Transformative impact of AI on technology and the economy
Transformative impact of AI on technology and the economy
Mark Schoeffel / 13 January 2026
The T. Rowe Price “Tech Tour” presentation provided an in-depth discussion of insights gathered during the firm’s annual end-of-year technology research trip to Silicon Valley in December 2025. More than 40 portfolio managers and analysts participated in meetings with senior executives from dozens of public and private technology companies, spanning infrastructure providers, semiconductor manufacturers, hyperscale cloud platforms, and emerging artificial intelligence (AI) application developers. The central theme of the discussion was the evolution of AI from a conceptual innovation into a large-scale economic force with material implications for productivity, capital spending, and long-term investment opportunities across the technology ecosystem.
Dom Rizzo framed AI as a productivity shock comparable to, and potentially greater than, the historical impact of electrification. He emphasized that long-term economic growth is driven by three inputs: capital, labor, and productivity. Historically transformative technologies, such as electricity, added approximately one percentage point to annual gross domestic product (GDP) growth over multi-decade periods. In contrast, Rizzo argued that AI, described as “digital intelligence,” may exceed that benchmark, citing recent strong nominal and real GDP growth in the United States as early evidence of accelerating productivity gains. He stressed that while AI adoption is capital-intensive and accompanied by speculative behavior in markets, it represents a durable, multi-decade driver of growth rather than a transient phenomenon.
A recurring point from meetings on the trip was that AI has progressed beyond being perceived as a novelty or experimental tool. As noted by Broadcom’s Chief Executive Officer (CEO), the industry has moved from debating whether AI was a bubble to recognizing it as a “wave.” This distinction was important for the portfolio managers, as productivity cycles historically involve periods of excess speculation, but waves imply sustained adoption and compounding economic impact. Rizzo and Wang both emphasized that their role as active managers is to navigate these cycles responsibly, distinguishing between structural growth and short-term excess.
Tony Wang expanded on the evolution of AI investment from infrastructure build-out toward monetizable applications. In the early phase, much of the economic benefit accrued to a narrow set of companies, particularly graphics processing unit (GPU) manufacturers such as NVIDIA. Over time, however, the “profit pool” has broadened across the supply chain. Bottlenecks have emerged in areas such as energy, memory, networking, advanced packaging, and optical components, creating additional investment opportunities beyond GPUs alone. Wang highlighted the importance of understanding where economic profits accrue at different stages of the adoption “S-curve,” which describes how technologies move from early adoption through rapid scaling to maturity.
A key focus of the discussion was whether AI capital expenditures (capex) by hyperscalers—large cloud service providers such as Amazon Web Services, Microsoft Azure, and Google Cloud—are sustainable. Wang framed this question around return on investment (ROI) at the level of “tokens,” the basic units of computation and output used by large language models (LLMs). As long as the cost per unit of compute continues to decline and AI applications generate incremental revenue or efficiency gains, spending can remain durable. The emergence of “agentic AI,” referring to semi-autonomous or autonomous software agents capable of executing complex decision-making processes, was identified as a potential next inflection point that could further reinforce this reinvestment cycle.
Rizzo contextualized the scale of AI investment with striking market size estimates. He described a transition from a roughly $45 billion AI chip market in 2023 toward an estimated $500 billion by 2028 and potentially $1 trillion by 2030. He argued that funding for this expansion is supported by the accelerating profitability of core businesses at major technology companies, including digital advertising platforms and cloud services, as well as by the rapid revenue growth of AI-native firms such as OpenAI and Anthropic. For example, he noted that global spending on software development, particularly coding labor, is estimated at approximately $3 trillion annually. Even modest productivity gains from AI-enabled coding tools could unlock hundreds of billions of dollars in economic value.
