Many of the companies involved would not survive the subsequent crash. The networks they built, however, would continue, to this day, to underpin the global internet.
It was the fibre-optics boom of the late 1990s.
Digital dreams
Between 1996 and 2001, companies such as Global Crossing and WorldCom carpeted land and sea alike in a vast network of fibre-optic cables in anticipation of the dawning digital age. In aggregate, an estimated 100 million miles of optical fibre was laid down across the world.
See also: The great misreading of China: Why exports are a consequence, not a strategy
In hindsight, the central irony of the fibre buildout is that it was not, in the conventional sense, a dotcom fantasy.
What proved fatal was not the thesis itself, but its timing and overreach. So much fibre was laid that it would have reached the sun, with length to spare. Returns, expectedly, proved elusive, and what followed is now well documented.
The bubble pops
See also: Investing where the economics take us: Today, it is America
Between March 2000 and October 2002, the Nasdaq Composite Index plummeted more than 75% from its March 2000 peak. Among the most prominent casualties was Global Crossing.
Founded in 1997 on US$35 million in seed capital, the company went public in 1998 and peaked with a market valuation of just under US$50 billion in 1999. By January 2002, Global Crossing’s transatlantic network dreams had collapsed. The company filed for bankruptcy, crushed under the weight of an overleveraged balance sheet.
Shortly thereafter, WorldCom imploded in what remains one of the largest corporate failures in American history, with investigators later uncovering an US$11 billion accounting fraud.
By late 2002, 23 telecom companies had gone bankrupt, with many more still under severe strain.
The telecom bubble had burst.
The cost of being early
Today, prominent failures of the dotcom era such as online pet store Pets.com are often viewed with derision. Yet, it can be argued that many of them were not bad businesses so much as badly timed ones. We can see their parallels in modern successes. Chewy, for instance, pursued a similar online pet-retail model, and went public in 2019 with a valuation of US$8.8 billion.
For more stories about where money flows, click here for Capital Section
The distinction between folly and foresight is often, exasperatingly, a matter of timing. And infuriatingly, it can be as fatal to be too early as it is to be wrong. As the adage goes, the pioneers take the arrows, settlers take the land.
The same is true of the fibre-optic companies. Vast transcontinental networks were built ahead of demand. Yet, today, the global internet we depend on — underpinning everything from social media to university research networks and, increasingly, AI workloads — rests on precisely this network of fibre infrastructure.
Unused capacity acquired at fire-sale prices became the foundations on which companies such as Google and Microsoft built their empires. Data centres could be sited wherever land and electricity were cheapest, safe in the knowledge that fibre connectivity was already widespread. The hyperscaler emerged from this combination of surplus connectivity and technological innovation.
In other words, had those billions not been sunk into the ground, today’s digital economy might look very different.
The new infrastructure race
Fast forward to the present, the scale of capital flowing in to AI infrastructure is poised to eclipse even the dotcom peak (see Chart 1). This time, it is the hyperscalers that have become the spendthrift lot.
According to Bloomberg estimates, Alphabet, Amazon.com, Meta Platforms, Microsoft and Oracle are set to pour more than US$700 billion into AI-related infrastructure this year alone, with further waves of spending still to come (see Chart 2). Other projections push the total even higher, beyond US$800 billion.
Much of this spending is directed towards constructing the massive data centres required to train and deploy frontier AI models.
In part, the investment spree is driven by intense competition from well-funded upstarts such as Anthropic and OpenAI. The logic is understandable, but also uncomfortably familiar. As in the fibre-optics boom, the prevailing view is simple: Build or be built around.
As a result, Big Tech has rapidly shifted from asset-light to infrastructure-heavy.
Pricing the future
In recent history, tech has predominantly been an asset-light sector. The phrase “software is eating the world”, coined by Silicon Valley venture capitalist Marc Andreessen in 2011, captured the growing dominance of software businesses.
Underpinning that dominance were attractive economics: low capital intensity, near-zero marginal costs and the ability to scale rapidly. Accordingly, investors rewarded those qualities with ever-higher valuations. Big Tech only got bigger.
AI, however, is now reversing this equation.
Much like the fibre builders of the 1990s, today’s tech companies are making large-scale bets on future demand. Hundreds of billions are being committed to infrastructure whose ultimate economic returns remain uncertain.
They may yet prove to be right. The long-term significance of AI is becoming increasingly difficult to dispute.
But as the telecom boom demonstrated, being right about the future does not necessarily mean capturing the profits from it.
Builder’s dilemma
The subsequent history of the telecom industry is instructive. By 2010, what had once been speculative capacity had finally begun to find its demand. Services such as Netflix and YouTube were beginning to scale rapidly. The iPhone, launched in 2007, ushered in the era of mobile internet traffic. The underutilised fibre networks of the prior decade became the superhighways for a new generation of data-intensive digital services.
Even so, telecom operators remained trapped at low valuations, constrained by the capital-intensive nature of the industry. Repeated rounds of capital expenditure on maintenance and network upgrades persistently weighed on returns.
Such is the typical fate of utility and utility-like businesses: They must bear the cost of massive upfront investment while facing uncertainty over how quickly demand will arrive, and whether returns will ever justify the capital deployed.
In the AI era, these questions remain unresolved.
The broader lesson
The lesson from the fibre-optic mania is not that transformative technologies fail. It is that being right about the future does not guarantee investment success.
Such technologies often require capacity to be built well ahead of demand. Case in point: The rails were laid out before commerce moved west; fibre was installed before data could flow; and, today, data centres are being built to support AI at scale. Yet, as both the railway and fibre eras suggest, the builders do not always become the beneficiaries.
The future often arrives on infrastructure that bankrupted its pioneers.
Who takes the land?
At present, AI’s economic landscape remains undefined. Having said that, it is tempting to assume winner-takes-all dynamics, in which one or two platforms capture most of the value, just as Google did in search and Amazon in e-commerce.
Perhaps.
It is precisely this belief that scale today translates into dominance tomorrow that is driving the colossal wave of capital expenditure now underway.
Yet, the fibre builders of the 1990s held a near-identical conviction, and they were right about everything except what mattered most: who would ultimately profit.
The danger for today’s AI pioneers, therefore, is that building the future may still leave them taking the arrows.
Only time will tell.
