Asian technology stocks have experienced their third deep correction of over 20% during the current AI-driven uptrend, with market pricing now reflecting pessimistic expectations of imminent earnings per share downgrades or capital expenditure cuts by hyperscalers. JPMorgan believes the reality is heading in the opposite direction.
According to a trading desk report, JPMorgan stated in its Asia tech strategy report released on August 5 that Asian tech stocks and the Philadelphia Semiconductor Index have recently fallen 25% to 30%, yet fundamentals show no signs of material weakness over the next 6 to 12 months. The bank explicitly states that EPS expectations will continue to be revised upward in the coming quarters, with the breadth of revisions expanding further, and hyperscaler capex guidance for 2027 trending higher.
JPMorgan declared in the report that now is the time to buy Asian tech stocks, arguing that valuations have become reasonable after this correction. Asian tech stocks, excluding memory, are currently trading about one standard deviation above their 10-year average price-to-earnings ratio, not excessively expensive.
AI Scaling Laws Intact, Compute Demand Accelerating
JPMorgan emphasizes that the core logic of this uptrend—the AI Scaling Laws—remains valid. The rule of thumb that deploying 10x more compute to train a model yields roughly 2x more intelligence is largely holding. Multiple leading AI labs are racing to advance model capabilities, with some making progress on recursive self-improvement, which is expected to accelerate the pace of frontier model evolution.
Public cloud revenue data confirms the strong momentum of AI compute consumption. The combined year-over-year revenue growth rate for the four major cloud providers—Google, Microsoft, Amazon, and Oracle—has accelerated for several consecutive quarters. In the second quarter of 2026, incremental revenue reached approximately $150 billion, nearly doubling from the first quarter. Meanwhile, the contract backlogs of the top three public cloud vendors continue to expand—Google Cloud's backlog exceeds $514 billion, Amazon AWS stands at $496 billion, and Microsoft's commercial remaining performance obligations total $678 billion—providing solid support for hyperscalers to maintain high capital expenditure levels.
The rise of open-source large language models is not seen as a negative. The bank believes that the increasingly competitive landscape of frontier models, with frequent changes in leadership, actually benefits the continued expansion of tech hardware demand and AI compute capex.
Hyperscaler Capex to Remain Strong in 2027, Financing May Bring Short-Term Volatility
Hyperscaler capital expenditure is expected to maintain strong growth in 2027. According to forecasts by analysts Doug Anmuth and Samik Chatterjee, total capital expenditure for the seven major hyperscalers—Amazon, Microsoft, Google, Meta, Oracle, CoreWeave, and SpaceX—is projected to grow approximately 103% year-over-year to about $901 billion in 2026, and further increase by 65% to roughly $1.49 trillion in 2027. SpaceX is also regarded as an emerging heavyweight capital expenditure player, potentially on par with traditional cloud providers.
Regarding market concerns about the sustainability of capex, free cash flow for hyperscalers is expected to turn negative in the second half of 2026 and into 2027. However, the report notes that their balance sheets remain generally healthy—as of the second quarter of 2026, the combined net debt-to-equity ratio for hyperscalers was about 12%. The bank expects these companies to supplement funding through equity and debt financing channels, without scaling back AI compute investments in 2027.
On the inventory front, no signs of stockpile build-up have been detected for key AI components such as GPUs, ASICs, or memory. This is distinctly different from the cyclical peaks seen in auto parts in 2022 or cloud DRAM chips in 2017. The bank forecasts that AI chip demand will continue to outpace supply chain capacity and data center power budgets over the next 12 months.
Semiconductor Equipment and IC Substrates Offer Best Value; Memory Narrative Has Risks
In terms of sub-sector allocation, JPMorgan believes semiconductor equipment is the most advantageous segment over the next 12 months. The bank expects significant upward revisions to capital expenditure forecasts for TSMC and major memory manufacturers, with equipment and cleanroom space potentially becoming the next bottleneck for 2027 to 2028.
IC substrates are identified as the most fundamentally sound sub-sector within the components space, benefiting from factors including the increasing size of AI accelerator packages, strong server CPU demand, the accelerated adoption of EMIB-T packaging technology from late 2027, and incremental demand from CPO-related solutions. Supply side concentration is high, with limited capacity expansion possible in the next two years. Profit margins remain well below previous peaks, leaving ample room for upward EPS revisions.
The memory sector is a notable exception, with a more cautious stance. While supply-demand fundamentals are healthy, with a supply gap expected to persist for 2 to 3 years, Nvidia and AMD are planning to launch lower HBM-density accelerators, such as Vera Rubin and MI455 with 8-HBM4 stacks, and reduce the SoCAMM memory configuration for the 2027 Vera CPU to address DRAM supply tightness and cost pressures. This behavior closely mirrors downshifting patterns seen in past cycles, which could cap valuation upside for memory stocks. After a deep correction of over 40%, memory stocks may see a strong rebound over the next 6 months, but are unlikely to recover the prior highs from May 2026 in the short term.
Interconnect Efficiency and Power Supply Could Be Next Key Variables
The report also offers forward-looking judgments on potential medium-term bottleneck shifts. Current model compute utilization rates for GPU and ASIC clusters are generally low, ranging from 20% to 40%, with some large clusters falling below 20%. As capital expenditure rises, power budgets tighten, and chip supply remains constrained, AI infrastructure providers and model labs will place greater emphasis on improving cluster efficiency. Interconnect technology—including CPO and optical connectivity, as well as network overhead optimization—is expected to become the next key bottleneck, driving accelerated adoption of 3DSoIC advanced packaging, SRAM memory tier expansion, and CPO on interposers.
Over a longer time horizon, the report suggests that AI chips will remain the primary bottleneck for compute infrastructure for most of 2026 and 2027. However, over the next 18 to 24 months, as semiconductor supply continues to ramp up, data center deployment delays and power supply constraints—including grid access and behind-the-meter power solutions—could replace chips as the core variable limiting AI compute expansion. This risk factor is expected to come into clearer view from the second half of 2027 through 2028.
Comments