The presenters devoted significant attention to AI “scaling laws,” which describe the empirical relationship between increased computational resources and model performance. In simplified terms, they explained that applying approximately ten times more compute can result in roughly a doubling of model intelligence. Achieving this scale requires massive, system-level infrastructure investments, including GPUs, high-bandwidth memory (HBM), networking equipment, optical interconnects, advanced semiconductor packaging, and power delivery systems. Rizzo illustrated the physical scale of these investments by noting that a single high-end GPU consumes roughly the same amount of electricity as an average U.S. household, implying that million-GPU data center clusters require power equivalent to that of a large metropolitan area.
Wang added that the increasing complexity and physical scale of AI systems has driven a redistribution of economic value across the semiconductor supply chain. Memory prices, particularly for advanced DRAM (dynamic random-access memory), have risen sharply due to demand outpacing supply. Optical components, including electro-absorption modulated lasers (EML lasers), are similarly constrained. Advanced packaging capacity at foundries such as Taiwan Semiconductor Manufacturing Company (TSMC) remains a bottleneck. These constraints have implications not only for enterprise infrastructure but also for consumer electronics, as higher memory requirements driven by on-device AI features may increase costs for products such as smartphones and personal computers.
The discussion also addressed semiconductor capital equipment (“semi-cap”) companies, which manufacture the tools used to produce chips. Wang noted that while AI initially represented a small portion of total wafer demand, conditions have shifted meaningfully. Improved demand visibility, normalization of shipments to China following export controls, renewed investment by manufacturers such as Intel and Samsung, rising memory prices, and continued strength at TSMC have combined to create what he described as an early-to-mid-cycle upturn. Rizzo emphasized the attractive business models of semi-cap companies, which typically exhibit high incremental operating margins and free cash flow generation during periods of rising demand.
Turning to software, Rizzo outlined a framework in which the technology sector is experiencing a multi-year period in which hardware economics dominate software economics. He argued that AI is simultaneously rewriting how software is developed and driving the marginal cost of software creation toward zero. In this environment, traditional application software providers face challenges unless they possess strong data strategies and platforms capable of enabling AI-driven workflows. He cited Palantir as an example of a company positioning itself as an “operating system for AI,” particularly in enterprise and government use cases, though he cautioned that valuation considerations remain important.
Rizzo also highlighted a potential resurgence in central processing units (CPUs) driven by AI agent orchestration. Unlike traditional applications that execute a task once per user input, AI agents may perform thousands of iterations to optimize outcomes, creating significant demand for general-purpose compute. This dynamic could benefit CPU providers such as Advanced Micro Devices (AMD) and Intel as AI adoption broadens beyond model training into real-time inference and enterprise deployment.
On valuation, Wang emphasized the importance of assessing where companies sit on the adoption S-curve. Early-stage companies may appear expensive on traditional metrics but justify valuations through rapid growth and expanding addressable markets. In contrast, more mature software firms tend to trade at lower multiples. The presenters argued that many large-cap technology companies, often referred to as the “Magnificent Seven,” remain reasonably valued on a price-to-earnings growth (PEG) basis, as their earnings growth has kept pace with or exceeded multiple expansion.
In concluding remarks, both portfolio managers expressed confidence that technology, and AI in particular, remains well positioned to lead markets over the long term. While leadership within the sector may evolve—from infrastructure to applications and services—the overall addressable market continues to expand. They emphasized that AI is not a zero-sum competition but a multi-layered ecosystem in which multiple companies can succeed simultaneously. Meetings with firms such as Palantir, OpenAI, Anthropic, and Momentum reinforced their view that AI is both a present-day economic reality and a long-duration structural growth driver.
Disclosure
This summary is based on information presented during a T. Rowe Price investment presentation featuring representatives of T. Rowe Price. The views and statements summarized above are those of the presenters and do not represent the views of Mark Schoeffel or iA Private Wealth Inc. Mark Schoeffel and iA Private Wealth Inc. have no responsibility for the content or accuracy of the information presented. This material is provided for informational purposes only and does not constitute investment advice, a recommendation, or an offer to buy or sell any security. Prospective investors should consult with their financial advisor to determine whether any investment strategy or security discussed is suitable for their individual circumstances and investment objectives.
Comments
Post a Comment