AI dominates venture capital: High capital concentration is rewriting North America's innovation rules.

The Fault Lines Behind the Numbers

PitchBook's latest data shows that total U.S. venture capital deals in 2025 reached $339.4 billion, with AI-related investments accounting for $222 billion, or 65.6%. By comparison, this proportion was only 47.2% in 2024 and a mere 10% a decade ago. Just one company—OpenAI—completed a $40 billion funding round in the first quarter of 2025, an amount nearly equivalent to dozens of times the total AI investment in the entire U.S. venture capital market for the whole of 2015.

Yet the real story lies not in "AI's cash grab," but in how this massive capital is allocated. PitchBook notes that half of all venture capital funds in 2025 went to just 0.05% of deals, with foundation model layers and AI infrastructure companies absorbing the vast majority. The total valuation of U.S. unicorns reached $4.3 trillion, with the top 10 companies alone holding a 51.8% share, primarily contributed by core AI players such as OpenAI, SpaceX, xAI, and Anthropic.

This polarization is creating a "capital isolation effect": a few companies become capital black holes, while the broader application-layer AI, non-AI tech sectors, and even other industries face funding squeeze. For North America's innovation ecosystem, this is no longer merely a market cycle fluctuation but a structural reset.

The "Independent Market" of the Foundation Model Layer

We can no longer view "AI investment" as a unified market. PitchBook's report explicitly states that foundation model investment itself has formed an independent niche market. OpenAI's single $40 billion funding round, the strategic support from Google and Amazon behind Anthropic, and the rapid expansion of xAI—these have transcended the scope of traditional venture capital and are closer to a capital game involving sovereign wealth funds and large tech enterprises.

Behind this "arms race" level of funding is a fundamental shift in business logic: foundation models are no longer early-stage trial-and-error startups but rather the entry points for national competition, industrial control, and next-generation infrastructure. Consequently, large LPs (e.g., pension funds, sovereign funds) and tech giants are willing to accept long-term unprofitable high valuations because what they are investing in is not just future returns but data sovereignty and ecosystem lock-in.

This directly leads to one consequence: the investment ecosystem below the foundation model layer is being distorted. Startups focused on AI applications find themselves not only competing with potential business expansions from foundation model companies on products but also vying for attention with these capital behemoths in the financing market. A Silicon Valley early-stage investor described it privately: "If your company isn't directly training large models or selling servers to large model companies, LPs will ask you why not."

AI Premium: Sustainable Valuation Dividend or Bubble Signal?PitchBook’s data reveals another important trend: the AI label brings a significant valuation premium. In the fintech sector, the median valuation of AI-enabled startups is 41% higher than their non-AI peers; in the early stage, the median valuation of AI fintech companies reaches $134 million, with a premium rate as high as 242%. This phenomenon also exists in verticals such as climate tech and healthcare.

The logic behind this “AI premium” is that the market expects AI to greatly improve efficiency, lower marginal costs, or create new revenue models in these industries. But we must be cautious: can the valuation premium translate into a lasting competitive advantage? From the history of North American industries, whenever a technology transition period brings cross-industry valuation premiums, it is often accompanied by bubbles and reshuffling. For example, during the dot-com bubble of the millennium, the “.com” suffix also brought a similar premium, but eventually only a few companies realized value.

What is more noteworthy is that the AI premium has begun to affect LP asset allocation decisions. Some funds deliberately package the AI concept to inflate their portfolio valuations, causing capital to further tilt toward AI and forming a self-reinforcing cycle. If the commercialization pace of AI falls short of expectations in the future, this valuation gap could trigger a chain reaction, especially among crossover funds that took over late-stage AI projects at high prices.

Infrastructure M&A: Computing Power Becomes the New Oil

While most attention is focused on the model layer, a consolidation around AI infrastructure is quietly taking place. In 2025, a consortium led by BlackRock’s Global Infrastructure Partners, together with NVIDIA, Microsoft, and xAI, acquired data center operator Aligned Data Centers for approximately $40 billion. This deal is not only one of the largest AI infrastructure M&A transactions of the year but also marks a trend: computing power infrastructure is being capitalized, assetized, and monopolized like traditional energy.

For North American supply chains and regional economies, this means that capital-intensive investment is reshaping industrial geography. States with lower electricity costs and ample land supply, such as Texas, Ohio, and Arizona, are becoming new hubs for AI data centers. This layout is not just about computing power; it is closely tied to energy supply and grid transformation, fueling the contrarian growth of AI climate tech—in Q4 2025, transaction volumes in this field reached a historic high.

From an investment perspective, the M&A wave in AI infrastructure indicates that capital is moving upstream along the industrial chain, seeking more stable and predictable returns. Compared to the extremely risky race for foundational models, data centers with long-term contracts, physical assets, and fixed cash flows are more favored by large infrastructure funds. For investors, this may mean that the second half of AI investment will place greater emphasis on “physical assets” and “energy binding.”

Who Will Benefit? Who Will Face Pressure?Beneficiaries: - Foundation Model Oligarchs: Companies like OpenAI and Anthropic have not only secured unprecedented capital reserves but also strengthened their positions through ecosystem lock-in; they will define the operating system layer of the AI era. - Large LPs and Late-Stage Investment Institutions: For example, a16z successfully raised $15 billion for a new fund. Capital is concentrating among a few large managers, creating a "mega-fund" effect where the strong get stronger. - AI Infrastructure Owners: Data centers, energy solutions, and specialized chip suppliers will directly benefit from sustained capital expenditure, especially those with supply relationships with the giants.

• Those Under Pressure:
• Early-Stage VCs and Vertical Startups: Non-AI or general AI application-layer companies face greater difficulty in fundraising, valuation pressure, and many promising entrepreneurs may be forced to rebrand their projects around AI concepts, increasing market noise.
• Traditional Tech Companies: SaaS and consumer internet companies that have failed to integrate AI in a timely manner or lack data advantages will face the risk of customer budgets being "crowded out" by AI.
• Small Investment Funds: Against the backdrop of capital concentration at the top and fund-raising hitting a low not seen since 2018, the living space for small GPs is extremely squeezed, potentially narrowing channels for innovation capital.

Future Trends: From Polarization to Diffusion?

Can the current polarized landscape be sustained? Looking at historical patterns, every technological wave experiences capital concentration, bubble accumulation, bubble bursting, and then diffusion to the application side. AI may be at its most concentrated stage, but there are already signs that capital is slowly seeping into the application layer: in Q4 2025, vertical applications surpassed horizontal platforms in both transaction volume and value for the first time—this could be an early signal.

Another key variable is regulation. North American regulators have not yet responded substantively to the monopolistic tendencies of AI foundation models, but as issues of data security, antitrust, and energy consumption intensify, policy interventions could shift capital flows, forcing some investments toward interpretability, safety, and decentralized applications.

For the North American regional economy, this wave of AI capital is intensifying interstate competition. States that can offer favorable electricity rates, fiber-optic networks, and talent policies will attract more data centers and AI companies, reshaping the economic geography of the United States. Texas, Georgia, and Ohio have already shown ambition, contrasting with the past Silicon Valley–centric model.

Ultimately, when the pendulum of the AI capital cycle begins to swing back, the truly valuable companies will not be those that merely gained a premium through labeling, but those that can deeply integrate AI into vertical industries and build sustainable business models. For investors, this means rediscovering overlooked hidden champions in the application layer amid the noisy AI narrative.