Everything About AI
NextFin News - Asia's earnings season is delivering a split verdict: the artificial-intelligence supply chain is posting record profits and lifting the region's profit forecasts to their fastest upgrade pace in more than three years, while the rest of the continent contends with a slowing growth outlook, elevated energy costs, and earnings that have yet to catch up with the optimism priced into regional equities. The divergence is redrawing the investment case for Asia. This is no longer a bet on a broad cyclical recovery; it is a wager on whether the AI build-out is a structural shift strong enough to carry an otherwise uneven economy — and on how long the cushion lasts if the build-out slows.
The Numbers Behind the Split
The headline index looks resilient. The MSCI Asia-Pacific Index has climbed roughly 12% this year, even as the S&P 500 has been essentially flat over the same period. But the resilience is narrow. Forward earnings expectations for the index's constituents rose 3.9% over the past month, the largest one-month upgrade in more than three years, and the lift came almost entirely from semiconductor firms in Taiwan and South Korea, according to LSEG IBES data. Among index constituents that have already reported second-quarter results, net profit jumped an average of 29.2% from a year earlier, the strongest growth in two years.
The weakness outside technology is not hypothetical; it is showing up in reported results. In China, Baidu's second-quarter revenue fell short of analyst expectations as online advertising revenue dropped 19% from a year earlier, even as its AI cloud infrastructure revenue grew 50%. Alibaba has missed consensus adjusted earnings per share in each of its last five reports, and analysts have trimmed the current fiscal year's revenue forecast by about 2.5% since the end of March. In consumer-facing businesses, growth is slowing: On Holding's second-quarter sales miss sent the stock down as much as 22%, and its Americas segment growth decelerated to 13% from 17% in the first quarter. These are not isolated misses; they are the earnings side of the same slowdown the IMF is flagging in its growth revisions.
The upgrades are not being shared evenly. Forward 12-month earnings-per-share forecasts for Chinese firms rose just 1.5% over the past month, a fraction of the index-wide gain. Minyue Liu, an equity investment specialist at BNP Paribas Asset Management, put the driver plainly: "The upgrades in Asian companies' earnings expectations are mainly due to upgrades in South Korea and Taiwan on the back of improving semiconductor earnings."
Beneath the equity market, the real economy is cooling. The International Monetary Fund, in its October regional economic outlook, cut its growth forecast for Asia-Pacific to 4.8% in 2026 from 5% in 2025, and expects a further moderation to 4.4% in 2027. The April forecast had been 4.4% for this year, meaning the fund is now pencilling in a slower path than it projected only six months ago. Emerging Asian economies are expected to slow to 5.1% in 2026 from 5.6% in 2025, before easing to 4.9% in 2027. Advanced economies in the region are the exception: projected to expand 3.2% in 2026, up from 2.5% last year, before moderating to 2% in 2027 — a direct reflection of how much of the AI dividend is accruing to the region's most advanced technology exporters.
"Think of countries like Bangladesh, Cambodia, Laos, all these countries are importing oil. They are not benefiting from the AI boom that much. So these countries are a lot more vulnerable and they also have fewer policy buffers," said Krishna Srinivasan, director of the IMF's Asia and Pacific Department.
Why the AI Cushion Is Structural, Not Cyclical
The first question investors need to answer is whether the AI boom cushioning Asia is a cyclical upswing that will mean-revert, or a structural regime shift that will not. The evidence points to structural, with a cyclical overlay that matters most for memory pricing and valuations.
The capital commitments are the clearest signal. Taiwan Semiconductor Manufacturing Company, the world's largest contract chipmaker, has earmarked between $52 billion and $56 billion in capital spending for 2026, up at least a quarter from 2025. It has also raised its full-year revenue growth forecast to slightly above 40% in U.S. dollar terms, up from a previous projection of above 30%. TSMC reported second-quarter 2026 revenue of $40.2 billion, up 33.7% year over year, and guided third-quarter revenue to $44.6 billion to $45.8 billion, roughly 12% sequential growth at the midpoint. These are not spot orders that can be cancelled when sentiment turns. They are multi-year capacity additions tied to data-center construction cycles that take years to permit, build, and equip. Once a hyperscaler commits to a data-center campus, the semiconductor demand embedded in it is locked in for the life of the build.
