As markets debate whether artificial intelligence represents an expensive "tech bubble," Bank of America Securities asserts in its latest report that AI-related spending is not only here to stay but will be "stronger and longer-lasting." In this report, NVIDIA (NVDA.US), Advanced Micro Devices (AMD.US), and Broadcom (AVGO.US) continue to be listed as the semiconductor sector's "preferred picks," with Micron Technology (MU.US) and Marvell Technology (MRVL.US) also included on the core list.
AI Spending to Be Not Only "Stronger" But Also "Longer" Based on the capital expenditure outlooks presented in recent first-quarter earnings reports by major hyperscale cloud providers such as Microsoft (MSFT.US), Amazon.com (AMZN.US), Alphabet (GOOGL.US), and Meta Platforms, Inc. (META.US), BofA anticipates that the total addressable market (TAM) for AI data center systems will surge to a staggering $1.7 trillion by 2030, up from a previous forecast of $1 trillion. Of this $1.7 trillion TAM, approximately $1.2 trillion is expected to come from AI accelerators, an increase from the prior $1 trillion estimate. This upward revision is attributed to increased shipments of custom application-specific integrated circuits (ASICs) by hyperscalers, such as Google's Tensor Processing Units (TPUs) or Amazon's Trainium chips. The data center CPU market TAM could reach around $110 billion (up from a previous forecast of $80 billion), while the AI networking market TAM may grow to approximately $316 billion (up from an earlier projection of about $240 billion).
Analysts explained: "Importantly, we note that compute/memory components continue to diversify as AI workload tail effects lengthen. However, we view this as additive to the overall TAM, not a replacement for existing workloads or components. For example, we expect CPUs (in their new standalone racks) to work alongside existing GPU-CPU compute racks, while SRAM-based ultra-low latency memory racks will coexist with HBM-based GPU racks."
From "Selling Shovels" to "Mining Gold": 2026 Seen as ROI Inflection Point Over the past two years, Wall Street's investment thesis in AI has largely focused on "infrastructure racing"—tech giants frantically ordering NVIDIA's GPUs to avoid falling behind. However, this "burn cash for the future" model has consistently faced questions about profitability prospects. BofA analysts believe that as AI unicorns like OpenAI and Anthropic approach initial public offerings (IPOs), the market will gain a clear view of how these "top-tier model producers" translate compute power into tangible commercial value. Once these startup giants demonstrate their profitable closed loops in the secondary market, the trillion-dollar expenditures from Microsoft, Amazon.com, Alphabet, and Meta Platforms, Inc. will have a rational and clear financial logic. It will no longer be blindly "buying shovels," but rather reaching the edge of "mining gold."
Analysts particularly emphasized that 2026 will be a key inflection point for the acceleration of AI sales and return on investment (ROI) realization, while 2027 will need to verify whether "token economic efficiency" (i.e., the decline in cost per unit of compute) can be achieved as expected.
Despite the bullish outlook, BofA retains a degree of caution. The report acknowledges that current supply chain bottlenecks may limit near-term shipments of cutting-edge components, and the sustainability of capital expenditures remains a persistent market concern.
Finally, BofA also raised its price targets for the following companies in the report: NVIDIA (from $300 to $320), Advanced Micro Devices (from $450 to $500), Marvell Technology (from $125 to $200), and Micron Technology (from $500 to $950). The price target for Broadcom was maintained.
A Dose of Reality for Investors While BofA's bullish stance is encouraging, rational investors should remain aware of several potential risks: First, the "IPO illusion" risk. If the public market debuts of OpenAI or Anthropic underperform, leading to a burst in private market valuation bubbles, it could trigger a chain reaction, shaking the investment confidence of cloud providers. Second, the "efficiency paradox." If algorithm optimizations (such as model compression, inference acceleration) significantly reduce the compute power required per AI task, hardware demand growth may fail to keep pace with market expectations. Finally, macroeconomic "black swan" events. The $1.7 trillion TAM forecast is based on the current relatively accommodative interest rate environment. If the Federal Reserve maintains high rates for an extended period, increasing the capital costs for cloud providers, they may be forced to cut long-term spending plans.
Nevertheless, BofA's report conveys a clear signal: AI infrastructure investment is not a fleeting "event-driven" phenomenon but a "structural bull market" that will last at least until 2030. In this feast, NVIDIA remains the "shovel seller," but Advanced Micro Devices, Broadcom, Marvell Technology, and Micron Technology are finding their own gold from different corners of the mine.
Comments