Cheaper AI, new earnings questions

U.S. equities do not look especially expensive on a forward price-to-earnings basis, but cyclically adjusted (Shiller CAPE) valuations remain historically rich. The difference reflects consensus expectations that today’s extraordinary earnings growth, supported by the unprecedented speed and scale of the AI buildout, will persist rather than revert to historical norms. See the chart. Whether that assumption holds is the question. The emergence of powerful Chinese AI models, including Moonshot's Kimi K3, could put it to the test. As competition intensifies, the debate is no longer just who will build the best model, but who will capture the economic rent. We think cheaper AI changes the winners, not the investment case. Instead, it reinforces our preference for AI infrastructure over the increasingly competitive model layer.

The cost of AI is emerging as a key concern for companies deploying it. Gartner expects worldwide spending on AI models and platforms to reach $64 billion in 2026, up 63% from 2025. As enterprise AI bills rise, companies have a stronger incentive to contain costs through model routing and lower-cost models. Reflecting that shift, OpenRouter data on the 10 most widely used AI models show Chinese models processing roughly 23 trillion tokens a week, compared with about 12 trillion tokens for U.S. rivals. Together, these trends could erode the pricing power of frontier model developers even as AI adoption accelerates. Meanwhile, AI sovereignty is encouraging countries and companies to build their own AI capabilities, reinforcing demand for open-weight models and the infrastructure needed to train, host, and deploy AI at scale.

It is too early for this quarter’s reported earnings to reveal the impact of lower-cost and open-weight AI models. Instead, the first clues are likely to come from management commentary on earnings calls. We will be listening for two sets of signals. From hyperscalers, we want to know whether they remain committed to the AI buildout through their capital spending plans, how they balance capital discipline in a higher cost of capital environment, and whether those investments are generating returns. Alphabet’s earnings last week sharpened that focus after free cash flow turned negative for the first time. Decisions on whether and how to deploy lower-cost models on cloud platforms also matter because hyperscaler cash flows — and increasingly, debt financing — fund much of today’s AI ecosystem. From companies across the broader economy, we will look for comments on how they are responding to rising AI costs and the increasing importance of AI sovereignty, including how that is shaping model routing, model choice, and returns on AI adoption. Together, these discussions could offer clues about whether greater competition changes the winners rather than the investment case.

Recent developments may shift where AI’s economic rent is captured. But cheaper models could also broaden AI adoption while AI sovereignty reinforces demand for AI infrastructure. We remain overweight on the AI theme, but it requires increasingly selective and active positioning. Rather than trying to identify long-term winners in the increasingly competitive model layer, we prefer investing around AI scarcity. More broadly, the U.S. still stands out for resilient corporate earnings, even outside the AI ecosystem.

Geopolitical risks returned to the fore last week. A sharp escalation in the Middle East conflict, alongside new global tariffs announced by U.S. President Donald Trump, pushed Brent crude oil prices briefly back above $100 a barrel and lifted 10-year Treasury yields to their highest levels since early 2025 as markets priced in a greater inflation risk. Technology shares also came under pressure after Alphabet raised its capital spending plans, with the Nasdaq ending the week down more than 1%.

Markets face a packed week against a backdrop of escalating Middle East tensions and President Trump’s new global trade tariffs. Investors will watch policy decisions from the Federal Reserve, Bank of England, and Bank of Japan, alongside U.S. GDP and PCE inflation data. Together, they will test whether resilient growth and sticky inflation still support our high-for-longer rate view.