The AI boom is threatened by an epidemic of unreliable numbers

The AI boom is currently threatened by an epidemic of unreliable numbers, according to a recent article in The Financial Times. OpenAI's revenue is reportedly nearing $50bn, but this figure is lower than the $70bn claimed by Axios just a week earlier. Tech stocks experienced a significant drop in response to these conflicting reports.

Investors expressed concern over the lack of clarity in the reported figures, while Hock Tan, the CEO of Broadcom, compared the markets to a casino, further emphasizing the uncertainty surrounding the numbers.

The issue of unreliable numbers has plagued the tech industry for years. The 1980s merger wave left corporate balance sheets cluttered with goodwill, making it difficult to measure intangible assets such as brands and patents in the 1990s and 2000s. The introduction of stock-based compensation in the 2010s further complicated the picture of corporate health, particularly for tech companies.

This pattern of acronyms, adjusted numbers, and valuation techniques has become commonplace in every investment boom, sometimes with the intention of clarifying the situation and other times to obscure it.

One of the main problems today is that the reported revenues of OpenAI and Anthropic are not actual revenues but rather vague interpretations of a popular Silicon Valley metric called ARR. ARR stands for annual recurring revenue, which is calculated by multiplying the sales from the past month by 12. However, Anthropic includes sales of its models through cloud providers like Amazon, while OpenAI only does so after subtracting its partners' cut.

This ambiguity leaves investors in a precarious position, as they are unable to accurately compare the firms' performance.

Investors are left in a state of uncertainty, relying on wacky valuations to gauge the worth of these AI labs. Coatue, an investment firm, estimates that the labs should be worth a multiple of their pre-training EBITDA, which excludes the cost of training their models. This approach is reminiscent of the oil industry's EBITDAX metric, private equity's creative accounting practices, and even the controversial "community-adjusted EBITDA" used by WeWork, a now-defunct office-leasing company.

Public markets provide some clarity, but not without its own set of issues. Last year, the accounting question at hand was whether big tech firms were writing down the value of their computing assets too slowly, overstating their earnings. This issue was merely replaced by two larger problems: the level of leverage held by the giants and the measurement of their return on capital.

In April, Microsoft reported an "annual revenue run rate" of $37bn for its AI business, while Amazon claimed $25bn in ARRR. Despite the extra "r" in Amazon's figure, these financial metrics remain opaque.

The most crucial aspect of the AI trade is understanding where the revenue is coming from. According to researchers at Goldman Sachs, a return between 10% and 20% on investments in data centers by cloud giants like Amazon, Google, Meta, and Microsoft, combined with a 30% margin for AI developers, would require an annual revenue of $1trn.

This means that the value of productivity gains companies extract from AI technology must far exceed this amount. However, measuring the impact of AI on the economy has become increasingly challenging. Soaring prices for memory chips have led to incorrect inflation calculations for some software products, potentially overstating GDP growth in America by 0.3 percentage points.

Additionally, funding cuts to data-collection bodies and the decline in competence of these organizations have further complicated the measurement of economic growth.

In summary, the AI boom is currently facing a significant threat due to the epidemic of unreliable numbers plaguing the industry. The conflicting revenue reports from OpenAI and Anthropic, along with the opaque valuation techniques employed by investors and companies, have left investors and analysts in a state of uncertainty. This lack of clarity has far-reaching consequences, impacting not only the AI trade but also the overall economy, as soaring prices for memory chips have led to incorrect inflation calculations and underreported GDP growth.