Three quarters of the way through a tumultuous year on the policy front with tariff related volatility, concerns over both persistent inflation, and signs of wobbly growth, clients often ask, “Why do the markets remain resilient?” One answer is that despite considerable headlines and policy noise, profits have not been heavily disrupted since many of the most damaging policies were not actually enacted. But another important part of that has been the surging AI investment cycle.
Looking at prior examples (such as railroads, electricity, telephones, cars, television, mainframe computers, PCs, and the internet), innovation cycles go through stages. First, a new technology enters the economy and early deployments start to catch on rapidly. Early inventors and investors (often the same: founders) see a high return on their investment. This is followed by a second stage where additional capital is invested, often the re-invested profits of the first stage, and the technology grows rapidly in a self-funded format. In the third stage, outside investors see the second stage success, and aggressively build out the technology even more, in a manner where the early stage returns and strong margins are projected onto the surge in new capital. At this stage, outside capital, especially debt, is used to fund growth.
The third stage runs the risk of entering the fourth: becoming a bubble, both from a fundamental and valuation standpoint. A valuation bubble means high expectations from early investment successes are projected into high market prices on future investments. A fundamental bubble means more capital is invested too quickly and returns fall, the result of too much of the technology relative to demand. When subsequent returns prove less fruitful than expectations, valuations need to adjust. This can involve lower equity values and, more importantly, bankruptcy as debts cannot be repaid.
AI is certainly going through an innovation cycle. Up until now, major investors like Microsoft, Alphabet, Meta, and Amazon have funded a large portion of the AI buildout through internal cash flow. OpenAI, without a legacy business to rely on, has been funded with venture and partner capital.
But unlike software, web services and social networks, AI is capital intensive. It not only requires racks of servers running cutting edge semiconductors, but massive data centers with cooling and power infrastructure. This cycle seems to have shifted up a gear. The notable event was on Sept. 22, when Nvidia agreed to invest $100 billion in OpenAI, so OpenAI could turn around and buy more Nvidia chips. There was also Oracle announcing a multiyear surge in projected revenues from an anticipated deal with OpenAI, and then tapping the bond market for $15 billion debt in order to fund the expected capex surge needed to earn these hoped for revenues.
Chart 1
A new technology and surging share prices have generated comparisons to the dotcom bubble that peaked in 2000. One market comparison that has become increasingly common is to compare the performance of the tech stock focused NASDAQ performance today versus that 1990s period. One comparison to look at is the November 2022 introduction of OpenAI’s ChatGPT relative to the 1995 to 2000 period, when the introduction of Netscape’s web browser is recognized as the start of the internet era and the “dotcom” stock market. As evident in this comparison (see Chart 1, above), the blue “AI” line has only reached 1997 on the 1995 to 2000 timescale. At that point in the prior period, there were still two years left until the cycle peak. However, the recent NASDAQ rally got underway in March of 2020, amidst the early days of the pandemic. This period was also a key date for technology stocks, as the surge in work from home technology helped to fuel a surge in cloud-based technology adoption, serving as the start of the narrow technology market. If this start data is used, as shown by the red line, the timeline has reached 1999 in length, but not in magnitude.
These two scenarios can bracket the potential outcome. If this is just “1997,” while this does not mean the current market will triple from here (as was the case from 1997 to 1999), periods of technology adoption can play out over an extended period and it is possible the current market rally can continue. On the other hand, if its “1999,” with the current market rally going back to the March 2020 start, it has now lasted as long as that period’s peak, but its 300% cumulative return, while large, is well below the nearly 700% return of the original 1990s peak. There is potential for downside if the cycle turns, but perhaps not the magnitude of the drop from the 1999 peak.
With AI driving the stock market through the influence of the mega-cap Mag7 technology stocks, as well as the economy through surging investment, understanding where the market is in this cycle is important. The recent shift in cross funding, debt funding and surging data center capex from outside investors bears watching. Yet at the same time, demand is outstripping current capacity and many of these headline-grabbing investment totals are to be deployed over the next three to five years. In fact, looking at the 1995 to 2000 comparison, you can even see how, after the 2001 decline, the market was barely higher than 1997. A cautious investor may have in fact been right, but also may not have been patient enough to wait the four years to be proven correct. And even then, the end of 2001 was 40% higher than in 1997.
No two market cycles are ever identical. But historic examples from railroads in the 1890s to the internet in the 1990s show both the reason for optimism on innovation and skepticism (not pessimism) that all investments work out. In fact, it is often the case that new technologies truly reach their mass appeal only after an overinvestment bust leaves excess infrastructure at low cost. For the market and economy to continue to deliver, it must answer the question Is it 1997 or 1999? Much like Fed policy will have markets closely watching the economic data, the importance of AI will have the markets closely watching data points on this topic, as well.
