The Chip Embargo That Backfired
When the US moved to restrict the export of advanced semiconductors and chip-making equipment to China, the intention was to slow China’s progress in artificial intelligence and advanced computing. In the short term, it did. But it has also had an unintended consequence: it has forced China to accelerate the domestic capability the embargo was meant to suppress.
Chinese AI developers, cut off from the latest US and Taiwanese chips, have found creative workarounds. One approach that has drawn attention is training AI models on rented compute capacity in international data centres; outside the reach of export controls, and then physically transporting the resulting training data back to China. A model’s training data, distilled down to a few terabytes, is small enough to carry on a hard drive or transfer online, effectively sidestepping the restriction on hardware entirely.
Just as importantly, the embargo has, ironically, been a lifeline to China’s domestic chip industry. Local semiconductor manufacturers, which may well have struggled to compete on price and performance against established US and Taiwanese producers, have instead been handed a captive market. With Chinese AI companies locked out of foreign supply, demand has poured into domestic chipmakers, funding the research and development needed to close the technology gap. The result is that Chinese semiconductor manufacturers are catching up to their US and Taiwanese counterparts faster than they might have absent the restrictions; a textbook case of protectionism inadvertently building the industry it was designed to contain.
Robots, Restrictions, and a Different Strategic Bet
A similar dynamic is unfolding in robotics. The US has moved to restrict Chinese-made humanoid robots from its market, mirroring the chip playbook. Yet China is already home to roughly 200 humanoid robotics companies, and the country’s manufacturing base has been racing toward “dark factories”; fully automated production lines that run without lighting, climate control, or, ultimately, human workers. As this automation matures, the need for human labour on factory floors becomes progressively less relevant.
What is notable is the difference in strategic emphasis between the two superpowers. The US is pouring capital into data centre infrastructure; the compute layer of the AI stack. China, by contrast, appears to be prioritising electrification and physical automation; the deployment layer, where AI meets the real economy through robotics and manufacturing. These are two different bets on where the value in AI ultimately accrues: in the infrastructure that trains the models, or in the physical and industrial systems that put AI to work.
A Question Worth Asking About Market Concentration
This distinction matters for investors because it cuts to a genuine risk building in US equity markets. AI-related companies; chipmakers, hyperscalers, and data centre operators, now make up an estimated 45% of the US stock market. That is a significant concentration of capital in a single investment thesis: that the buildout of AI infrastructure will translate into commensurate returns for the shareholders who funded it.
But infrastructure spend and shareholder return are not the same thing. History offers a cautionary parallel in the dot-com era buildout of fibre-optic cable; the infrastructure was genuinely transformative and is still in use today, but many of the companies that built it did not survive to enjoy the benefits, and their shareholders were largely wiped out. The risk for today’s AI infrastructure names is analogous; the market may eventually distinguish between the companies building the picks and shovels of the AI economy and the (often different) companies and sectors that ultimately capture the productivity gains AI delivers throughout the broader economy.
Positioning for the Correction We Can’t Time
We don’t know when or how sharply this realisation might hit markets, but a correction concentrated in AI infrastructure names is a real enough possibility that it warrants portfolio discipline today, not after the fact. This is precisely the argument for staying diversified across both sectors and regions, rather than concentrating exposure in the narrow set of companies currently driving index returns.
Two sectors Morningstar find particularly interesting from a valuation standpoint are healthcare and consumer staples; both currently look undervalued relative to the broader market, and both stand to benefit meaningfully from the proliferation of AI through the economy, even though neither is typically thought of as an “AI stock”:
- Healthcare stands to benefit as AI materially compresses the time and cost of drug discovery, accelerates clinical trial design and patient recruitment, and improves diagnostic accuracy; all of which flow through to R&D productivity and, ultimately, earnings for pharmaceutical and healthcare companies, without those companies needing to spend billions on their own data centres.
- Consumer staples stand to benefit as AI-driven demand forecasting, supply chain optimisation, and automated manufacturing (including the robotics and “dark factory” trends discussed above) compress costs and improve margins for household-name manufacturers and retailers; translating AI’s productivity gains directly into more resilient earnings, in a sector that already trades on defensive, less demanding valuations.
In both cases, these are sectors positioned to capture the benefits of AI adoption without carrying the concentration risk, or the capital intensity, of the companies currently building the infrastructure. As we navigate what remains a genuinely transformative but increasingly crowded AI trade, that combination of participation and diversification is where Morningstar see an attractive risk-adjusted opportunity today.