At the top, the slide shows how demand from end users and capital from investors flow into developers of AI tools, including model builders, application creators, enterprise integrators and platform enablers. This application layer is where many frontier AI companies sit, from leading model labs to enterprise software firms and sector-specific applications in areas like healthcare, legal and financial services. In the middle, compute providers act as the bridge between software demand and physical infrastructure, translating AI adoption into spending needs across the broader ecosystem.

At the bottom, the chart highlights how broad the AI value chain has become: one company’s capex is another company’s revenue, and AI exposure now reaches far beyond chips into semiconductor design and production, data center real estate, construction, servers and networking, cooling, electrical equipment, regulated utilities, independent power producers, transmission, reliability and communications infrastructure. The key takeaway is that AI is creating opportunities across growth and value sectors, across market caps and geographies, but selectivity remains critical—competition is intense, technologies are changing quickly and not every business model in the ecosystem will prove durable.

The chart compares major U.S. and Chinese AI models by performance and cost, with the vertical axis showing capability and the horizontal axis showing the cost per task on a logarithmic scale, where models farther left are cheaper. Blue dots represent U.S. models and orange dots represent Chinese models; U.S. models still occupy many of leading performing positions, but Chinese models have moved closer to the frontier and often appear at lower cost points, suggesting a narrowing performance gap and a stronger emphasis on compute efficiency.

The key takeaway is that U.S.–China dynamics in AI are multifaceted, not a single scoreboard. The U.S. continues to lead at the frontier—pushing the most advanced models, compute and ecosystem development—while China’s advantage is increasingly around diffusion at scale, delivering “good enough” AI into real workflows at lower cost. Each new generation of more capable and cheaper models can unlock additional enterprise and consumer use cases, supporting further infrastructure spending, but the constraints are diverging: China remains more compute-limited, while the U.S. is increasingly facing power, permitting and data-center execution bottlenecks.

This chart shows the share of weekly tokens consumed on OpenRouter, a developer-focused model aggregation platform, by model origin. Over the period, usage shifts meaningfully from U.S. models toward Chinese models, which overtake their U.S. counterparts and sustain the lead toward the end. The gray line captures other-region models, which also rise meaningfully late in the period, suggesting developers are spreading usage across a broader set of credible options from Europe, India and other markets.

The key takeaway is that Chinese models are gaining share on developer and agentic platforms, helped by lower costs and strong appeal to price-sensitive users optimizing for tokens-per-dollar. However, this should not be read as a complete picture of AI adoption: OpenRouter excludes major enterprise channels such as Azure, Bedrock, Vertex and direct provider APIs, where U.S. models like GPT, Gemini, Claude and Llama are more heavily represented, so U.S. usage may be undercounted and its decline may look steeper than reality. Importantly, falling share does not necessarily mean falling usage, as absolute token volumes could still be growing; rather, the chart points to natural fragmentation as more Chinese and other open-weight models become available and competitive.

The lines show paid AI adoption rising across major industries since 2023, led by technology and media, with finance and insurance also moving meaningfully higher. The key takeaway is that the story is shifting from AI makers (AI spenders and AI earners) to AI users: adoption is broadening, but businesses now need to prove that AI can be deployed with discipline around people, processes and price. As compute costs, capacity constraints and monetization questions remain in focus, investors should expect volatility, but diversification is not anti-AI—it is a way to stay exposed as leadership potentially rotates from the companies building AI infrastructure to the industries using AI to improve productivity, workflows and long-term fundamentals. 

On the left, the bars show China’s annual installations of industrial robots, while the blue line shows China’s share of global installations. The key message is that China has become a dominant force in physical AI adoption: robot installations have risen sharply over the past decade and now account for roughly half of global demand, underscoring both the scale of China’s manufacturing base and the speed at which automation is being deployed.

On the right, the donut chart breaks down AI-related investment in China by sector in 2025. Robotics is the largest category at 38%, followed by AI hardware and technology, AI native applications, traditional industrial applications, and autonomous driving, highlighting that China’s AI pathway is less about frontier models alone and more about localization, cost efficiency, and real-economy diffusion. Taken together, the slide suggests that China’s AI opportunity may increasingly come through “AI users”—companies applying AI to improve productivity in manufacturing, logistics, mobility, and industrial processes. 

