What A.I. Investors Can Learn From The Ericsson Crash In 2001

Dear investor,

The Ericsson supercycle was rooted in the global adoption of the Global System for Mobile Communications (GSM). By the mid-1990s, GSM had transitioned from a nascent European standard into a de facto world standard. Ericsson, leveraging its early lead in software-controlled switching and radio technology, captured a 40% share of the global mobile market by 1997. In 1998, the number of GSM subscribers totaled 100 million, but by the end of 2000, this figure had exploded to 465 million, a number that notably exceeded the 399 million internet users estimated at the time.

The narrative sustaining this growth was built on the premise of “Mobile Miracles.” Ericsson’s internal research and technology reviews from the early 1990s had already begun to forecast a world where mobile devices would become the primary interface for human interaction. This vision prompted a massive internal shift toward third-generation (3G) technology. Even as early as 1992, at the World Administrative Radio Conference (WARC), spectrum was being allocated for 3G on the assumption that traffic would soon be dominated by internet access rather than simple voice calls.

A common misconception regarding the dot-com era is that it was comprised entirely of profitless internet startups. Ericsson, however, was a profitable company at its peak. The year 2000 represented a record-breaking performance. The company’s success was not merely a product of stock market manipulation but was driven by a genuine, albeit unsustainable, surge in infrastructure orders from global telecom operators who were racing to build the backbone of the mobile internet.

By the end of 2000, Ericsson had increased its turnover by SEK 58 billion in a single year, with net profits reaching a historic high of SEK 21 billion. The company was reinvesting a massive SEK 42 billion into research and development, attempting to maintain its lead in the 3G race. This profitability provided the financial justification for the New Economy narrative, as analysts argued that the traditional rules of valuation no longer applied to a company that appeared to be the sole arms dealer for the digital revolution.

In Sweden, the Ericsson stock became a national phenomenon. By March 2000, the Stockholm General Index reached historic highs, with the total stock value on the exchange reaching twice the value of Sweden’s entire GDP. Ericsson was the most frequently traded stock on the OM exchange, with a market capitalization reaching SEK 1,800 billion at its peak. The hysteria was such that turnover on the exchange rose by 107% in the first two months of 2000 alone.

The mass media and financial analysts played a critical role in sustaining this fever. Even as the bubble began to show signs of instability in the spring of 2000, major Swedish publications like Dagens Nyheter and Svenska Dagbladet urged calm, asserting that a significant financial downturn was improbable. This collective optimism was reinforced by government policy; in April 2000, the Swedish Riksdag allowed general pension funds to be invested in equities, essentially directing the retirement savings of the populace into the market just two days before the New York stock market began its historic collapse.

The narrative of the late 90s was also defined by a fierce technological competition over 3G standards. While 2G had been fragmented between GSM in Europe and various analog and digital systems in the US and Japan, 3G was envisioned as a common global standard. However, this framework eventually split into competing tracks: Wideband Code Division Multiple Access (WCDMA), championed by Ericsson and Nokia for the European and Asian markets, and CDMA2000, favored by US-based firms and companies like Qualcomm.

Ericsson’s strategy was predicated on WCDMA becoming the dominant global protocol. The company invested heavily in proving the viability of this technology, conducting the first 3G call for Vodafone in the UK in 2001. The perceived inevitability of WCDMA adoption drove much of the infrastructure ordering in 1999 and 2000, as operators feared being left behind in a world where data speeds were expected to jump from the 9.6kbps of early GSM to the 2Mbps promised by 3G.

The cause of the sentiment shift and the following market crash was not a failure of Ericsson’s technology, but a failure in the capital allocation of its customers, the mobile network operators. Seeking to capitalize on the 3G hype, European governments moved away from traditional way (where licenses were awarded based on technical merit and social benefit) toward competitive auctions designed to maximize state revenue.

The UK’s 3G auction in March 2000 served as the canary in the coal mine. After 150 rounds of bidding over 36 days, five operators spent a combined £22.5 billion for licenses, an order of magnitude higher than the official pre-auction forecast of £1-3 billion. In Germany, the auctions raised even more in absolute terms. Collectively, European operators paid just over 130 billion Euro (SEK 1,200 billion) for the right to build 3G networks.

