JPMorgan Chase: "Open source disruption" and "AI security" are not the real issues; there is still room for capital expenditure over the next two years; semiconductor equipment will become the "new bottleneck."

JPMorgan Chase: "Open source disruption" and "AI security" are not the real issues; there is still room for capital expenditure over the next two years; semiconductor equipment will become the "new bottleneck."

Market confidence in AI infrastructure investment has recently wavered. The rise of open-source big data models, increasing calls for AI safety regulation, and the negative free cash flow of hyperscale cloud vendors—these triple pressures have led to a continued downturn in sentiment in the technology hardware sector.

According to TrendFocus, JPMorgan analysts Gokul Hariharan and others addressed the three concerns mentioned above in their Asia Technology Strategy Report released on September 16. The conclusions are: the fundamentals of computing power demand remain unchanged; concerns about AI security regulation are short-term disturbances; the capital expenditure cycle will continue; and semiconductor equipment will become the most critical bottleneck in the entire supply chain around 2027.

Purchasing computing power is also a good business.

The market's first concern is whether "buying computing power" (i.e., AI model manufacturers and vertical AI application providers) can also continue to be profitable, while "selling computing power" is certainly profitable.

Analysts point out that multiple reports indicate that the gross margin of the current inference business has reached 60% to 80%.

The report further estimates that, assuming a model vendor sells AI tokens, each GW of computing power could generate $20 billion to $40 billion in revenue annually , far exceeding the approximately $10 billion per GW in 2025.

The report also cited data from Sapphire Venture, stating that more than 80 vertical AI companies have achieved annual recurring revenue (ARR) exceeding $100 million in a very short period of time, confirming the value creation in the downstream of computing power consumption.

Open source is not a threat, but a catalyst for demand.

The second market concern is that the rise of open-source large models will depress token prices and erode the profit margins of model vendors.

JPMorgan Chase holds a different view. Analysts believe that the open-source model has historically driven a continuous decline in token costs (with a long-term decrease of 80% to 90% in cost per token), and this trend will accelerate the penetration of AI across industries, rather than weaken the overall demand for computing power.

Analysts pointed out three points:

  1. Given the current tight supply of computing power, token consumption for both proprietary cutting-edge models and open-source models will continue to grow strongly.
  2. The widespread adoption of open source will drive AI to penetrate into a wider range of applications;
  3. More vertical AI application companies will leverage low-cost open-source tokens to solve specific industry problems and create new revenue streams.

Analysts specifically pointed out that the temporary rebound in token costs in the first half of 2026 was an "outlier" caused by a severe shortage of computing power coupled with the rise of AI agents, and not a trend reversal.

AI safety oversight: Short-term disruptions will not alter the training rhythm.

The third concern comes from the regulatory front. According to Bloomberg, Anthropic's CEO recently publicly called for a slowdown in the pace of AI model improvements, raising concerns about the future demand for computing power.

Analysts believe this concern has also been amplified.

He judged: " AI models are evolving at a faster pace, and the approaching RSI (Recursive Self-Improvement) precisely demonstrates that the laws of technological evolution and AI expansion have not slowed down. "

Analysts further point out that even if the release of new models may be constrained by factors such as alignment limitations, the training pace of cutting-edge models is unlikely to slow down. At the same time, open-source models will not stop, which will instead force leading model labs to accelerate training to maintain their lead.

JPMorgan Chase believes that the AI security issues stem from regulatory scrutiny triggered by the rapid pace of technological evolution, rather than a sign of shrinking demand. The report predicts that generative AI is poised to open up entirely new market opportunities within the next one to two years, similar to the programming and intelligent agent workflows of the past 18 months.

Capital expenditures: Leverage remains low, cash flow is accelerating.

The capital expenditure financing capabilities of hyperscale cloud vendors are the fourth concern in the market. Many cloud vendors have entered a phase of negative free cash flow, leading to increased external financing pressure.

JPMorgan Chase believes that capital expenditures will still have room to grow in the coming years.

There are three reasons:

  1. Large and hyperscale cloud vendors currently have a net debt-to-equity ratio of only 13%, indicating that their leverage level is still relatively low.
  2. The growth rate of public cloud services is accelerating, and with higher pricing, it will become an important source of operating cash flow growth;
  3. Equity financing from hyperscale cloud vendors and downstream buyers of computing power can provide additional financial support.