The memory market tells a parallel story: structural demand layered on a cyclical price recovery. SK Hynix, which commands roughly 58% of global high-bandwidth-memory revenue, posted a record $33 billion operating profit in fiscal 2025 as HBM sales more than doubled. The United States now accounts for about 65% of SK Hynix's revenue, anchored by a partnership with Nvidia that alone contributed 24% of total 2025 sales. Samsung Electronics estimated third-quarter operating profit of 107.4 trillion won, about $80 billion, up 782.5% from a year earlier — the first time a South Korean company has cleared 100 trillion won in a single quarter, and 1.6% above the Yonhap Infomax consensus of about 105.75 trillion won.
The cyclical counterpoint is real and should not be ignored. SK Hynix recorded a 7.73 trillion won operating loss in 2023 during the last memory downturn, when a brutal pricing collapse wiped out margins across the industry. DRAM and NAND pricing, consumer electronics demand, and the pace of hyperscaler capital expenditure all cycle — they always have. But the composition of demand has changed in a way that matters. High-bandwidth memory and advanced packaging are tied to AI accelerator deployment and data-center throughput, not to the smartphone and PC replacement cycles that drove previous memory supercycles. HBM is sold on multi-quarter contracts into a supply chain with only three qualified producers. That scarcity, not the broader electronics cycle, is setting the price today.
This is the distinction that makes the cushion structural rather than cyclical. A cyclical rebound lifts all boats as inventory restocking ripples through the chain. A structural shift reroutes the chain itself — and Asia's AI-linked exporters are now on the new route, while the region's consumer and industrial exporters are still waiting for the old one to recover.
The Second-Order Effect: A Two-Speed Asia
The first-order effect of the AI boom is obvious: chipmakers win. The second-order effect is what is redrawing the map of Asia. Capital, talent, and policy support are concentrating in the economies and sectors plugged into the technology supply chain, while the rest of the region is left with fewer buffers and slower growth. Three transmission channels are doing the work.
The equity channel is the most visible. Information Technology makes up 37% of the MSCI Asia-Pacific Index and was the best-performing sector in the latest session, up 1.81%, followed by Communication Services at 1.05% and Health Care at 0.72%. Consumer Discretionary, at 10% of the index, managed only 0.62%, and Real Estate, at 2%, gained just 0.45%. The earnings revisions are following the same path as the prices: upgrades concentrated in semiconductors, flat-to-modest elsewhere.
The currency and fiscal channel runs deeper. Technology exporters such as Taiwan and South Korea are gaining foreign-exchange inflows and tax revenue from the AI boom, which gives their central banks and treasuries more room to respond to shocks. Energy importers such as Bangladesh, Cambodia, and Laos face the opposite dynamic: elevated oil prices, weaker traditional export demand, and limited policy space. The IMF's growth revisions — up for Korea, Taiwan, and Hong Kong; down for inflation-hit economies without technology gains — are the official recognition of this divergence.
The supply-chain geography channel is the least discussed and the most durable. As AI infrastructure spending compounds, the value is accruing not just to chip designers but to the foundries, memory makers, advanced-packaging houses, and equipment suppliers clustered in Taiwan, South Korea, and parts of Southeast Asia. Each new node shrink and each new HBM generation raises the technical barrier to entry, which concentrates margins among the incumbents rather than diffusing them across the region.
There is, however, a concentration risk embedded in the cushion. The IMF warned that elevated valuations, growing investor leverage, and market concentration could amplify turbulence, and it stated plainly that "risks remain tilted to the downside." The fund added that "a correction in AI-related investment or asset valuations could weigh disproportionately on Asia given its central role in global technology supply chains." When one theme carries a region's earnings upgrades, the region's fortunes become correlated with the sustainability of a single investment cycle — and Asia's exposure to that cycle is larger than America's, not smaller.