This created a perfect storm of financial distress. To win these licenses, operators like BT and Vodafone were forced to bid up to their entire enterprise value, fearing that without 3G spectrum, they would be dead in the water. This massive debt-funded expenditure left the operators’ coffers empty. They possessed the legal right to build the next generation of networks but had no remaining capital to purchase the actual base stations, routers, and software from vendors like Ericsson.

In the fall of 2000, Ericsson’s CFO, Sten Fornell, was still downplaying the market decline as temporary. However, the reality of the operators’ financial exhaustion finally broke through in early 2001.

On March 12, 2001, Ericsson issued a profit warning. The company reduced its first-quarter growth forecast from 15% to zero. The drop in orders was unprecedented; in February 2001, major customers like AT&T began canceling written orders for mobile systems, a phenomenon management had never previously encountered. Within a single day, Ericsson’s share price plummeted by 21.5%.

This warning signaled the end of the “New Economy” delusion. The market suddenly realized that the 3G license costs were not merely “sunk costs” as economists had theorized, but were a direct drain on the future capital expenditure required to sustain the equipment vendors. From mid-2001 to mid-2002, investments by telecom operators swung from an annual increase of 25% to a cutback of 25%, effectively halving the purchases from Ericsson and its competitors.

Between 2001 and 2005, Ericsson reduced its global workforce from 107,000 to 47,000.

Comparing the Infrastructure Cycles: 1990s Telecom vs. 2020s AI

In 2025 and 2026, the global technology industry is undergoing what analysts describe as an investment supercycle. The top four hyperscalers, Amazon, Google, Microsoft, and Meta, are expected to spend more than $350 billion on capital expenditures in 2025 alone, representing a mid-30% year-over-year increase. By 2026, the top five hyperscalers (including Apple) are projected to deploy over $600 billion in data center infrastructure.

The scale of the AI build-out is comparable to the telecom peak in terms of its contribution to GDP, but the underlying health of the balance sheets is radically different. While the 1990s boom was led by highly leveraged operators, today’s cycle is driven by the world’s most cash-rich companies.

The most critical distinction lies in how the infrastructure is being paid for. The 3G auctions and the network builds were funded by a massive expansion of corporate debt. Companies like Vodafone and BT took on staggering debt loads to buy spectrum, leaving them vulnerable when the market corrected. In contrast, the current AI capex is being funded almost entirely from existing operating earnings. In 2024, companies like Microsoft and Meta demonstrated that their growing revenue from existing advertising and cloud businesses could more than fund their surging data center spending.

Furthermore, the vendor financing that plagued the 1990s, where companies like Nortel and Ericsson essentially lent their customers the money to buy their gear, is largely absent in the current AI cycle. While there are examples of circular financing (such as Microsoft investing in OpenAI, which then pays for Azure services), the systemic risk is mitigated by the massive free cash flows of the anchor tenants.

One of the most damning indictments of the 1990s boom was the lit fiber statistic. By mid-2001, telecom companies had installed 39 million miles of fiber, but only 10% was active, and each active fiber was utilizing only 10% of its available capacity. This meant that 99% of the investment was effectively sitting idle.

In the AI era, the opposite is true. The demand for compute power and GPUs is so extreme that vacancy rates in primary US data center markets have fallen to a record low of 1.6%. In hotspots like Northern Virginia, vacancy is effectively zero. Every unit of compute being deployed is currently active, driven by the enormous training workloads required for frontier language models.

Despite the stronger financial foundation of the AI cycle, there is a significant parallel to the 3G era in the monetization gap. The fundamental reason for the Ericsson crash was that while the infrastructure for the mobile internet was being built in 2000, the applications that made it indispensable (like the iPhone, mobile video, and app stores) did not arrive at scale until 2007–2010.

3G technology was capable of video calling and internet browsing as early as 2001 in Japan, but the “Mobile Miracle” did not materialize immediately because devices were expensive, battery life was poor, and the networks were initially unreliable. This led to a multi-year “air pocket” where operators had built expensive networks that were utilized only by a tiny fraction of their subscriber base.