The slide shows robotics progressing from fixed-function industrial machines, such as SCARA robots and cobots, toward increasingly intelligent and generalized systems, including autonomous mobile robots, mobile manipulators and humanoids. As generalization improves, robots become capable of operating across less structured environments and performing a wider range of tasks.

The key takeaway is that robotics is expanding beyond industrial automation into commercial and household applications. Shared formats such as cobots, drones, humanoids and quadrupeds can support activities ranging from manufacturing and construction to healthcare, education and elderly care, broadening the market opportunity across the economy.

On the chart, the purple bars show the share of tasks in each occupational category that could theoretically be covered by AI, while the grey bars show where AI coverage is already being observed in practice. Exposure is highest in knowledge-based roles such as computer and mathematical, business and financial, legal, office and administrative, and architecture and engineering jobs, while more physical or location-based roles such as construction, transportation, farming, and personal care show much lower exposure.

The key takeaway is the large gap between AI’s potential reach and its current adoption. This suggests the productivity impact is likely to be phased and uneven rather than immediate, depending on whether companies can redesign workflows, reskill workers, and integrate AI into day-to-day processes. For investors, that creates dispersion: the opportunity is not just in AI technology itself, but in identifying the companies and sectors that can convert theoretical exposure into measurable productivity, cost, and earnings benefits—especially in an economy where slower labor-force growth means future growth must rely more on productivity than simply adding workers.

On the left, the blue bars show capex from the major U.S. hyperscalers, while the grey dashed line shows the proportion of capex spending as a share of operating cash flow. The chart highlights the scale of the U.S. AI buildout: spending has accelerated sharply since 2023 and is expected to approach USD 1trillion by 2028, with capex absorbing a much larger share of cash flow than in the past. This underscores both the strength of AI-related demand and the growing investor debate around whether hyperscalers can convert this infrastructure investment into durable revenue, margins and returns.

On the right, the orange bars show capex from the major Chinese hyperscalers, with the same grey dashed line tracking capex intensity relative to operating cash flow. China’s AI spending is rising from a much smaller base, but the direction is similar: capex is expected to climb meaningfully over the next few years, while taking up a larger share of cash generation. Taken together, the slide shows that the AI infrastructure race is global, but the U.S. remains much larger in absolute scale, while China’s buildout is more about cost efficiency, localization and domestic adoption. For investors, the near-term spending cycle still supports AI infrastructure and supply-chain beneficiaries, but the longer-term winners will be those that can turn heavy capex into visible monetization.

On the left, the chart helps reframe AI exposure as a global ecosystem rather than a single U.S. mega-cap trade. The U.S. remains the center of frontier models, hyperscalers and premium innovation, but Asia provides much of the physical infrastructure that allows AI to scale—memory, High Bandwidth Memory (HBM), advanced chips, servers, equipment, components and manufacturing depth. The key takeaway is that diversification across AI does not mean stepping away from the theme; it means owning different parts of the value chain where earnings support remains strong and valuations may be more attractive.

On the right, the supply-chain breakdown shows why Asia ex-Japan is so central to the AI buildout, particularly across semiconductors, hardware, logic chips, and assembly, testing and packaging. Korea is most directly tied to HBM and memory, Taiwan to foundry and AI server manufacturing, Japan to equipment and materials, and China to localization, domestic cloud and enterprise adoption. For investors, the geography gap is important: as AI leadership rotates from spenders to earners and eventually users, the opportunity set should broaden across regions and sectors, favoring active selection of companies that can turn AI demand into visible earnings, productivity gains and returns.

On the left, the chart shows why AI infrastructure is not just about adding more data centers, but upgrading them for much more power-intensive workloads. Global data center capacity is around 125 gigawatts today, but only 10–13 gigawatts is estimated to be AI-capable, because AI servers require far higher power density, liquid cooling rather than traditional air cooling, and significantly more electrical capacity.

On the right, the stacked bars show that global data center capacity is expected to grow significantly through the end of the decade, with AI workloads shown in blue becoming the dominant driver of new demand. The key takeaway is that the AI buildout is a broad infrastructure cycle—not just chips and servers, but also power, cooling, networking, equipment and financing. For investors, that means the opportunity extends well beyond the obvious AI leaders into the companies enabling the physical capacity needed to run AI at scale.

