By Boo Kok Chuon and Clarie Chan Mei Lee

A few days ago, over lunch, I found myself sharing a family story with my soul mate and my co-author for this article, Clarie. Beyond being my wife, she is also my fellow classmate in NTU when we did our Chartered Valuer and Appraiser programme previously. She is also my most trusted sounding board whenever we take on valuation assignments.

That afternoon, our conversation turned to whether the assumptions we routinely adopt in valuation work are still relevant in a world increasingly shaped by artificial intelligence. It was in that context that I revisited a family history that had quietly shaped my own entry into finance: a story I had previously recounted in an interview with The Straits Times in 2015, but one that felt newly relevant.

I shared with Clarie the story that began long before I became a professional, during the dot-com era.

I was thirteen at the time, when technology stocks dominated market sentiment. My father, like many investors then, had exposure to companies such as IPC Corporation, Creative Technology, and Chartered Semiconductor. These were not speculative fringe bets. They were credible, familiar names, closely associated with the technological future that was visibly unfolding, including in our own household, which I still remember being a proud owner of an IPC desktop personal computer running Windows 95.

I remember that day, we had just returned from a family cruise. It was meant to be an ordinary, happy family day when my father’s broker called.

The market had crashed. There was a margin call. The question was whether he wanted to cut his losses.

As I had been commercially groomed by my parents since I was nine, I understood clearly what had transpired: margin financing, forced liquidation, and the implications of holding positions as markets turned decisively against you.

I still remember my father’s response. Calm. Measured. Grounded in the valuation logic of the time.

He told them to hold.

That instruction would ultimately cost my family approximately S$300,000 of our family’s savings. This is not a small sum in the late 1990s and early 2000s, especially in the context of our humble middle-class family. It was capital that did not disappear in a single dramatic collapse, but steadily evaporated as technology stocks were re-rated and the assumptions underpinning their valuations failed.

I told Clarie I was not recounting this story out of nostalgia. I was recounting it because, as valuation practitioners, we were now asking a disturbingly similar question over lunch: are the assumptions we rely on today’s age of Artificial Intelligence still valid, or are we once again pricing the future using a map drawn for the past?

From that conversation, we began examining whether the structural conditions that led to valuation failure during the dot-com era may be re-emerging today, albeit through a different mechanism.

Dot-Comera vs AI era

The dot-com era is often remembered as a period of excess and irrational speculation. That framing is convenient, but it misses the more instructive failure. The damage did not arise simply because investors believed in technology. That belief was, in many respects, correct. The damage arose because capital continued to deploy using valuation models whose assumptions no longer matched the economic reality being created by that technology.

In the late 1990s, technology fundamentally altered distribution, scale, and cost structures faster than valuation frameworks adapted. Companies that appeared credible, productive, and aligned with the future were priced as though early participation in a technological wave guaranteed durable value capture. When that assumption broke, the correction manifested not as philosophical insight, but as impairment, devastatingly, quietly, and permanently.

What makes the present moment unsettling is not the presence of enthusiasm around artificial intelligence, but the familiarity of the pattern. Once again, a general-purpose technology is collapsing long-standing constraints. Once again, valuation models continue to price assets as though those constraints remain intact. And once again, capital is being committed under assumptions that feel reasonable until they no longer are.

The concern, therefore, is not whether AI will transform industries. The concern is whether we are repeating a structural error: using yesterday’s valuation logic to price participation in tomorrow’s economic reality.

How Traditional Startup Valuation Works — and Why Its Assumptions Are Fraying

Traditional startup valuation is often presented as a collection of methods: scorecards, comparables, milestone pricing, early discounted cash flow.

In practice, these approaches share a common intuition. They attempt to estimate future value not by precise forecasting, but by assessing whether a startup appears capable of executing better, faster, or earlier than others.

At the early stages, when revenue is limited or nonexistent, valuation becomes a proxy for confidence. Investors look to the founding team’s background, the speed at which a product has been built, early user adoption, and the apparent momentum of the business. These signals are taken as evidence that the startup can navigate uncertainty more effectively than its peers and convert opportunity into durable value.