Analysts predict that the combined capital expenditures of seven companies—Amazon, Microsoft, Google, Meta, Oracle, Coreweave, and SpaceX—will increase from $443 billion in 2025 to an estimated $933 billion in 2026, and further to $1.577 trillion in 2027, representing year-on-year growth rates of 71%, 110%, and 69%, respectively.

Analysts also cautioned that the rising interest rate environment and changes in risk appetite in some AI infrastructure financing sectors remain variables that require continued monitoring.

Semiconductor Equipment: The Next Bottleneck is Forming

At the supply chain level, JPMorgan Chase believes that by 2027, semiconductor equipment (SPE) will shift from being a bottleneck in "cleanrooms and other areas" to becoming the most critical constraint on the entire supply chain.

Demand-side drivers include:

  • TSMC will significantly increase its equipment investment in N2, N3 and A14 processes;
  • Intel, Samsung foundry, and challengers such as Terafab will also increase capital expenditures;
  • Mature process wafer fabs embark on their first new investment cycle in four years;
  • DRAM and NAND Flash manufacturers will rapidly expand equipment procurement in 2027 and 2028 as cleanroom capacity comes online;
  • China's investment in memory chips is also expected to accelerate significantly over the next two years.

On the supply side, JPMorgan Chase points out that semiconductor equipment manufacturers have begun to enjoy a rare price increase – due to tight capacity, rising input costs, and rush orders from multiple customers.

Packaging and substrates: Bottlenecks persist, optical interconnects rise.

At the component level, JPMorgan believes that IC substrates and advanced packaging remain key bottlenecks.

The supply of substrates is highly concentrated among leading manufacturers such as Unimicron and Ibiden, with capacity expansion cycles lasting approximately 2.5 years. As AI accelerator chips drive increased usage of single-package substrates, and with the introduction of EMIB-T packaging technology by the end of 2027, the supply-demand gap is expected to widen further.

In terms of optical interconnects, JPMorgan Chase predicts that Google, AWS, and Nvidia will fully adopt NPO (Near-Package Optics) solutions within the next two years to address performance bottlenecks in interconnects between GPUs, between GPUs and CPUs, and in storage. CPO (Co-Package Optics) is a long-term trend, but the supply chain maturity will still take time.

Storage: Fundamentals are healthy, but sentiment still needs to recover.

The storage sub-segment is an area where JPMorgan Chase has taken a relatively cautious approach.

The report points out that concerns about HBM downsizing, the reduction in memory usage due to improved algorithm efficiency (such as loop Transformer), and the migration of KV cache to lower-level storage such as DDR or NAND have made the investment logic in this sector quite chaotic.

However, the fundamentals are not pessimistic: supply and demand equilibrium is not expected to occur before 2028, and prices should continue to rise until 2027. JPMorgan believes that if AI sentiment recovers in the fourth quarter of 2026, storage stocks are likely to follow suit. But before that, investor participation may remain lower than in other technology sub-sectors.

Which sectors will see the strongest price increases in 2027?

JPMorgan Chase has identified the sub-sectors where price increases are expected to widen in 2027 compared to 2026:

  • Foundry : TSMC is expected to raise prices for all process nodes in 2027 (except for advanced processes in 2026); the price increase for mature 8-inch processes may reach 10% to 15%, higher than the 8% to 10% in 2026;
  • OSAT : Price increases extended to wire bonding and mature flip chip packaging;
  • Substrate : The price increase cycle has begun, and the introduction of EMIB-T will further tighten supply and demand;
  • High-end CCL (M7 and above): Driven by both AI demand and server specification upgrades, supply growth continues to lag behind demand.
  • Storage : The pace of price increases is expected to slow;
  • PCB materials : The momentum for price increases is also weakening.

Analysts point out that the EPS revision trend in the Asian technology hardware supply chain remains healthy, and valuations are attractive.

Current market sentiment is weak, but there are no clear signs that a downward cycle in capital expenditure is imminent. The key catalyst for the next round of market activity will still come from further validation of the monetization capabilities of the model layer and AI application layer.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.