economic value, leading some private equity capitalists to look beyond the mere economic gains. As Masayohshi Son said, “AGI is the only thing I care about.”

§2.3 The tech companies that have invested vast resources in AI, such as Meta, Amazon, and Google, currently have a price to earnings ratio of ~35 to ~50 which is much lower than the ratios of dotcom companies that reached PE ratios of 200.

What does this mean for AI safety?

Given my beliefs, it’s irresponsible to develop a generally intelligent species that may make humanity obsolete without waiting for a global democratic consensus on how we should go about it. However, in the context of investing, understanding how much will go to safety and assurance is important.

The most comprehensive work on this question is the AI Assurance Technology Market Report, that anticipates a 108% yearly growth to 2030 of the AI assurance (/safety) market, which leads to a total valuation of $276B. It projects that the AIAT market will be 15.2% of the total AI market from 0.8% in 2023. The 15% number is a conservative estimate that bases itself on the 5%-15% anticipated costs of complying with the EU AI Act. As a baseline, the cloud security market is about 6.3% the size of the total cloud infrastructure market.

Due to §2.1, I think 108% YoY growth is a reasonable target and that we may see even higher growth during the initial stages of general machine intelligence market adoption, 2026 through 2029.

For any investor, this means that AI safety and assurance is a massive future market that, if you get in now, allows you to earn a lot of money if society persists. If you care much about impact, it is also an amazing opportunity to accelerate the development of new technologies for existential security.

Conclusion

In summary, the AI market is probably not as overblown as some would have you believe, nor is it responsible to support AI’s further development. However, if you have money and wish to grow that money ethically, the best market seems to be AI safety and assurance.

The answer to AI’s $600B question is safety.