From record highs to questions about the bill

Wall Street began the week with plenty to celebrate. Artificial intelligence continued to underpin expectations for another strong earnings season, and US equities were setting fresh records. Yet, as the week progressed, a more uncomfortable question began to creep into the conversation. Just who is going to pay for the enormous infrastructure being built around AI, and at what cost?

The change in mood was reflected in the market. On Tuesday, the SPX500 and NAS100 closed at record highs, helped by steadier oil prices and some relief in Treasury yields. By Thursday, however, the NAS100 had lost 1.22%, with semiconductor stocks among the hardest hit. Investors were beginning to look beyond the excitement surrounding AI and examine the financing arrangements, revenue assumptions and eventual returns underpinning the investment boom.

One development helped bring those questions into sharper focus. According to Reuters, OpenAI told investors that its annualised revenue in September was approaching $50 billion, rather than the roughly $70 billion previously indicated. The distinction matters. It did not mean that $20 billion in revenue had suddenly disappeared. Much of the difference reflected how revenue was defined, particularly sales made through cloud partners. Even so, the episode was a reminder of how heavily investors are relying on financial information from private companies, where reporting conventions are not always consistent or easy to compare.

Then came the financing headlines. Reuters reported that SpaceX was discussing a $40 billion package to purchase Nvidia chips, involving approximately $30 billion in investment-grade debt and $10 billion in loans. Separately, the Wall Street Journal reported that Broadcom was arranging around $50 billion in financing for OpenAI-related hardware, while Oracle was also exploring funding options.

None of these reported arrangements should be mistaken for completed transactions. Nevertheless, they offer some indication of the extraordinary sums involved. Building the infrastructure required to support AI is becoming an increasingly significant financing exercise, not simply a technology investment story.

The bond market is setting tougher terms

Unfortunately for companies contemplating such commitments, the cost of money is hardly moving in their favour.

US Treasury yields remain elevated as investors contend with stubborn inflation, heavy government borrowing and an economy that has proved remarkably resilient. On Thursday, the 10-year yield approached 5.34% before easing towards 5.23% after a well-received 30-year Treasury auction. The auction offered some reassurance that demand for US government debt remains intact. It did little, however, to change the broader reality that borrowing over longer periods remains expensive.

Oil has added to the difficulty. UKOil settled at $103.91 a barrel on Thursday as concerns resurfaced over Middle Eastern shipping and disruptions to production in the US Gulf. Prices subsequently eased towards $103 in early trading today following indications that US-Iran negotiations were continuing.

For financial markets, these developments are about considerably more than the price of a barrel of oil. Persistently high energy costs squeeze corporate margins, complicate the inflation outlook and make it harder for central banks to contemplate easier monetary policy.

That concern was evident in the minutes of the Federal Reserve's September meeting, released on Wednesday. Policymakers pointed to higher energy prices and the AI infrastructure boom itself as sources of inflationary pressure. Most viewed another interest-rate increase before year-end as likely, although the eventual decision would depend on incoming economic data.

There is an uncomfortable irony here. The very investment boom helping to drive economic activity may also be contributing to the inflation pressures that keep its financing costs elevated. For companies embarking on multiyear expansion programmes, this is hardly an ideal combination.

Nor is the issue confined to the United States. European government bond markets have also been unsettled by inflation and fiscal concerns, particularly in France. With sovereign borrowers competing for capital, companies seeking to raise substantial amounts of long-term debt face an increasingly demanding financing environment.

The next phase demands proof of returns

None of this necessarily undermines the investment case for artificial intelligence. Demand continues to grow rapidly, and semiconductor manufacturers are already seeing tangible benefits. According to LSEG estimates, analysts expect third-quarter earnings for S&P 500 companies to rise approximately 30.6% from a year earlier.

That remains a forecast, of course, rather than an actual earnings result. Nevertheless, it goes some way towards explaining why equities have held up as well as they have in the face of higher borrowing costs.

The difficulty is that strong demand for computing equipment is not necessarily the same thing as an attractive return on the capital used to purchase it.

For a semiconductor supplier, a large order may translate into revenue relatively quickly. For the company building and operating the data centre, the financial consequences are very different. There are years of financing costs to absorb, equipment to depreciate and electricity bills to pay. Ultimately, sufficient revenue must be generated to justify the investment.

That distinction becomes particularly important when leverage enters the picture. If demand falls short of expectations, or takes longer than anticipated to materialise, the consequences can be amplified by debt.

Private credit, leasing structures and other financing arrangements may help spread the burden across different investors. They do not, however, make the economic risk disappear. Someone still has to earn an adequate return on the capital committed.

This is likely to become one of the defining questions for markets over the next 6-24 months. Investors will need to pay closer attention to actual AI revenues, operating margins, free cash flow after capital expenditure and the utilisation of newly installed computing capacity. Credit spreads and refinancing conditions will matter increasingly, as will the direction of energy prices and the implications for monetary policy.

There is also a distinction worth keeping in mind. An investment cycle requiring enormous amounts of upfront capital is not necessarily a bad investment cycle. Some of the most important technological advances in history have demanded precisely that. The real question is whether the assets being created will generate returns sufficient to compensate investors for their cost, risk and economic lifespan.

For now, the evidence is not sufficient to conclude that AI investment has become excessive. What this week has demonstrated, however, is that investors are becoming more discerning about the financial assumptions behind the technology. The enthusiasm has not disappeared, but it is beginning to share the stage with a more sober assessment of costs and returns.

That may prove to be an important shift. For much of the AI boom, attention has centred on technological breakthroughs, chip demand and the scale of planned investment. Increasingly, the answers may lie in credit markets, balance sheets and cash-flow statements.

Artificial intelligence may well transform the global economy. But even a technological revolution must eventually demonstrate that it can pay its own way.

Sources and References

• reuters.com
• reuters.com
• reuters.com
• fidelity.com
• reuters.com
• federalreserve.gov
• reuters.com
• reuters.com