This logic emerged in a world where replication was genuinely difficult. Building software used to require specialised talent, long development cycles, and significant capital. Iteration was slow, and mistakes were expensive. Under those conditions, a team that could ship early, attract users, and learn faster than competitors deserved a valuation premium. Time, talent, and speed were scarce, and scarcity justified price.

The difficulty today is that these conditions are changing faster than valuation logic has adjusted, especially with the rise of no-code AI development platforms such as Base44. Artificial intelligence has dramatically reduced the cost of building, testing, and iterating products. Tasks that once required large teams and months of effort can now be accomplished by small teams in weeks, sometimes days. Execution capability, once a key differentiator, is becoming increasingly widespread.

As a result, many of the signals that traditional valuation relies upon have lost some of their meaning. A polished product no longer implies exceptional execution skill. Rapid iteration no longer guarantees a lasting lead. Early traction, while still valuable, is easier to replicate when competitors face fewer technical and operational barriers. What once indicated scarcity may now reflect temporary accessibility.

The problem is not that traditional startup valuation is irrational or poorly constructed. It is that it continues to price assumptions that are quietly eroding. When execution speed, team size, and early momentum become less scarce, their ability to predict long-term value weakens. Valuations built on these signals may still feel reasonable, until time reveals that the advantages being priced were not as durable as assumed.

This is where the relevance question arises. If valuation models continue to reward characteristics that are becoming increasingly common, they risk overstating differentiation and understating how quickly competitive parity can emerge. In such an environment, capital is not misled by hype, but by familiar signals whose economic meaning has shifted.

That shift, subtle as it may appear, is precisely how valuation models become obsolete.

Why the Dot-Com Parallel Still Matters Under Modern Accounting Frameworks

A common objection to comparing the current AI cycle with the dot-com era is that technology companies today do not become publicly listed as quickly as they did in the late 1990s. That observation is correct, but it misunderstands where valuation risk now resides. The risk has not disappeared; it has migrated.

In the dot-com period, technology companies entered public markets early, often before their economics were proven. When valuation assumptions failed, the correction was immediate and visible. Share prices collapsed, and losses were borne directly by retail investors through market price discovery.

Today, most technology startups remain private. However, their funders often do not. Listed companies, listed funds, and institutional vehicles now hold significant exposures to private startups through venture arms, strategic investments, and minority stakes. These exposures sit on balance sheets, not on trading screens. As a result, valuation failure no longer manifests primarily through sudden price crashes, but through impairment testing under financial reporting standards.

This is where the relevance of the dot-com analogy becomes precise rather than metaphorical.

Under Singapore Financial Reporting Standards, impairment is governed primarily by FRS 36 – Impairment of Assets. The standard requires an entity to assess, at each reporting date, whether there is any indication that an asset may be impaired. Where such an indication exists, the entity must estimate the asset’s recoverable amount, defined as the higher of its value in use and its fair value less costs of disposal.

Crucially, FRS 36 is not triggered by failure alone. It is triggered by changes in economic conditions, including technological changes that adversely affect the entity or the market in which it operates. When artificial intelligence materially alters the economics of replication, substitution, and competitive advantage, it directly affects the assumptions underlying future cash flows and terminal values used in impairment assessments.

For many listed funders, startup investments are not consolidated operating assets but are recognised as financial assets or investments in associates. Where such investments fall under FRS 109 – Financial Instruments, they are measured either at fair value through profit or loss, or at cost subject to impairment where fair value cannot be reliably measured. Where influence is significant, FRS 28 – Investments in Associates and Joint Ventures applies, and the investor is required to assess whether there is objective evidence that the investment may be impaired, again by reference to recoverable amount.

Across these standards, the logic is consistent. Carrying values are justified only to the extent that future economic benefits remain recoverable under current conditions. When the cost to recreate a capability collapses, when substitutes emerge rapidly, and when early advantages lose persistence, the assumptions used to support carrying values must be reassessed.

This is why AI matters from an accounting perspective.

