Want to understand the current state of AI? Check out these charts.

Stanford’s 2026 AI Index says AI model performance, adoption, and industry spending are all still accelerating, with AI data centers now drawing 29.6 gigawatts of power and GPT-4o water use potentially exceeding the needs of 12 million people. The US and China are nearly tied on model performance, while the US leads in capital and data centers and China leads in publications, patents, and robotics. The report also flags fragile chip supply chains, rising regulatory activity, and early signs of labor-market pressure, including a nearly 20% drop in software developer employment for ages 22-25 since 2022.

Analysis

The market is underpricing how quickly AI is shifting from a model race to an infrastructure choke point. When frontier performance converges, incremental returns migrate from software differentiation to whoever controls power, chips, memory, networking, and deployment capacity. That argues the next leg of alpha is less about picking the “best model” and more about owning the bottlenecks that scale with every inference request, especially where capex intensity is still being revised upward.

TSM remains the clearest structural winner on the supply side, but the setup is more nuanced than a simple foundry bull case. Heavy concentration in one fabrication node creates pricing power, yet it also concentrates geopolitical and execution risk; any export-control escalation, Taiwan-premium de-rating, or customer diversification effort could cap multiple expansion. The more interesting second-order trade is that the beneficiaries of AI capex may rotate toward adjacent semicap equipment, advanced packaging, HBM, and grid/thermal management providers, where demand is still early in the cycle and less crowded.

On the China side, BIDU and BABA are less about raw model leadership and more about domestic distribution, enterprise adoption, and policy-supported localization. If model quality continues to converge, the key variable becomes cost-per-task and how quickly Chinese incumbents can bundle AI into search, cloud, commerce, and workflow software without depending on US IP. That creates upside optionality if domestic deployment accelerates, but the trade is vulnerable to any renewed compute restrictions or a slower-than-expected monetization curve.

GOOGL is the cleanest way to express the “AI is monetizing faster than the market thinks” view, but not for the obvious reason. The important catalyst is margin mix: if AI traffic can be monetized without proportional TAC or capex leakage, the market will re-rate the franchise as an operating leverage story rather than a capital-spending story. The contrarian risk is that AI search and agentic use cases cannibalize high-margin query economics before new formats fully replace them, so the stock likely trades on confidence in monetization cadence over the next 2-4 quarters, not on model performance alone.

Key Decisions for Investors

• Long TSM / short SOXX into any post-earnings volatility: own the manufacturer with the tightest capacity bottleneck while fading the broader semi basket where multiple expansion is already crowded; target 6-12 months, with downside if AI capex decelerates faster than expected.
• Buy GOOGL on 3-6 month horizon via call spreads: the asymmetric bet is that AI features lift engagement and pricing without blowing out traffic-acquisition or inference costs; risk/reward improves if the market remains focused on capex rather than unit economics.
• Relative-value long BIDU vs short BABA over 3-9 months: BIDU has more direct AI leverage if domestic model adoption and search monetization inflect, while BABA faces higher execution dispersion as AI becomes a feature inside a broader commerce platform.
• Add a tactical long in semicap / advanced packaging exposure versus mega-cap AI software beneficiaries: the second-order winners should have cleaner earnings sensitivity to incremental fab buildouts and are less exposed to product-cycle narrative risk.
• Avoid outright shorting AI leaders until there is evidence of monetization failure; instead use put spreads on the names most exposed to capex disappointment or regulatory friction, since the near-term trend is still supported by adoption and balance-sheet strength.