Six Key Issues in the Early Stage of the AI Mega Cycle

With the onset of the AI mega-cycle, the global technology industry is in the early stages of structural transformation. Although there are still discussions in the market regarding capital expenditure and return rhythms, the potential enhancement of productivity by AI has established the theme of its long-cycle growth. The year 2026 will be a key window for observing constraints in AI infrastructure, the evolution of chip competition, and the integration of end applications.

According to news from the Chasing Wind Trading Desk, HSBC's Nicolas Cote-Colisson analysis team released a report on the 7th, stating that AI is currently in the initial stage of a mega-cycle. Although there was a situation in October 2025 where capital expenditure expectations led early revenues, considering the speed of AI development and its potential to drive over $110 trillion in global GDP, the market remains optimistic about the prospects for 2026. The analysis suggests that risks, opportunities, and narrative focus will concentrate on six core dimensions.

At the macro level, HSBC holds an optimistic view on the performance of the US stock market in 2026, setting a target price of 7,500 points for the S&P 500. Analysts recommend that investors expand AI trades from merely infrastructure suppliers to users and drivers.

Cloud Capacity Constraints and Rising Capital Expenditure

The three major global cloud giants (Amazon, Microsoft, Alphabet) are facing severe order backlogs, but capacity expansion is limited by data center construction cycles. HSBC expects the capacity tightness to persist until 2026.

Against this backdrop, the willingness of tech giants to invest remains strong. Alphabet, Meta, and Amazon have all hinted at "significant" growth in capital expenditure in 2026. HSBC predicts that the aggregated capital expenditure growth rate for related companies will reach 44% in 2026. Most giants, except for Oracle and CoreWeave, are expected to rely on free cash flow for financing. HSBC believes that the current expenditure rhythm is more constrained by "construction capacity" rather than "investment willingness," and there are upward risks to the expenditure guidance for 2026.

Power Supply Becomes the Core Bottleneck for AI Development

The biggest obstacle to capacity expansion is not funding, but electricity. According to HSBC's analysis, the order delivery cycle for heavy gas turbines has been extended to over four years, while small modular reactors (SMRs) in nuclear energy are unlikely to contribute capacity before 2035. This means that the power gap for data centers will continue to exist in the short term.

In contrast, the Chinese market exhibits different characteristics. Domestic brands are rapidly penetrating the large engine market with shorter delivery cycles. Additionally, liquid cooling technology will become a hot topic in 2026, as chip power increases, making liquid cooling a necessity rather than an option.

Significant Growth in Capital Expenditure

Alphabet, Meta, and Amazon have all hinted at "significant" growth in capital expenditure in 2026, and HSBC expects most large tech companies to take similar actions when announcing their 2026 guidance (at the end of January/early February). HSBC's forecast indicates that total capital expenditure for the year will grow by 44%. Overall, it is expected that funding will come from current free cash flow generation, except for Oracle and CoreWeave, which will require additional financing. Given the above capacity constraints and strong demand, the report expects capital expenditures to exceed initial guidance in an opportunistic manner. Therefore, there is an upward risk to the guidance for 2026.

The Rise of ASIC Chips Challenges the Dominance of GPUs

In the AI chip sector, NVIDIA's GPUs remain the preferred choice for cloud service providers, but competition from ASIC chips is intensifying. Alphabet's Ironwood TPU and Amazon's Trainium3 are among the representatives. ASICs are attracting more self-developed investments from large-scale enterprises due to their better cost-performance ratio and cost advantages.

HSBC expects that by 2027, the share of ASICs in capital expenditures of cloud service providers will rise from 2% in 2023 to 13%. Although external chip sales may not make a substantial contribution to financial statements until 2027, discussions about orders and technology agreements in 2026 will impact valuations. NVIDIA's recent licensing agreement with Groq is seen as a positive move to address the trend of reducing inference costs.

Competition Among Frontier Models Tends Towards Oligopoly

The competition among frontier large models is entering a reshuffling period. HSBC believes that high sunk costs will lead to market rationalization, ultimately forming an oligopoly dominated by a few giants, with small specialized participants coexisting. Currently, the intelligence level of open-source/open-weight models is rapidly catching up with closed-source models, with the gap reduced to about three months.

Gemini's intelligence is growing faster than ChatGPT. The focus in 2026 will be on high-end model charging models and the commercialization of AI advertising, with HSBC expecting that by 2030, AI chat advertising will account for 2% of digital advertising spending.

AI Integration and Form Innovation in End Hardware

2026 is seen as a key year for the integration of AI into smartphones, as well as the inaugural year for new hardware to challenge traditional platforms. Apple is expected to significantly enhance the AI capabilities of its hardware between 2026 and 2027, including an upgraded Siri and possibly a foldable screen or a 20th-anniversary edition iPhone.

Smart glasses are viewed as an important carrier for the power of AI large models. Meta's collaboration with EssilorLuxottica is currently in the lead, while Samsung, Google, Apple, and Amazon all have related product plans set to launch in 2026-2027. These devices are expected to partially replace smartphone functions through voice dialogue and contextual information capture