Similarly, in 2025 and 2026, the focus is on building massive training clusters. The risk for hyperscalers is that the current focus on training frontier models might not immediately translate into consumer or enterprise revenues. CFOs currently report difficulty in quantifying the ROI of AI, despite it being a mission-critical priority. If generative AI initiatives do not demonstrate tangible productivity improvements or revenue enhancement within a 12-18 month window, the market may face a significant correction similar to the 2001 sentiment shift that punished Ericsson.

In the 1990s, the primary bottleneck for the telecom industry was spectrum, the finite “air” required to transmit data. In the 2020s, the new primary constraint is power. AI data centers consume as much electricity as 100,000 homes, and the largest new sites are expected to consume twenty times that amount.

Power scarcity has shifted from a temporary inconvenience to a primary constraint on hyperscale growth. Utilities across the US and Europe are revising their load forecasts upward, and grid interconnection timelines are stretching into the late 2020s. This effectively caps near-term expansion, regardless of how much capital companies are willing to spend. This power imperative creates a different kind of risk than the 3G era; rather than over-building a network that no one uses, companies may find themselves with stranded assets, data centers that have the hardware but lack the electricity to run it.

Conclusions and Lessons for the Modern Era

The rise and fall of Ericsson remains the definitive cautionary tale for infrastructure-led technology cycles. The company’s trajectory from a record-breaking SEK 21 billion profit in 2000 to a SEK 21 billion loss in 2001 illustrates how quickly market sentiment can turn when the capital allocation of an entire industry is misaligned with the economic reality of its customers.

Key Takeaways

• Ericsson was at its most profitable when its stock peaked. High net income can mask a looming systemic collapse if that income is driven by debt-funded customer spending that is about to evaporate.

• The “Mobile Internet” narrative was factually correct, the technology eventually changed the world, but the timing of the “Killer Apps” was off by nearly a decade. Investors who focused on the vision rather than the cash flow of the operators were the ones who suffered the most.

• The fundamental reason the AI bubble is likely to behave differently than the 3G bubble is the source of capital. Funding infrastructure via free cash flow (the hyperscalers) is structurally more resilient than funding via spectrum auctions and junk bonds (the 2000s operators).

The Ericsson story suggests that the technology evangelists are often right about the long-term destination but wrong about the speed of arrival. The mobile miracle did happen, but it required the near-collapse of the industry’s largest player to reset the cost structure and make it sustainable. As the AI infrastructure build-out reaches its peak in 2026, the question is not whether AI will transform the world, but whether the current titans of industry have the discipline to wait for the killer applications to arrive before their capital reserves, or the market’s patience, are exhausted.

History shows that being the “arms dealer” for a digital revolution is only profitable as long as your customers have the capital to keep buying the weapons. If the “AI Miracle” takes as long to materialize as the “Mobile Miracle” did in 2000, even the most cash-rich hyperscalers may eventually face a sentiment shift that punishes their valuation.

But what if you didn’t have to guess when the AI cycle would peak? What if you could invest in a tech company that has spent the last 3 decades building a fortress that doesn’t require a single cent of AI investment to dominate?

Here are the facts:

• Zero debt

• Consistent free cash flow margins around 30%

• Less than 5% of its cash flow is used for capital expenditures

• Digital infrastructure player that’s essential for its niche

• Dividend of around 4%

• Trading at a free cash flow yield of around 5.2%

• No competition

• Founder-led for more than 3 decades

• Small-cap

👉 Download the report 👈 and see if this matches what you look for in a long-term holding.

Until the next issue.

If you enjoy The Onveston Letter, let me and the algorithm know by clicking the “Like” button ❤️.

And if you aren’t a subscriber yet, then sign up below to not miss out on future articles.

For new readers: Check out all my previous posts here

Disclaimer: This analysis is not advice to buy or sell this or any stock; it is just pointing out an objective observation of unique patterns that developed from my research. Nothing herein should be construed as an offer to buy or sell securities or to give individual investment advice.

Thanks for reading! This post is public so feel free to share it.