(Disclaimer. We are investment analysts, not AI or technology industry insiders. This material is intended solely to provide a framework for evaluating the evolving AI investment landscape from a public markets perspective. Nothing contained herein should be construed as an investment recommendation. Readers should conduct their own independent research and due diligence. Please refer to the footnote for additional disclosures and disclaimers.)
The first wave of the AI investment cycle has rewarded hardware companies, semiconductor manufacturers, hyperscalers, and data center operators. The next phase will likely favor inference-layer economics, industrial automation, scarce infrastructure owners, and power-constrained hardware ecosystems.
Training is the process of building AI models — teaching them to understand language, generate images, write code, and perform reasoning tasks. It is highly capital intensive, requiring enormous amounts of data and energy. The primary players are concentrated among a small group of companies such as OpenAI, Anthropic, Google, and Meta, while the major beneficiaries have been GPU manufacturers, cloud providers, and data center operators.
Inference begins once a model has been trained. It is the ongoing process of using the model in real-world applications. Every ChatGPT prompt, AI-generated search result, coding assistant request, or enterprise copilot interaction requires inference. While each individual query is relatively inexpensive, inference occurs continuously and at massive scale.
As AI adoption expands across consumers and enterprises, the long-term economics of AI may increasingly shift from training toward inference. That transition could favor companies focused on inference efficiency and infrastructure scalability, including power providers, networking firms, cooling systems, edge AI chips, and businesses that reduce the cost of running AI workloads at scale.
Historically, technology benefited from the economics of bits. Creating a product was difficult, but distributing it to an additional user was nearly free. That was the core advantage of internet and software — a principle often summarized as: bits are cheap, atoms are expensive.
“Bits": digital products and information - software, media, data, algorithms, and online services.
“Atoms”: physical assets - factories, hardware, energy, transportation, raw materials, and infrastructure.
The economics of AI are fundamentally different. Scaling AI depends heavily on physical infrastructure:
In other words, AI turns bits back into atoms.
Each new generation of AI models requires enormous amounts of compute infrastructure. Even as models become more efficient, demand often grows faster than efficiency gains — a dynamic similar to Jevons paradox in energy markets, where improved efficiency leads to higher overall consumption rather than lower usage.
This shift has major implications for investors. AI may ultimately behave less like traditional technology and more like infrastructure — rewarding companies tied to semiconductors, power, data centers, networking, and industrial-scale compute capacity.
“Compute” refers to the underlying infrastructure required to power AI systems — including GPUs, servers, networking equipment, memory, data centers, cooling systems, and electricity.
Training AI models requires massive upfront compute investment, while inference consumes compute continuously. Every AI-generated response, search query, or autonomous workflow requires ongoing processing power, networking, electricity, and data center capacity. As AI adoption scales, compute is becoming a scarce and capital-intensive economic resource.
That is why BlackRock CEO Larry Fink recently argued that compute itself could evolve into a standalone asset class. His view is that AI infrastructure is beginning to resemble energy infrastructure: capacity-constrained, economically monetizable, and strategically valuable.
Trades tied to training have become crowded. The next phase of AI investing will likely focus on inference bottlenecks - the scarce infrastructure and operational constraints required to scale AI economically.
Access to reliable power may become as important as access to GPUs.
The ability to reduce inference cost could become a major competitive differentiator.
As AI directly intersects with labor shortages and productivity constraints, the long-term opportunity may extend into industrial infrastructure and automation.
AI is also increasing the value of scarce enabling assets:
This material is provided for informational and educational purposes only and is intended solely to offer a high-level framework for evaluating the evolving artificial intelligence (“AI”) investment landscape from a public markets perspective. The authors are investment analysts and market participants, not AI researchers, engineers, or technology industry operators. As such, portions of this discussion involve interpretation, synthesis of publicly available information, and forward-looking views that may ultimately prove inaccurate.
The structure, framing, investment themes, and conclusions presented herein are those of the authors. Certain summaries, organizational elements, phrasing assistance, and preliminary content generation were supported in part by generative AI tools, including ChatGPT, and were subsequently reviewed and edited by the authors. While reasonable efforts were made to verify factual accuracy, no representation or warranty is made regarding completeness, accuracy, or reliability.
Nothing contained herein constitutes investment, legal, accounting, tax, or other professional advice, nor should any content be construed as a recommendation to buy, sell, or hold any security, sector, or investment strategy. References to specific companies, industries, technologies, or market themes are illustrative only and do not represent investment recommendations or endorsements.
The AI industry is evolving rapidly and is subject to significant technological, regulatory, competitive, geopolitical, and macroeconomic uncertainty. Forward-looking statements, forecasts, estimates, and thematic views are inherently speculative and are based on assumptions that may change without notice. Actual outcomes may differ materially from those discussed.
Readers should conduct their own independent research and due diligence and consult with their own financial, legal, tax, and investment advisors before making any investment decisions. Past performance is not indicative of future results, and all investments involve risk, including possible loss of principal.
All company references, industry observations, and market commentary are based on publicly available information believed to be reliable as of the date of publication, but accuracy and completeness cannot be guaranteed.