The Counter-Thesis: What Could Break the Cushion
The strongest argument against the structural-cushion thesis is that the AI build-out is itself a capital-expenditure wave that will roll over, leaving Asia overexposed to a demand cliff. The scale of the spending is unprecedented: a trillion-dollar global investment bonanza betting on AI evolving at breakneck speed. Regional policymakers are already beginning to factor in a moderation: Singapore's central bank has been cited expecting the AI boom to ease in the latter half of 2027, and the IMF's growth downgrade for the region signals that officials are no longer assuming the current investment pace extends indefinitely.
The bear case has a specific mechanism, not just a mood. Hyperscalers are spending today on the expectation that AI applications will generate enough revenue tomorrow to justify the infrastructure. If monetization lags — if enterprise adoption of agentic AI is slower than capex, if inference costs do not fall fast enough to unlock mass-market use cases, or if regulatory constraints slow deployment — then capital expenditure becomes the variable that adjusts. When it does, it adjusts first in the most cyclical link of the chain: memory pricing. That is where Asia's cushion is thinnest, because memory is where the AI premium is most recently added and most easily reversed.
This counter-thesis is serious enough to define the specific signals that would falsify the cushion view. The structural thesis breaks if hyperscaler capital-expenditure growth falls below roughly 15% year over year for two consecutive quarters, or if high-bandwidth-memory contract prices roll over for two consecutive quarters — either would show that AI demand is cycling rather than compounding. A second falsifying signal: if MSCI Asia-Pacific earnings excluding technology fall more than 5% year over year for two quarters while technology earnings growth drops below 10%, the AI cushion is no longer offsetting the broader weakness; it is merely delaying the recognition of it.
What to Watch: Three Horizons
Short term (this earnings season): The market is already pricing the split. TSMC traded at 2,420 Taiwanese dollars, up 0.62% on the day, while Samsung Electronics fell 0.67% to 222,000 won and SK Hynix dropped 1.31% to 1,277,000 won. The divergence between chip-stock resilience and broader index weakness is the tell: even the beneficiaries are not immune to profit-taking, but the selling is selective rather than indiscriminate.
Medium term (through 2027): Watch whether the earnings upgrades extend beyond semiconductors into equipment makers, advanced-packaging firms, and the industrial suppliers that feed data-center construction. If upgrades stay confined to memory and foundries, the cushion is narrow and fragile. If they spread to the second and third tiers of the supply chain, the structural case strengthens materially.
Long term (structural): The durability of the AI infrastructure cycle itself. Agentic AI is beginning to increase the role of central processing units in AI data centers, adding another layer of silicon demand alongside accelerators — a sign that the demand base is broadening rather than narrowing. The more workloads diversify across chips, the harder the cycle is to break.
Investors should treat the concentration itself as the risk metric. As long as technology remains roughly 37% of the MSCI Asia-Pacific Index and semiconductor earnings drive the majority of upgrades, the region's beta to a single theme will stay elevated. Diversification is not a moral preference here; it is a risk-management necessity for any portfolio that wants Asia's growth without Asia's single-theme exposure.
The base case is that the AI boom continues to cushion Asia through 2027, with Taiwan and South Korea outperforming and energy-importing emerging markets lagging. The upside case is that AI demand broadens into inference workloads and edge devices, pulling more of the region's supply chain along and turning a narrow cushion into a broad recovery. The downside case is a hyperscaler capex slowdown in late 2027 that triggers a memory-price correction and exposes how much of Asia's earnings resilience was borrowed from a single theme.
Asia's earnings season is not telling investors that the region is healthy. It is telling them that health has become concentrated. The AI boom is a cushion, but a cushion only protects what lies beneath it; everything else is still falling.
Explore more exclusive insights at nextfin.ai.
