AI Boom or Bust

Today, the Wall Street Journal published an analysis of Projected Capital Expenditures vs Projected Returns in the Data Center industry. That journalists will write these articles is not surprising in the midst of an unprecedented boom in AI Capital Expenses across the board. Chips, networking, power and several other aspects of the AI Supply Chain are in a massive boom that is not slowing. In charts, the article says that current data center is 70.8 GW and is estimated to be at 109.2 GW by end of 2027. For context, the United Kingdom produces roughly 35 GW per hour. In terms of square footage, data centers will take up about 4 times of all Costco warehouse space by 2027. In terms of dollars, by 2030, another analysis quoted in the article projects a total of $5 trillion in investments. Then, the article goes on to say that to provide an annual 10% return on capital, the industry will have to create and additional $650 billion in revenues (Apple is $400 billion roughly).

This is a good analysis considering the fundamentals from an outside analyst's perspective. I compared this article and several others in recent days to the testimony that Jensen Huang and Satya Nadella offered on a couple of podcasts. These executives have the glass half full. First of all, from a corporate standpoint, they have already made their monies and there is no reason to hold back now. Microsoft has already made 10 times what it put into Open AI. Additionally, they have the fastest growth rate among the big 3 hyperscalers on a higher base (not adjusting to the fact that Microsoft has a larger portfolio than AWS). Building and renting data centers is a core business for hyperscalers. They are all investing in vertical integration too. They are all working on their own chips, they are working on their own AI models and everything else that is or will be needed in the world in the future. Why should they not have a glass half full outlook and will an investor reward them for holding back?

There is a need for caution and there is abundant caution in the airwaves right now when journalists and analysts write about the coming reckoning in valuations. Compare what is happening today with what happened in the years leading up to 2000. Back then, there was a long and broad market rally. the world was experiencing a long peaceful era, the 80s and 90s had delivered a financial engineering boom, supply chains were going global, the knowledge economy was starting to take hold and costs of increasing productivity were on the decline. The US government had turned a surplus at the end of 1998. Everything was looking way up. In the midst of that, the dot com boom led to a frenzy of investment in several business models. Fiber Optics was booming, Telecom, Satellite Comms, eCommerce - all of those were booming. There was an accompanied strength in labor market unprecedented in several decades. That industrial activity produced a lot of fiber capacity that went unused for a decade or more. There is a term for that - "dark fiber". We simply did not have enough demand for the capacity that was produced in 2000. If one only focuses on the capacity over production, yes, we are in similar times now. Run before the sky falls! However, what is starkly different now is that there is massive demand for AI from those who do not produce AI or are not in the AI industry. For all the talks of circular revenues, the demand is not circular. Lets build a narrative for this demand. First, this is not a new demand that is emanating from thin air. All industries (bar none) have built their operations on Information Technology and Consumer Technology products over the last 25 years. They all can benefit from AI. IT itself is reliant on writing code and making sense of data that AI can excel at. AI is simply the next leap forward in these types of operational technologies. All aspects of human life will be impacted positively (or negatively) using AI. That is a huge demand. Second, companies are building robots that will autonomously perform physical functions. Count in all autonomous (driverless) cars, trucks, port equipment, warehouse equipment, earth moving equipment, boats and even aircrafts. Sooner or later these will become ubiquitous and will need massive amounts of AI data center capacity to run. Third, we are in the early innings of AI model technology. In the next few years, the fortune 100 companies and sovereign nations will have a strong need to build independent AI models instead of relying on the top 3 or 4 companies to take a prompt and turn it into tokens (token factory). Meaning, there will be many more models deployed all over the world that create tokens in a token factory. Where will these companies and nations host these models? Fourth, structurally too, the way data centers are built today, they are concentrated in large warehouses in rural areas. The expectation is that all devices on the edge (such as your phone or your car) will always have a reliable network connection to support fast decision making (for example when your car needs to decide suddenly whether to swerve to the left or to the right). It is plausible in future to build more distributed, purpose build, much smaller compute stations that are built into every street intersection or in air balloons or in buoys so that all types of robots and edge devices can have a low latency access to an aggregated compute (fast access to a computer that is connected to many other edge devices). In other words, Compute, Memory and Network itself will become ubiquitous as fiber optic cable and radio waves (its everywhere).

Of course, there is no telling how to future will unfold and all of the above is simply a narrative to contrast against hard numbers that analysts are putting forth. Going back to the glass half empty argument, I do believe that a lot depends on how fast all of these new technologies are adopted. There will be a lot of pain for investors in the coming years when several of these technologies fail to pan out or are far more mediocre than the creators will have us believe. On my part, I am prepared to lose some in the short term and gain it all back over the long term.