Bank of America Predicts AI Regulation Will Boost Nokia and STMicroelectronics as Spending Shifts

Bank of America Challenges Narrow Market Assumptions on AI Regulation

Bank of America strategists are pushing back against the market's conventional view that tighter AI regulation will simply slow frontier-model development and weaken semiconductor demand1

1

. Instead, the bank argues that AI regulation could redirect spending toward a broader set of hardware categories, fundamentally reshaping the semiconductor industry cycle2

2

. This shift matters because it challenges investor assumptions about which companies will capture value as AI governance frameworks tighten globally. Rather than concentrating capital in advanced GPU development, regulation may distribute spending across inference, monitoring, and control infrastructure, creating opportunities for European IT stocks that have been overshadowed by hyperscale AI training investments.

How Regulation Reshapes AI Infrastructure Spending

Regulation may reduce the frequency of large frontier training runs, but production AI agents will require monitoring, verification, containment, and auditability across prompts, outputs, and tool calls1

1

. Bank of America cited Anthropic research showing that using Claude 3.5 Haiku as a safety filter for Claude 3.5 Sonnet increased inference costs by approximately 25%, though this figure is architecture-specific and representation reuse could reduce overhead substantially1

1

. OpenAI's guardrail framework spans input, output, and tool controls, with tool checks potentially running before and after every function call, suggesting that policing workloads could scale with agent complexity rather than simply user-query volumes1

1

. This regulatory architecture creates sustained demand for distributed infrastructure rather than episodic spending on massive training clusters.

Nokia and STMicroelectronics Positioned as European Winners

Bank of America named Nokia and STMicroelectronics as potential beneficiaries of this spending shift1

1

2

2

. Nokia is viewed as a potential European beneficiary through datacenter switching and optical transport, with the bank highlighting its expanded portfolio of high-speed switches featuring 1.6Tb Ethernet interfaces designed for AI training and inference workloads1

1

. STMicroelectronics was cited for its optical interconnect exposure1

1

. Bank of America maintains buy ratings on both Nokia and STMicroelectronics, with price targets of 16 and 85 respectively2

2

. Notably, Ericsson was not mentioned in the analysis, with BofA previously assigning the stock an underperform rating2

2

.

Custom ASICs and CPUs Gain Ground in Regulated AI Workloads

The shift toward regulated AI agents could favor custom ASICs, CPUs, networking switches, optical interconnect, and memory1

1

. CPUs should see increased content because agentic inference requires scheduling, routing, policy logic, memory management, and safety checks1

1

. Networking intensity should rise as agents communicate with models, retrieval stores, tools, security services, and other agents1

1

. Secure inter-datacenter transport may become more valuable as regulated workloads require privacy, resilience, and encryption across distributed infrastructure1

1

. This represents a fundamental architectural shift from concentrated frontier training toward distributed inference and control systems.

Semiconductor Equipment Companies Remain Relatively Neutral

Bank of America views semiconductor equipment companies as largely agnostic to this potential spending shift, though the bank noted a possible longer-term benefit if custom ASICs and connectivity silicon increasingly adopt leading-edge manufacturing nodes1

1

. This assessment suggests that equipment makers may capture value regardless of whether spending concentrates in training infrastructure or distributes across inference and monitoring systems. However, a migration toward advanced process nodes for connectivity silicon could eventually drive incremental demand for cutting-edge lithography and deposition tools.

Key Risks That Could Derail This Investment Thesis

Bank of America flagged two key risks to its thesis: embedded guardrails could end up minimizing incremental compute demand, or regulation could slow AI deployment more than the expansion of policing infrastructure offsets it1

1

. If safety mechanisms become sufficiently efficient through architectural optimization, the anticipated infrastructure buildout may not materialize at scale. Alternatively, overly restrictive regulation could dampen AI adoption rates faster than compliance infrastructure expands, creating a net negative for the semiconductor industry cycle. Investors should monitor regulatory implementation timelines and efficiency improvements in safety filtering architectures to assess whether this spending shift materializes as Bank of America anticipates.