On the left, the bars show forecast data center electricity consumption in the U.S. and China, while the lines show data centers’ share of total power demand. The key message is that power is becoming the key constraint for AI: after decades of limited U.S. electricity demand growth, data centers are now driving a sharp increase in load, with their share of U.S. power demand expected to rise meaningfully by 2030.

On the right, the stacked area chart shows U.S. electricity generation by source, highlighting why the AI buildout will require an “all hands on deck” approach across renewables, natural gas, nuclear, grid upgrades and efficiency gains. The takeaway is that AI infrastructure is no longer just about chips, servers or data centers; it is also an energy and transmission story. For investors, this broadens the opportunity set toward power generation, grid modernization and infrastructure assets, while also suggesting that AI compute may remain supply-constrained within U.S. in the years ahead.

On the left, the chart shows South Korea’s export growth by category, with dark blue line capturing AI-related technology, light blue line showing other technology, grey line showing non-tech exports, and the green line showing earnings growth. The key message is that Korea’s export cycle has re-accelerated sharply, led by AI-related technology, and earnings are now responding even more strongly—consistent with Korea’s role as a major beneficiary of memory and HBM demand in the AI supply chain.

On the right, Taiwan shows a similar but somewhat steadier pattern, with AI-related technology exports again driving the improvement and earnings growth turning higher alongside the export recovery. Taken together, the slide reinforces the geography gap in AI investing: while the U.S. may dominate the software and hyperscaler narrative, Korea and Taiwan are capturing the physical infrastructure demand behind the AI buildout, with export momentum increasingly backed by earnings rather than sentiment alone. 

On the left, the bars compare the technology makeup of major equity markets, separating “hard tech” in blue from “soft tech” in grey. The key message is that Asia’s AI exposure is very different from the U.S.: Taiwan and South Korea are dominated by hard-tech representation, reflecting their central role in semiconductors, hardware and the physical supply chain, while offshore China has more soft-tech exposure versus China A-shares, which show a more balanced but still hardware-led profile.

On the right, the chart shows price returns across AI-related industries in Asia ex-Japan, with the table highlighting index weights and valuation levels. Semiconductors and hardware have delivered the strongest recent performance, while software and hyperscalers have been weaker, reinforcing that Asia’s AI opportunity is more closely tied to the “picks and shovels” of the buildout—memory, chips, hardware and infrastructure—than to the software narrative alone.

The broader takeaway is that AI optimism has not disappeared, but it has become more rational and selective. Investors still recognize AI as a long-term trend and are willing to support hyperscaler investment, but they are also asking tougher questions around capex, cash-flow pressure and when ROI will become visible. That creates a higher bar for software names, where AI monetization is often less directly visible and more dependent on adoption, pricing power and workflow integration, reinforcing the need to focus on companies that can convert AI demand into measurable earnings, productivity gains and returns.

The chart compares how semiconductors, shown in blue; hardware, shown in green; and other sectors, shown in pale blue, contribute to earnings growth across major markets. The key message is that the AI capex boom is now flowing directly into earnings, especially in Asia: Korea and Taiwan stand out with very large semiconductor and hardware contributions, while Japan, the U.S. and China A-shares also benefit meaningfully. This helps explain why international and emerging market equities have been supported by earnings upgrades rather than just multiple expansion.

The more durable takeaway is that semiconductors and hardware are becoming a major earnings engine across global markets, with the strongest visibility in regions tied most directly to AI infrastructure, memory, chips and hardware. For investors, this reinforces the idea that AI exposure is not limited to U.S. mega-cap technology; the earnings beneficiaries are increasingly global, especially in the Asian supply chain. 

The chart shows the combined weight of the top 10 companies across emerging markets, the U.S., Europe and Japan, with colors indicating each company’s sector and white lines separating individual names. The key message is that market concentration is high across regions. Within emerging markets, the green technology segment is the largest, led by TSMC, Samsung Electronics and SK Hynix, underscoring how much emerging market exposure is tied to the AI hardware and semiconductor supply chain. But the AI linkage is broader than the green tech bars alone: companies such as Tencent and Alibaba sit in communication services and consumer-related categories, yet remain important AI beneficiaries through cloud, platforms, data, applications and digital ecosystems. The takeaway is that AI exposure is highly concentrated, but also more diversified across sectors than the headline sector labels suggest—reinforcing the need to look beneath index classifications when assessing where market leadership and AI-related earnings may broaden next.