Citi Redirects Semiconductor Investments Toward Power and Infrastructure Amid AI's Growing Structural Constraints

Semiconductor Industry Shifts: From Compute to Infrastructure

The semiconductor sector is undergoing a significant transformation as the AI-driven market matures. Initially, the focus was on computational power, with investments concentrated in a select group of chip designers and large-scale cloud providers. However, as AI systems expand, the limitations of current infrastructure—specifically in memory, storage, and energy supply—are becoming more pronounced. This evolution signals a move away from the traditional cyclical nature of technology hardware toward a foundational, long-term infrastructure build that will support AI for years to come.

Memory Market: The New Growth Engine

One of the clearest indicators of this shift is the memory segment. AI servers now require far more memory than conventional workloads, driving demand that is increasingly independent of consumer electronics cycles. As a result, memory pricing is becoming less tied to device sales and more reflective of infrastructure needs. High-bandwidth memory (HBM), which works in tandem with advanced GPUs, is scaling rapidly alongside the complexity of AI models. With supply expected to remain constrained, the memory market is forecasted to expand at an annual rate approaching 80%. Similarly, enterprise SSDs used for AI inference are projected to grow by about 50% each year. These trends are already impacting financial results, with DRAM and NAND prices anticipated to climb 70–80% through 2026.

Analog and Power Management: Beneficiaries of Infrastructure Expansion

This transition is creating strong momentum for analog and power management semiconductor companies. Unlike consumer-focused chips, these components are closely linked to capital investments in physical infrastructure. As data centers require more electricity and increasingly intricate system designs, the demand for efficient power solutions rises. This change is shifting the sector’s growth from being driven by product cycles to a steadier, infrastructure-oriented trajectory.

Institutional Strategies Reflect the Shift

Major investment firms are responding to this trend. For example, Citigroup’s recommendation to overweight the Information Technology sector is a direct endorsement of the infrastructure thesis. The sector’s impressive 28% gain so far this year underscores the ongoing flow of capital into these foundational technologies. While the compute segment will continue to experience its own cycles, the broader story—encompassing memory, storage, and power—is setting the stage for more sustainable growth. For investors, this means shifting from a focus solely on AI chips to a more diversified approach within IT, emphasizing companies poised to benefit from ongoing capital expenditures.

Top Investment Picks: What’s Driving Outperformance?

Citigroup’s favored stocks are carefully chosen to align with the infrastructure narrative, each representing a different aspect of the AI ecosystem—from core computation to power management.

• Broadcom and Nvidia: These companies remain at the forefront of AI computing, with data center sales serving as their primary growth driver. Their leadership in networking and GPU technologies positions them to capture the most lucrative segments of infrastructure spending. The institutional consensus is that their scale and quality make them strong long-term investments as capital expenditures in data centers persist.
• Texas Instruments: TI is recognized for its potential to improve profit margins as capital spending stabilizes. As the initial surge in AI server deployment slows, the emphasis is shifting to efficiency and optimization. TI’s increased focus on data center applications positions it to benefit from the growing complexity and power requirements of next-generation infrastructure.
• Monolithic Power Systems: Seen as the purest play on infrastructure growth, Monolithic Power is expected to outpace the industry, driven by robust enterprise data sales. The company’s specialization in power management solutions for servers and enterprise systems directly addresses the escalating energy needs of AI workloads.

Collectively, these selections represent a strategic pivot within the semiconductor space—from cyclical, compute-focused growth to the more stable, foundational layers of the industry. This approach is designed to capture the benefits of sustained capital investment while preparing for a future where efficiency and power management are as vital as processing speed.

Portfolio Strategy and Risk Considerations

The prevailing investment strategy is a balanced, or “barbell,” approach. This combines high-growth, high-volatility leaders in AI compute with more defensive, infrastructure-oriented companies. The goal is to participate in the rapid expansion of AI while mitigating risk through exposure to the steadier, capital-expenditure-driven demand for core components.

• Compute Leaders: Broadcom and Nvidia are favored for their dominant positions and ability to benefit from ongoing data center investments, fueling the sector’s impressive year-to-date rally.
• Analog and Power Management Specialists: Texas Instruments and Monolithic Power Systems offer more stable growth, less dependent on product cycles and more closely tied to the ongoing expansion of AI infrastructure.

This dual approach provides diversified exposure, capturing both the cyclical upside and the structural tailwinds supporting the sector.

However, a key risk is the industry’s heavy reliance on AI as a single growth engine. By mid-2025, generative AI chips are estimated to account for nearly $500 billion in revenue—about half of global chip sales. This concentration means that any slowdown in AI demand could significantly impact the sector’s outlook. Portfolio managers must remain vigilant to this potential vulnerability.

Valuation is another important consideration. While the sector overall is performing strongly, not all stocks are equally overvalued. This divergence allows investors to build more resilient portfolios by overweighting high-conviction growth names and seeking value in infrastructure plays, which tend to offer more consistent growth and are less susceptible to sharp downturns.

Key Catalysts and Monitoring Points

The shift toward semiconductor infrastructure is a forward-looking investment thesis, and its success depends on several near-term indicators.

• Quarterly Earnings: For analog and power management companies, attention will be on data center-related sales and margin performance. Texas Instruments’ ability to maintain pricing power and operational efficiency as capital spending normalizes will be closely watched. Similarly, Monolithic Power Systems must demonstrate accelerating revenue from server and data center markets to justify expectations of industry-leading growth.
• Capital Expenditure Trends: The spending patterns of hyperscalers and telecom providers are critical, as they are the primary customers for infrastructure semiconductors. Any reduction in their investment plans—due to economic factors or a slowdown in AI adoption—would directly affect demand for analog and power management chips.
• Broader Economic Indicators: Trends in industrial demand and global manufacturing activity will also influence the sector. Positive developments support analog companies with diversified exposure, while declines—such as a drop in China’s auto production—could weigh on performance, especially for firms like Texas Instruments with significant industrial and automotive business.

In summary, the infrastructure-driven investment thesis for semiconductors is dynamic and will be tested by real-world data in the coming quarters. The current rotation into foundational technology names reflects a bet on continued data center investment and a structural change in semiconductor demand. Ongoing monitoring of earnings, capital expenditures, and macroeconomic trends will be essential to validate this strategy.