AI versus the dotcom bubble: 8 reasons the AI wave is different

Since the artificial intelligence (AI) big bang of November 2022, AI-related technology stocks have outperformed significantly evoking comparisons and fears about a dotcom-like bubble, as we saw in the late 1990s. Most commentaries including AI comparisons provide hindsight assessments and rarely capture the full context of the era. In our previous article: Are tech sector investors too bullish on AI’s promise? we briefly touched on the key differences and similarities. Here we provide our views based on reflections of the team’s experience of having invested through the dotcom period, and explain why this time round very different dynamics are at play.

Over the last 20 years or so technology stocks have returned more than twice that of broader global equities, with technology having outperformed more than 80% of the time and also delivered a positive return of more than 80% of the time through that period.(1) Despite this persistent outperformance, the fear of a dramatic drawdown, like that seen between 2000-2002 after the dotcom crash has continued to haunt many investors.

Why is AI not just a theme but an evolving multi-year tech wave?

We believe the adoption of AI is different. We consider AI to be the fourth wave of technology. Importantly, waves are different to themes. Themes for us include 3D printing, cybersecurity, and electric vehicles; they tend to have specific end markets and a narrower focus.

Waves, however, are defined by their impact across the whole economy. They create thematic ripples and importantly require investment across every layer of the technology and communication stack. This ranges from silicon, compute, storage, networking to software, devices, power and connectivity. Each wave has sequentially connected more people, devices, and required incrementally more investment. The investment required for each wave has taken longer and been greater in magnitude than investors anticipate for multiple years.

Waves of compute are not isolated, separate entities. The reality is that one wave lays the foundation for the next and typically they tend to build rather than end abruptly. For example, PC internet shifting to laptops with broadband connectivity laid foundations for true mobile compute with apps enabled by cloud infrastructure. This then opened the door to the evolution of AI.

Importantly, the dotcom bubble is effectively the only compute wave to have an abrupt end. Rather than fade and mature as other waves have – it reached a crescendo after six years (with two years of generating in excess of 70% returns) before it abruptly crashed.(2)

How does the AI boom differ from the dotcom bubble?

Recognising some of the unique aspects of the internet bubble are important to avoid false analogies with the current AI wave:

1. Y2K was a unique contributor to the dotcom crash

The role of Y2k is often overlooked as a contributory factor to the manner in which the internet bubble burst. The switch to year 2000 was viewed as an expiry date for all existing legacy computer systems, which couldn’t cope with the move from two digits: 99 to 2000. It was a global, non-negotiable deadline that pulled forward IT spending. The fear of mass outages resulted in huge IT projects, accelerating replacement capacity not just new capacity for the internet. The condition of forced upgrades and expiry has never been repeated in any subsequent technology wave.

2. Fraud inflated the competitive internet capex environment

The rapid pace of capital expenditure (capex) during the internet wave was partially fuelled by companies that were not subject to the same audit standards as today. The collapse of WorldCom was not only symptomatic of the dotcom bubble, but also contributed to the overestimation of demand.

WorldCom was a key player driving the fibre optic cable building frenzy of the late 1990s. It led to aggressive spending on mergers and acquisitions as well as capex to build infrastructure for the immense new internet traffic expected from dotcoms and the adoption of the World Wide Web. Fibre build outs were commodities, with no competitive differentiation other than how much capacity could be built. WorldCom’s spending was a mixture of aggressive and genuine spending, but also massive fraudulent accounting (more than US$11 billion) that other companies of that era tried to compete with. This artificially inflated demand. The Sarbanes-Oxley Act of 2002 was a legislative response to WorldCom (and Enron, which also had a broadband business) and has ensured higher audit standards that have effectively avoided that type of behaviour since. Ironically, NVIDIA replaced Enron in the S&P 500 in 2001 as a result. But despite greater transparency in this tech wave, there are still concerns around funding and finance structures.

3. More robust capex and start-up funding for AI

Private equity and credit funding for AI capex has been very robust but has been focused on larger company raises and datacentre projects by Open AI, XAI, and Anthropic, with seven of the highest valued private tech companies on paper worth more than US$1.3 trillion. According to Forge Global, a marketplace for private investments, 19 AI firms have raised US$65 billion this year – over three quarters of all private market capital.(3) With this level of funding available privately, there is little incentive to go public early and is evident in the initial public offering (IPO) market of now versus the internet era.

In addition, while today almost 20% of technology companies are unprofitable, that is significantly lower than the 36% observed back in the internet era. Tech companies also now have the strongest balance sheets across sectors.

Figure 1: Less tech companies are unprofitable versus dotcom era

While funding fueling AI and the dot.com era are very different, there are some emerging similarities in financing complexity that bears watching:

4. Risks from financing complexity: Circular references and vendor financing

As the internet capex boom developed, vendor financing and take or pay contracts became increasingly common. Vendor financing is when a company provides significant amounts of credit for companies to buy their products. In the internet bubble, now defunct telecom equipment company Lucent Technologies at its peak had over $15 billion in vendor financing versus operating cash flow of only around $300 million.(4) The massive projects that WorldCom agreed to were often on a take delivery or pay a penalty model whether they had demand or not.

