When Numbers Leave You Numb

Key Capital Private, Investment Note - Issue #32

When Numbers Leave You Numb 

In the second Austin Powers movie, there is a scene where Dr Evil makes a blackmail demand for $100bn, not fully appreciating the implications of the fact that he has travelled back in time to 1969. The response is that this amount is impossibly large - far more than exists in the world - similar to saying, “I want a kajillion bajillion dollars.”

We are committed numbers people who seek to derive meaning and insight from analysing data; however, the current market sometimes leaves us feeling that our grasp of fundamental arithmetic is no better than Dr. Evil’s miscalibrated ransom demands.

The contemporary financial landscape has reached a juncture where the sheer magnitude of numbers has begun to render traditional comprehension mechanisms inadequate. This shift in quantum is making it more difficult, if not impossible, to process and contextualise financial scale.

Musk’s Pay Demand

Tesla's proposed $1 trillion compensation package for Elon Musk over the next decade represents perhaps one of the most extreme examples of how financial numbers have transcended meaningful scale. The package is contingent upon achieving a company valuation of $8.5 trillion (c. 6 times its current value), selling one million autonomous taxis and AI robots, and significantly increasing Tesla’s profitability (source: Yahoo Finance). As an aside, on the 1st of October Musk achieved an intraday net worth of half-a-trillion dollars, which compares to a worth of ‘just’ $25bn in March 2020 (source: Forbes).

Oracle’s Share Price Jump

Oracle added c. $230bn in market cap in one day following a huge one-day increase of c. +36%, which was Oracle’s largest one-day percentage gain since 1992. The catalyst for this was an astonishing contract backlog jump from $130bn to $455bn - an increase of 250% quarter-over-quarter (source: CNBC).

Keith Weiss, an analyst at Morgan Stanley, said: “An extraordinary $332bn in bookings in Oracle’s Q1 represents not only the biggest bookings number we’ve ever seen in software, but a fundamental shift in the business model towards data centre operator.”

OpenAI’s reported $300bn, five-year commitment to Oracle raises serious questions about its capacity to honour such an obligation. OpenAI doesn’t have $300bn. In its last funding round in August, it raised $8.3bn, valuing the business at $300bn. This was part of a $40bn funding round of which SoftBank will be investing $30bn ($7.5bn already invested and $22.5bn due in before year-end). Its current annual revenue run rate of approximately $12bn still leaves it operating at a significant loss. Indeed, industry estimates suggest OpenAI could burn through over $44bn in cumulative losses before reaching profitability around 2029 (source: dataconomy).

Oracle is contracted to deliver approximately 4.5 gigawatts of data centre power capacity to OpenAI, which could power c. 3.4m homes or is equivalent to 4.5 large nuclear reactors. As highlighted in a Moodys report following the announcement, Oracle has a credit rating at the lower end of investment grade (Baa2 credit rating with a negative outlook), and this data centre build is effectively one of, if not the world’s largest, project financing. Moodys also highlighted the counterparty risk of relying on large commitments from a small number of AI companies. This is especially true when it is likely that cash flow will be negative for an extended period before reaching breakeven.

Could the sheer scale of the multi-year order be intended to hinder independent verification and bottom-up modelling?

Zuckerberg’s Recruitment Drive

The AI revolution has also spawned an investment boom in talent as companies seek to own the new post-AI future. This has resulted in the big tech firms offering packages worth hundreds of millions to AI experts.

Sam Altman has accused Mark Zuckerberg of personally calling AI developers with the promise of $100m sign-on bonuses. The NY Times reported that Zuckerberg offered Mark Deitke, a 24-year-old AI researcher, around $250m over four years, with potentially $100m of that to be paid in the first year (this was a revised offer after Deitke initially turned down an initial $125m approach). To best represent their interests, AI talent (similar to sports stars), are using agents to help them navigate the range of options and maximise the value for their skillset.

AI Capex Numbers

Four large tech firms (Alphabet, Amazon, Meta and Microsoft) are forecast to spend $380-400bn on AI Capex this year, which implies a 20-25% capex-to-revenue ratios (source: 8Figures). This compares to a projected $300bn at the beginning of the year (source: Irish Times). The implications of this level of spending are far-reaching. “AI machines - in quite a literal sense - appear to be saving the US economy right now,” Deutsche Bank stated recently. “In the absence of tech-related spending, the US would be close to, or in, recession this year.” Bain & Company recently reported that $2 trillion in annual revenue is needed to fund the computing power needed to meet anticipated AI demand by 2030.

Market Capitalisations

The normalisation of trillion-dollar valuations has created systematic distortions in market behaviour. Nvidia's achievement of a c. $4.5 trillion market capitalisation, means that it has added c. $3 trillion in market capitalisation since the end beginning of 2024. This is an extraordinary event that requires careful analysis. Instead, such market capitalisation movements are increasingly treated as routine occurrences.

Summary

Certain aspects of the current market have created a disconnect between numerical precision and cognitive comprehension, which makes deploying analytical skills more challenging. Austin Powers simply wanted to make an outlandishly large ransom demand, but his lack of understanding of the financial realities of the 1960s made him look foolish.

The Greek alphabet was used for numbers, creating a cumbersome system where arithmetic was difficult, limiting mathematics to geometry and theoretical proofs. So, suppose we revert to some Greek era analysis. In that case, we caution that stock market concentration is significant, capex spending is huge, AI revenues are MIA, and AI spending increasingly supports US GDP. One other aspect of the Greek mathematical system that hindered progress was the absence of a zero, which meant they couldn’t express concepts such as negative numbers. This is something that the tech/AI sector hasn’t had much use for of late.