AI Valuation Trends in 2025 and the Forces Shaping Them

Artificial intelligence today sits in a moment that invites comparison to the early commercial internet. When ChatGPT was released in late 2022, it did for AI what Netscape Navigator did for the web in 1994: it turned a powerful but abstract technology into something many people could use directly. Silicon Valley Bank’s State of the Markets: H2 2025 report leans into that analogy and asks how far the comparison can reasonably go, and what the data actually says about scale, concentration, and risk in the current AI cycle.

The report takes the importance of AI as a given. The focus is not on whether AI matters, but on how today’s capital flows and valuation levels relate to underlying revenue and business models, and how that pattern resembles, or differs from, the dot-com era.

A Wave Powered by Infrastructure Spending

One of the clearest themes in SVB’s data is the extraordinary level of spending directed toward AI infrastructure. The report highlights that capital expenditure (CapEx) among major hyperscalers (Amazon, Apple, Google, Meta, and Microsoft) has risen sharply since the release of ChatGPT. Over this period, combined CapEx among these firms has increased by roughly 67%, with a substantial portion flowing directly into NVIDIA’s revenue.

NVIDIA’s growth curve across 2021–2025 sits almost directly on top of the hyperscaler CapEx trend, underscoring how intertwined model development, data center expansion, and chip demand have become. Google and Meta, the two companies making the largest investments in AI infrastructure, have seen their own revenue climb significantly, by roughly 80% since 2021, according to the report.

This level of infrastructure investment is one of the reasons the AI economy today is much larger and more top-heavy than the early internet. As SVB notes, “the money is flowing freely,” and a meaningful share of it is aimed at foundational platforms and compute capacity.

When Venture Capital Becomes Part of the Revenue Stack

Alongside this infrastructure surge, SVB calls out a structural feature of the current ecosystem that deserves attention. The report states that many LLM customers are themselves VC-backed companies. In other words, a portion of the demand driving revenue for foundational AI platforms is funded by the same pool of venture capital that supports their customers.

This creates a kind of circular system. Infrastructure providers report strong growth, but part of that growth is a function of venture capital subsidizing experimentation and deployment at the application layer. If downstream startups later pull back on spending, change their usage patterns, or fail to scale, that revenue can adjust quickly. SVB explicitly notes that this pattern will sound familiar to anyone who remembers segments of the dot-com bubble, where some of the demand for internet services came from other speculative businesses rather than from mature end markets.

The report does not present this circularity as an imminent risk. Instead, it treats it as an important ingredient in understanding how the current AI economy is being built: a real technology shift, partly financed and accelerated by risk capital that may or may not sustain the same pace over time.

The Growing Weight of a Small Number of Companies

The scale of individual AI companies is another point of contrast with earlier cycles. The report compares the cumulative value of the top five US AI unicorns with the cumulative value of all dot-com IPOs. SVB notes that OpenAI, xAI, Anthropic, Databricks, and Scale AI together are now worth more than $500 billion. That combined private value exceeds the total value of all IPOs from the dot-com boom, after adjusting for inflation.

At the same time, the vast majority of AI value creation remains in private markets. That concentration of value in a small set of private companies is one of the reasons the report frames the current environment as “fueling a rocket or inflating a bubble.” A large portion of expectations about AI are effectively bundled into a handful of highly valued names.

Major technology companies such as Microsoft, Meta, Google, Amazon, and NVIDIA are making direct investments into these AI leaders, alongside venture firms and other investors, highlighting the interconnectedness of the ecosystem: hyperscalers fund AI companies, buy their services, and simultaneously sell them compute.

Capital Concentration in Venture Deals

SVB then steps back to look at venture capital more broadly. The report shows that roughly 36% of US VC deals now involve AI companies, yet those deals account for about 58% of total capital deployed. The implication is straightforward: a little more than a third of the deal volume is now responsible for well over half of the capital flow.

This is what the report calls “the new math of concentration.” The venture market has not only tilted toward AI in narrative terms; it has done so financially. Larger rounds and higher valuations for AI startups mean that a significant portion of venture portfolios is now tied to how this particular technology wave plays out.

How Public Markets Are Pricing AI

While most AI value remains in private markets, the report does look at how public markets have responded since the launch of ChatGPT. The report plots Nasdaq revenue multiples (total enterprise value divided by revenue) for two periods: the dot-com era, indexed from 1994 to 2000, and the current GenAI era, indexed from 2022 onwards.

In the late 1990s, the Nasdaq revenue multiple climbed from the low single digits to a peak of roughly 6x revenue, before falling sharply. In the GenAI period, the multiple has moved from around 3x to peaks above 5x. The absolute levels are lower than during the dot-com peak, but the trajectory has a similar rising shape.

SVB uses this chart to suggest that the psychology of the market and the way enthusiasm builds into valuations has familiar contours, even if the starting point, participants, and underlying businesses are quite different.

Interpreting the “Bubble” Question

Many people in markets are already asking: are we in an AI bubble? SVB’s answer is careful. The authors write that the answer is “almost certainly yes” in the narrow sense that parts of the market show signs of valuation inflation, heavy capital concentration, and a degree of circular revenue. But they immediately add that the more important question is whether returns from the largest winners will offset losses from the many companies that will not succeed.

The report also emphasizes how different today’s market structure is from the late 1990s. Startups remain private longer, which concentrates risk among institutional and corporate investors rather than retail holders. Many funds model portfolios on the assumption that a high percentage of companies will fail. Roll-ups, acquihires, and other forms of partial exits already play a role in returning some value to investors and employees when standalone stories do not work.

In that sense, “bubble” in the SVB report is less a forecast of collapse and more a shorthand for a phase where prices in some segments appear ahead of fundamentals, even as the underlying technology continues to advance.

Research Analyst

Simran Mehla

Frequently Asked Questions

Q: What does the SVB 2025 report say about AI valuations?

A: The report finds that AI companies command larger round sizes and higher valuations than non-AI peers across stages, with a significant share of overall venture capital now flowing into AI deals.

Q: Why does the report compare AI to the dot-com era?

A: SVB uses the dot-com period as a reference to show similar patterns in narrative momentum and market psychology, while also highlighting how today’s scale, capital intensity, and private-market structure are different.

Q: How concentrated is value in the AI ecosystem?

A: According to the report, the top five US AI unicorns together exceed $500 billion in value, more than the total value of all dot-com IPOs, and AI deals account for 58% of US venture capital despite being about 36% of deal count.

Q: Why is venture capital considered part of the AI revenue stack?

A: SVB notes that many large language model customers are VC-backed startups, meaning some infrastructure revenue is indirectly funded by venture capital rather than by fully mature end customers.

Sources

1. State of the Markets: H2 2025. Silicon Valley Bank, 2025.