The AI bubble meets reporting season

There is little doubt that artificial intelligence is a transformative technology. It is already changing how businesses operate, how software is developed and how information is found and processed. At Forager, we use it extensively ourselves.

But a great technology does not automatically make every company associated with it a great investment. As we wrote recently in my letter to investors, what began as a rational boom has increasingly taken on the characteristics of a bubble. Investor enthusiasm, enormous capital spending commitments and some remarkable valuations have all been built on the assumption that the eventual financial returns will be equally remarkable.

The current earnings season is providing an important test of that assumption.

From spending at any cost to showing the returns

Until recently, investors rewarded almost any announcement of higher AI investment. New data centres and larger capital expenditure budgets were treated as evidence of future growth. If share prices are anything to go by, the questions are now more nuanced.

How much revenue is the spending generating? How much cash is left after paying for the infrastructure? And what return will companies ultimately earn on the trillions (yes, trillions) of dollars being invested?

Alphabet (NASDAQ: GOOG) provided an early indication of how the mood might be changing. Google’s parent reported exceptional growth in its cloud business, but its shares still fell by around 7% on the day of its results after the company increased its expected capital expenditure guidance.

A year ago, the combination of strong cloud growth and higher AI spending might have been enough to push the shares higher. This time, investors focused more closely on the cash going out the door.

Meta (NASDAQ: META) faced a similar reaction on the day it released its results, with the share price falling almost 12%. Revenue continued to grow strongly, helped by its highly profitable advertising business. But free cash flow fell significantly as spending on data centres, equipment and AI talent accelerated.

The business is still growing. The issue is that the cost of pursuing its AI ambitions is growing even faster.

That does not necessarily make the spending irrational. Meta has a huge user base, valuable data and one of the world’s strongest advertising platforms. If AI improves engagement or advertising returns, the investment could be highly profitable. But investors are no longer treating the outcome as automatic.

Shares in both of these stocks have since recovered at least some of the losses alongside a rally in the rest of the big-tech sector.

Good results can still be rewarded

The AI boom has not ended, and markets are still willing to reward companies that can demonstrate tangible results.

Microsoft (NASDAQ: MSFT) received a much stronger response after reporting faster-than-expected growth from its cloud infrastructure business Azure and continued adoption of its Copilot products. The company also reassured investors that it expected to remain strongly cash-generative despite its enormous investment program. Its shares rose by almost 16% following the result.

Amazon (NASDAQ: AMZN) also showed evidence that its AI and data centre spending is translating into demand. Amazon Web Services reported its strongest growth in several years, while contracted future revenue continued to rise.

The market rewarded the result, with the share price increasing over 15% on the day, but the numbers also showed the size of the bet being made. Amazon is spending heavily on new capacity while its free cash flow has come under pressure.

Investors are currently comfortable with that trade-off because cloud demand is strong and much of the future capacity appears to be spoken for. Whether those investments produce attractive returns over their full lives remains an open question.

The contrasting reactions to Microsoft, Amazon, Alphabet and Meta are important. They suggest investors are starting to distinguish between AI spending that is producing measurable revenue and spending that still relies heavily on promises about the future. Elon Musk’s SpaceX (NASDAQ: SPCX) is perhaps the poster child for the latter, and its share price has fallen well over 30% since its post-IPO peak.

This is a healthy development for markets and economies. Allocating capital to its most productive uses is one of the most important contributions investors make to the economy. But it is also a huge risk for the share prices of those companies that have ridden the AI-wave up and haven’t yet produced any results of substance.

The physical economy is pushing back

This reporting season is also highlighting another constraint: AI does not exist solely in the digital world.

It requires enormous amounts of electricity, land, advanced chips, memory, cooling equipment and construction capacity. Those resources are not unlimited.

The rapid expansion of AI data centres is already placing pressure on global supply chains and energy systems. Companies are competing for chips, engineers, electricity connections and suitable sites.

That is one reason the current AI build-out looks different from many previous software booms. Technology businesses are accustomed to scaling quickly with relatively little physical capital. Building AI infrastructure looks more like constructing railways or electricity grids.

These projects can create significant economic value. They can also produce overcapacity, falling prices and poor returns for the investors funding them.

A data centre may be full today, but investors still need to consider what it will be worth in five or ten years. Chips also become obsolete quickly. New competitors can reduce prices, and customers may develop cheaper ways to train and operate AI models.

The amount of money being spent makes these questions increasingly important.

A bubble does not mean everything is expensive

Calling parts of the AI market a bubble is not the same as predicting that share prices will collapse tomorrow.

Bubbles can persist for much longer than seems rational. Strong results from companies such as Microsoft and Amazon can extend enthusiasm further.

Nor does it mean every AI-related business is a poor investment. Some companies will earn excellent returns from the current spending cycle. Others will become valuable long-term businesses that do not yet exist.

The more useful observation is that the enthusiasm is concentrated.

As money has poured into a relatively small group of AI beneficiaries, large parts of the sharemarket have been ignored or actively sold. Profitable software companies have been treated as potential AI casualties regardless of their individual economics. Smaller companies, old-economy businesses and entire markets remain priced for much less optimistic futures, as you can see in the chart below.

That is why we believe the current environment looks more like the technology bubble of 2000 than the broad excesses preceding the Global Financial Crisis. The expensive part of the market is extremely expensive, but it has not taken everything else with it. There are still sensible, cash-generative businesses available at valuations that do not require heroic assumptions.

There are some seen as AI losers that have been severely punished over the past 12 months.

Recent results for software companies Salesforce (NASDAQ:CRM) and Atlassian (NASDAQ: TEAM) have notably challenged the "AI-loser" narrative. After a torrid year where the share prices of both companies tumbled dramatically, both businesses reported strong results for the June quarter that the market rewarded with significant share price rallies. Atlassian in particular has seen a share price rise of more than 60% over the past month.

Earnings are beginning to matter again

The most important message from this reporting season is not that the AI boom is over.

It is that investors are beginning to ask harder questions.

Alphabet delivered outstanding growth and was punished because spending and cash flow disappointed. Microsoft was rewarded because it showed stronger evidence of commercial demand and ongoing cash generation. Meta demonstrated that a strong existing business can still face scrutiny when the cost of its ambitions rises dramatically. Amazon showed that markets will tolerate enormous spending, but only while growth remains exceptional.

Salesforce and Atlassian showed that they aren’t dying just yet and that low expectations are easier to beat.

That selectivity may come and go. Another wave of enthusiasm could easily overwhelm concerns about valuations and returns. But the economic reality will eventually matter. Trillions of dollars are being invested, and those dollars will need to earn an acceptable return.

Artificial intelligence will almost certainly change the world. The current earnings season is beginning to test whether today’s share prices have already assumed that every dollar spent changing it will be extraordinarily profitable.

History suggests that is a much less certain proposition.