Recently, the AI space has become increasingly interwoven with suppliers funding customers, rising customer concentration, revenue sharing between customer and vendor, take-or-pay contracts, and vendor repurchase agreements. NVIDIA (NASDAQ: NVDA) for example, is entering a strategic partnership with OpenAI to build and deploy 10 GW of datacentre capacity using NVIDIA systems, supported with $100 billion investment in OpenAI.(5)

OpenAI has also entered into an agreement with NVIDIA competitor AMD, with an option to acquire up to 10% of AMD over time.(6) OpenAI requires vast amounts of computing power and has agreed deals to lease this from companies like Alphabet (Google Cloud), Oracle and CoreWeave. OpenAI makes up a significant portion of Oracle’s remaining purchase obligations (RPOs) as well as of other neoclouds like CoreWeave creating customer concentration risk.(7)

We have mixed views on this as OpenAI requires significant funding for these deals that have yet to be secured. In contrast the US$100 billion investment by NVIDIA is made possible because of its exceptional balance sheet and cash flow, giving it a stake in OpenAI. Additional funding for OpenAI can drive Alphabet and other companies to continue to competitively invest more in AI infrastructure. As Google (NASDAQ: GOOGL) CEO Sundar Pichai said in January this year, “The risk of overinvestment is less than the risk of underinvesting.”(8) However, this does create circularity, and risk that the pace of AI demand creation is overestimated. If this were to become the primary or sole source of AI funding, that would be of increasing concern.

5. Focused AI capex means better supply and demand visibility

In the internet wave, small technology companies came to equity markets to raise capital to buy semiconductors, servers, storage, networking equipment and software that were needed to ramp up volume. As the adoption of internet services and ecommerce took longer to ramp than anticipated, it created overcapacity and grey markets for a wide range of equipment. Inventory-to-sales ratios became bloated across the sector.

Now, thanks to the mobile cloud era and hyperscalers ramping cloud capacity, companies no longer have to raise capital for equipment; they can simply ‘rent by the minute or second’ according to their capacity needs. Many of the companies building out AI capacity like Microsoft (NASDAQ: MSFT), Alphabet and Amazon (NASDAQ: AMZN) have amongst the highest free cash flow and strongest balance sheets in the equity market – this in contrast to the IPO-fuelled internet demand previously.

Figure 2: Technology sector has the strongest balance sheets

These companies are also experts in compute logistics, having managed not only their own enormous demands, but that of hundreds of thousands of companies over the last decade. A concentrated group of seasoned technology companies on both the cloud and semiconductor side are managing this AI supply ramp. Demand visibility is also higher than it was for the dotcom start-ups.

Demand for accelerated compute for AI continues to outstrip supply. Generative AI uses more than 1000x the compute that perception AI needed, and as we shift to agentic AI (building context and reasoning) a further 30-100x increase in compute power is required. Foundational models are moving from training improvements in 2023-24 into inferencing, driving a further exponential increase in demand.

6. Geopolitics: Globalisation then versus deglobalisation now

In the late 1990s and early 2000s globalisation accelerated, driven by landmark trade agreements such as the launch of the euro, the US granting China permanent normal trade relations, and the Information Technology Agreement (ITA) on telecom liberalisation, which resulted in rapid multinational expansions and cross-border investments, rising commodity prices, and accelerating global trade. Conversely, 2025 has been characterised by a shift in trade policy, increasing tariffs and a focus on localising manufacturing as well as data sovereignty. These factors are increasing the demand for AI capex as more countries see AI as a strategic national priority, needing a manageable local supply chain and source.

7. Macroeconomics: Rates and inflation dynamics

In 1999 in the US, inflation was rising, unemployment had fallen to 4%, commodity prices were rising sharply, and interest rates were firmly in a tightening cycle, with rates rising from 4.75%, reaching 6.5% before the dotcom bubble burst in 2001. At present, while the current government shutdown has reduced data transparency, we appear to be in a period of weakening employment, signaling a loosening rather than tightening monetary backdrop.

8. AI valuations underpinned by earnings growth

The sharp run up in share prices in AI-related names has been driven by corporate earnings growth, rather than pure multiple expansion (company valuation). As noted earlier, with fewer start-ups, the public market large technology stocks have not shown signs of excess in the way that they did in 2000. In 2000, the technology sector traded at more than double that of the broader equity market. Currently it trades lower, at around 1.34x the broader equity market.(9)

NVIDIA is a good example of how this early phase of AI has been led by earnings growth. We continue to see strong positive earnings momentum from the technology sector driving economic growth. A sign of a market bubble is when the multiple (valuation) expands without profit growth – something we have yet to experience in the current AI wave, unlike in the dot.com era.

Figure 3: Tech valuations: A long way from 1999/2000 highs

Relative Technology to World forward P/E

​Conclusion

We continue to believe that AI represents a major new wave in technology. These waves typically take multiple years to play out, and while we are no longer in the very early stages of the AI wave, we remain excited about the breadth of investment opportunities that continue to emerge. Despite parallels with the dotcom era in terms of the levels of spending and disruption, we believe that this wave is more likely to ebb and flow – creating higher returns accompanied by higher volatility as it evolves – rather than bursting as the internet bubble did given multiple unique characteristics.

Managed Fund

Janus Henderson Global Research Growth Fund

Global Shares

View