Filenews 1 December 2025
By the Treffis Team
"This time it's different" are the five most dangerous words in the financial sector. They are usually said just before the market crashes, to justify the outrageous valuations. However, when comparing the rise of AI in 2025 to the dot-com bubble in 2000, this expression may hold some theoretical validity. The development of AI does not look like a repeat of the dot-com decline, but it is shaping a whole new financial cycle.
In 2000, we experienced a valuation bubble, where stock prices were disconnected from reality. In 2025, we may face a capacity bubble (where infrastructure spending is not related to current utility). First, let's look at the two companies that are at the center of developments: Cisco in 2000 and Nvidia today. And why Broadcom may be one of the best choices in the field of artificial intelligence.
In 2000, Cisco temporarily became the most valuable company in the world with a market capitalization of more than $500 billion. dollars, with the forecasts at the time not giving a realistic picture of growth. Nvidia is currently writing a different scenario: high profit margins, cash galore and a high valuation relative to its growth momentum.
Let's take a comparative overview of Cisco and Nvidia:
Investors usually make the following mistake: they speculate that Nvidia, because its market capitalization exceeds 4 trillion. dollars, is "expensive" as Cisco was. The data show the opposite.
– P/E (Price/Earnings ratio): 200x for Cisco based on its highest share price in 2000 and its earnings for the year. This contrasts with Nvidia's roughly 38x future earnings.
Growth: Cisco had a CAGR (Compound Annual Growth Rate) of 55% revenue from 1997 to 2000, while Nvidia had 70% in the last 3 years on a significantly larger basis.
Cash flow: Approximately €1.3 billion. per quarter for Cisco, compared to the impressive $25 billion. per quarter for Nvidia.
Customers: Vulnerable dot-com start-ups versus highly liquid tech giants — many of Cisco's customers went bankrupt, while Nvidia is far from doing so.
Cisco was a "bubble" because its stock price was exorbitant. Nvidia can be a "bubble" because demand can be ephemeral.
1. Capability vs. Utility: The Big Mismatch
Therefore, the issue is not the share price, but the usefulness of the product.
We are currently observing a historical discrepancy between capability (the infrastructure developed) and utility (the value derived from it).
Capacity (spending): Microsoft, Google, and Meta collectively spend over $200 billion annually on data centers and AI chips. OpenAI is reportedly committed to spending over $1.2 trillion. dollars in the coming years.
Utility (The Performance): Sequoia estimates that AI needs an annual revenue of $600 billion. dollars to "consolidate". At the moment, OpenAI has revenue of $20 billion. dollars per year.
The reality: The actual revenue from customers who purchase AI services (chatbots, etc.) has a small share of this amount. OpenAI has about 800 million. users per week — just 5% to 10% are subscribers ($20-200/month). Over 90% use its services for free.
2. The financing of expenditure
The second important difference is how this "bubble" is financed. In 2000, the bubble was backed by supplier funding — companies like Cisco provided loans to unprofitable start-ups so they could get Cisco routers. When the start-ups collapsed, the funds were lost.
The "safe" category: Microsoft, Google, and Amazon cover capacity costs with cash flow from operating activities. It is among the most profitable companies in history. If they fail in the field of artificial intelligence, they will suffer losses, but they will not go bankrupt. They are solvent customers.
– The Neo-Clouds: Companies like CoreWeave and Lambda Labs are investing billions of dollars in advanced graphics cards (GPUs) for leasing purposes. They operate with small profit margins, fluctuating demand, and significant leverage. Much of this industry relies on loans secured directly due to GPUs. This carries a risk, as these chips quickly become obsolete with each new generation that emerges, with the previous generation losing value.
The key risk: If lease prices fall due to excess capacity, the value of the collateral decreases. This could lead to forced sales of GPUs to repay lenders, flooding the market with "product" and leading to a further drop in prices. This creates a classic cycle of negative feedback.
Circular Investment Model: In addition, chip manufacturers are enhancing the cycle. Nvidia funnels significant sums into leading AI powerhouses such as OpenAI, who then use those funds to buy Nvidia's chips. Nvidia may invest up to $100 billion. dollars to OpenAI.
3. How the situation could develop
If AI productivity growth fails to meet expectations, companies will cut back on spending, hyperscalers will cut capital expenditures, PC prices will plummet, and leveraged entities like neo-clouds may default, leading to a decline in chip orders as cyclical deals fall apart one after another.
"Thorns" can appear anywhere: a disappointing announcement of financial earnings results from a large silicon or cloud company, a private credit restructuring, or any other crisis that will cause turbulence in the system.
Ultimately;
Is it a "bubble"? Very likely. Is it like the dot-com bubble? No.
The year 2000 was a "garbage bubble": In addition to the large network and software infrastructure companies that still exist today, investors fed money to stocks that did not have viable business models behind them.
2025 is an "expectation bubble": Investors believe they are backing the most important infrastructure project in the history of mankind, but this is before its usefulness and economic feasibility are fully confirmed.