Traditional startup valuation embeds assumptions about durability: that execution advantages persist, that early traction compounds, and that capabilities are costly to replicate. AI directly challenges these assumptions by reducing replacement costs and accelerating duplication. When those assumptions weaken, projected cash flows and exit values used in impairment models become overstated, even if the startup continues to operate and generate revenue.

Importantly, FRS 36 does not require a market crash or a failed business model to recognise impairment. A reduction in expected future cash flows, a shortening of useful economic life, or a deterioration in the technological moat can all constitute impairment indicators. In an AI-compressed environment, these indicators may emerge quietly and incrementally, rather than dramatically.

This is the modern equivalent of the dot-com reckoning.

In the early 2000s, valuation failure surfaced through collapsing share prices because assets were publicly traded. Today, valuation failure is more likely to surface through impairment charges recognised by listed funders, often explained as prudence, conservatism, or changing market conditions. By the time impairment is recognised, the economic loss has usually already occurred.

The relevance, therefore, is not historical nostalgia. It is structural continuity. Then, capital mispriced technological participation and the correction arrived through markets. Now, capital may again misprice technological participation, but the correction arrives through accounting.

The mechanism has changed. The underlying error has not.

In this sense, artificial intelligence does not eliminate dot-com-type risk. It relocates it from stock charts to balance sheets, from retail portfolios to institutional financial statements, and from sudden crashes to delayed impairment.

That is why the comparison is not only relevant, but necessary.

Conclusion

This article does not argue that artificial intelligence is a bubble, nor that startups are destined to fail. The technology is real, its impact is observable, and its adoption is already reshaping how work is done. The risk lies elsewhere.

The concern is that valuation logic has not kept pace with the economic consequences of that transformation.

The dot-com era taught a painful lesson that is often misremembered. The failure was not belief in technology; it was belief that early participation in a technological wave translated automatically into durable economic value. That belief was embedded in valuation models that assumed scarcity, persistence, and defensibility where none ultimately remained. When those assumptions broke, the losses were real, permanent, and borne by those who trusted the map.

Today, artificial intelligence compresses execution, accelerates duplication, and lowers replacement costs in ways that directly weaken the assumptions underpinning traditional startup valuation. What once signalled advantage increasingly signals accessibility. What once implied durability increasingly reflects temporary alignment. Yet capital continues to be deployed under familiar frameworks, because those frameworks still appear coherent and disciplined.

The difference from the dot-com period is not the absence of risk, but the path through which it materialises. Technology companies now remain private for longer, while their funders are often listed. As a result, valuation failure is less likely to surface through abrupt market crashes and more likely to emerge through impairment testing under financial reporting standards. The correction, when it arrives, will be quieter, slower, and more technical, but no less destructive.

Impairment is not an accusation of mismanagement. Under FRS, it is the formal recognition that assumptions about future economic benefits no longer hold. In an AI-compressed world, that recognition may become increasingly necessary as substitution accelerates and the economic half-life of advantage shortens.

The purpose of this analysis is therefore not to predict a crisis, but to highlight a structural tension. When valuation models continue to price scarcity in an environment defined by abundance, the gap between carrying values and recoverable amounts widens. History suggests that such gaps do not close themselves.

They are eventually closed by accounting.

This is why remembering the lessons that history taught us is not nostalgic. It is cautionary.

About the Authors

Mr Boo Kok Chuon is the Group Chief Executive Officer of Iconomy Group of Companies. He has a diverse background spanning corporate finance, valuation, accounting, and corporate litigation. His contributions in this article focuses on, valuation methodology, capital structure, and the interaction between technological change and financial reporting. He has been interviewed by The Straits Times on matters relating to finance and entrepreneurship.

Ms Clarie Chan is the Group Chief Financial Officer of Iconomy Group of Companies and the Group Chief Executive Officer of Iconomy Consulting Group. She is a senior corporate finance professional with extensive experience in financial reporting, valuation, and advisory work under Singapore Financial Reporting Standards. Her work in this article centres on impairment analysis, accounting judgement, and ensuring that valuation assumptions remain aligned with economic reality.

Together, they work closely on valuation and advisory assignments, with a particular focus on how emerging technologies affect capital allocation, valuation assumptions, and financial reporting outcomes.