In recent weeks, investors have seen a pickup in large partnership announcements across artificial intelligence (AI) model developers, hyperscalers and chip companies. These deals are expected to deploy capital over a number of years, with contingencies tied to the continued execution and leadership of the firms. Yet they also signal urgency—a race to meet exploding demand for compute. AI leaders are coordinating across the value chain in an attempt to ensure that supply keeps up with the speed of innovation.

Still, the scale and circular nature of these commitments, where suppliers, customers and investors overlap, has prompted comparisons to the late-1990’s tech bubble. While some caution is warranted, we think the better question is not whether today's deals resemble the dot-com era, but whether the underlying fundamentals do. 

We identify three key distinctions:

1. Robust balance sheets

In the 1990s, much of the buildout was financed by companies with limited profitability and heavy reliance on external capital. By contrast, today's wave is largely funded from the hyperscalers’ own free cash flow and robust margins. Considering that many past bubbles burst amidst tightening credit conditions, this buildout appears more resilient to that kind of stress.

 Many have also drawn comparisons to the vendor-financing loops of the ‘90s, where telecom infrastructure companies financed each other to inflate growth¹. However, today’s deals look different. Capital is chasing AI, not the other way around—AI has captured 1 in every 2 VC dollars this year² — and spending is anchored in physical infrastructure like chips, electrical equipment and data centers.

2. AI revenue momentum

Whereas early internet firms built first and monetized later, AI is monetizing as it builds. Hyperscalers are already seeing returns through increased cloud demand and productivity gains in coding, advertising and enterprise tools. The model developers have more nascent business models, but with a 99% U.S. market share in global LLMs³, revenues are scaling. Meanwhile, enterprise adoption is gaining traction. KPMG’s latest AI Survey shows average enterprise AI investment rising 14% from 1Q to USD 130M, supported by visible productivity and profitability gains from AI use cases⁴. 

3. Demand outpacing supply

Any wave of heavy capital investment runs the risk of overbuilding. At the peak of the dot-com era, only about 7% of the fiber-optic network was being utilized, leaving vast excess capacity that took years to absorb. But today, data center vacancy rates are at record lows⁵ and utilization levels hover around 80%. Demand for compute continues to far outpace supply—more data has been created in the last three years than in all of history⁶, and AI workloads are growing by the magnitude. 

Lessons from history

Still, we don’t think caution around AI is unwarranted. The scale of spending is enormous, the pace unprecedented, and some assumptions around ROI, such as the useful lives of assets, remain open questions. History reminds us that enthusiasm can run ahead of reality. Yet so far, today’s players are far better capitalized than those of the dot-com era, AI monetization is underway, and the risk of overbuilding seems limited in the near term. As this story unfolds, investors should focus on selectivity, leaning into active management to separate transformative winners from bubbly valuations. 

¹In the late 1990s, many telecom equipment makers extended loans or equity stakes to their own customers to finance network expansion. The practice boosted reported sales and demand on both sides, but when credit tightened, those circular flows unraveled, contributing to the sector’s collapse and broader dot-com crash.

²Source: Pitchbook, as of August 31, 2025.

³Source: Statcounter.

⁴Source: KPMG AI Quarterly Pulse Survey, September 2025.

⁵According to CBRE, the vacancy rate in North America's data center markets has reached a record low of 1.6% in 1H 2025.

⁶Source: IDC, as of 5/31/2024.

09pw252210033733