HSBC Global Research continues to express confidence in the global artificial intelligence investment landscape, yet it cautions that the market's next upward move hinges on a pivotal piece of evidence: whether enterprise-level AI monetization can keep pace with the massive capital expenditure programs of major cloud providers such as Meta, Amazon, Google, Microsoft, and Oracle.

The revenue growth trajectory for cutting-edge AI models remains robust. Industry reports indicate that as of July, Anthropic's annualized revenue run-rate had climbed to $65 billion, with OpenAI reaching $40 billion. Despite these impressive figures, enterprise adoption is still in its infancy. Data from Ramp AI reveals that the median enterprise spends merely $12 per employee per month on AI models, while the top 10% of spenders allocate up to $650 monthly—both metrics are expanding rapidly. Corporate budget capacity appears sufficient to support a significant increase in AI spending. If S&P 500 companies were to invest an average of $650 per employee monthly, annual expenditure would total $250 billion, representing just 5% of the index's EBITDA, 8% of wages, and 27% of R&D spending. Currently, AI accounts for only 20% of software budgets, but this is shifting swiftly, with new IT and software firm budgets already dedicating 50% to AI.

Enhancements in model capabilities are expected to boost corporate willingness to pay. According to METR data, the Claude Mythos model achieves an impressive 80% success rate on tasks exceeding three hours in duration, whereas the Claude Opus model, released in May 2025, only reaches similar success rates on tasks lasting about twenty minutes. This progression from assisting with singular prompts to executing complete end-to-end workflows not only broadens the potential market but also strengthens the value proposition that justifies corporate AI expenditures.

The capital investment climate presents a temporal dilemma. Market consensus projects that U.S. hyperscale data center operators will allocate $770 billion to capital expenditure in 2026, rising to $1.1 trillion by 2027; since the beginning of the second-quarter earnings season, these forecasts have been revised upward by $170 billion and $260 billion, respectively. HSBC notes that S&P Global Ratings estimates U.S. hyperscalers will spend over $7 trillion on data centers and AI-related capital projects between 2025 and 2030. Free cash flow has already turned negative and is expected to remain under pressure through the end of 2027, as spending precedes the realization of full revenue opportunities by several years. Hyperscalers are bridging this gap through external financing: debt issuance so far this year has approached $250 billion, alongside more than $1 trillion in non-cancellable lease commitments. Despite these strains, demand indicators remain strong. The remaining performance obligations of Microsoft, Amazon, Alphabet, and Oracle reached approximately $2.35 trillion in the second quarter of 2026, up from $815 billion a year earlier. Even with rapid capacity expansion, rental prices for NVIDIA's B200 and H100 GPUs continue to climb, signaling that available computing power remains scarce. HSBC anticipates that while the return on invested capital for hyperscalers will decline, it will hold at a robust 17% through 2028.

From a global portfolio perspective, HSBC's strategic recommendations favor sectors with the strongest demand and earnings prospects that have not yet seen valuations become excessive. The firm prefers emerging markets, particularly South Korean memory chips, Taiwanese semiconductors and advanced packaging, and mainland China's semiconductor equipment and power infrastructure. These markets continue to exhibit robust demand, constrained capacity, and valuations that are generally more attractive than certain segments of the U.S. AI sector. Within the U.S. market, HSBC ranks semiconductors as the top preference, followed by hyperscalers and then software. The firm points out that semiconductor and cloud infrastructure companies derive direct benefits from increases in training and inference volumes, with their revenues being less dependent on which specific foundational model ultimately prevails. HSBC downplays the recent rotation into software, attributing it more to short-covering dynamics rather than a fundamental improvement in business conditions.