Bytes and Bubbles: Comparing the 90s Dot-Com Bubble and the AI Race

Every disruptive technology sparks excitement, investment, and the risk of a bubble—and AI is no different. Before deciding if AI dreams will materialise, the parallels between past bubbles and the current AI race need to be understood.

The discourse on AI and the ongoing AI arms race over recent years has been inundated with a cacophony of sceptical voices and millenarian projections about an AI-driven age of abundance. Due to the increasing integration of AI into national strategies and geopolitical dynamics, the continuation of AI innovation and development is not just relevant for market leaders like the US but also for emerging players across the world. Every financial year, a recurring question has been whether AI companies are overvalued and whether AI products will live up to expectations. Furthermore, given that the ‘Magnificent 7’ represents over 30 percent of the US market, distress signals like the launch of DeepSeek R1 in January 2025 can have cascading effects, causing global market and supply chain readjustments. Any hint of an innovation slump between the release of frontier AI models raises apprehensions regarding the AI companies and their ability to keep the tech market buoyant. Questions regarding diminishing returns, overvaluation and the inability of AI companies like OpenAI and Anthropic to effectively monetise their products have led to frequent comparisons between the AI race and the dot-com bubble of the 1990s. This paper aims to explore the nature of tech bubbles along with the continuities and discontinuities between the dot-com bubble of the 1990s and similar speculations about the current trajectory of AI development.

AI scepticism

Bubbles in economic sectors emerge when speculative demand and irrational investment cause the price of assets like stocks to rise significantly above their value leading to a capital drought followed by market collapse that can have cross-sector effects. During the 1990s dot-com bubble, there was a surge in demand for assets based on information and telecommunication technologies like the internet that led to valuations significantly outmatched the revenues of tech companies. Although the internet has profoundly changed the global economy, its real impact was not felt till the mid-2000s. The longer-than-expected time scale caused a sell-off of tech stocks during the early 2000s, erasing trillions in market value.

The current tech market shows some signs of a bubble forming, particularly in the AI space. Although platforms like Meta, Microsoft, and Google have seen profits in the last two years due to the range of products they offer, market leading AI-focused entities like OpenAI continue to generate billions in losses regardless of the success of their products. Chipmaking giant NVIDIA which briefly surpassed Apple in January 2025 to become the most valued company in the world continues to generate profits but with shrinking margins. The leading cause of worry is that promises of AI-driven paradigm shifts across societies made by developers may not materialise. While AI systems like ChatGPT have continued to improve in various aspects like text and media generation, coding and research, the revolution still seems a way off. Furthermore, the development of increasingly powerful frontier AI models has not been accompanied by a proportionate rise in productivity and monetisation. Microsoft CEO Satya Nadella stated in an interview that in the absence of GDP growth, “[AI developers] self-claiming some AGI milestone, that’s just some nonsensical benchmark hacking”. The real marker of AI progress, according to Nadella, is wider economic growth that can recursively secure investment in the AI sector.

Market uncertainty, particularly for semiconductors, is also being driven by rising geopolitical tensions and competitiveness between the United States and China. President Donald Trump’s tariff regime can challenge semiconductor production growth given the globally distributed nature of the semiconductor supply chain. The out-going Biden-administration’s Framework for Artificial Intelligence Diffusion (FAID) published in January 2025 (before the launch of DeepSeek R1) intensified export controls to further isolate China from the AI value chain established by the West. The FAID is supposed to streamline the process of acquiring export licenses for advanced AI chips; however, the framework was drafted prior to the release of open-weight Chinese AI models like DeepSeek-R1 and the more recent Manus AI agent that could compete with Western frontier models. The consistency of private sector investment in AI will depend on complicating factors like Chinese innovation in the AI sector, the question of how the Trump administration enforces the FAID or whether the framework (along with the 2022 CHIPS act) even continues to exist or is replaced by a tariff-based licensing framework.

Slow but Steady Uptake

Tempering high expectations is necessary to avoid economic distress, but caution should be applied in a measured manner to avoid unnecessary obstacles that may hinder innovation at the frontiers. While the AI-driven boom may rhyme with the dot-com bubble, it is not a repetition. A critical component of the dot-com crash was over-investment in network architecture (like fibre-optic cables) that grossly outmatched demand at the time. Conversely, a bottleneck for developing frontier reasoning AI models is that demand is outpacing supply. The data and energy-intensive nature of scaling inference and computing is causing a surge in demand for data centres. According to a Goldman Sachs report, the global market capacity of data centres will double by 2030 leading to the global power demand for data centres to increase by 165 percent. Even in a scenario that sees demand for AI plateauing or decreasing, the strategic relevance now accorded to the technology has enabled unprecedented public sector investment across developed and developing economies. For instance, the Stargate Project initiated by the Trump administration established a funding pipeline for AI infrastructure and established AI as a priority for national security. Following the 2025 Paris AI Action Summit co-hosted by India and France, the EU announced the InvestAI initiative to facilitate a 200 billion euro investment in the AI sector with 20 billion euro earmarked for AI gigafactories. Before the Summit, the United Arab Emirates (UAE) entered an agreement with France to invest 30 to 50 billion euros towards building an AI campus and a 1 gigawatt data centre in the country.

AI adoption and maturity across economic sectors have also been steadily increasing since 2021. A report by Boston Consulting Group states that data-rich sectors like tech and finance show the highest levels of maturity while historically lagging sectors like automotive and industrial goods are now exhibiting the fastest adoption rate. AI maturity is also on the rise across geographies with emerging markets like India and the UAE growing faster than many developed countries. AI models are also being adopted by the public sector. For instance, the United Kingdom (UK) government has signed an MOU with Anthropic to collaborate on integrating AI to increase the efficiency of public services. OpenAI has launched ChatGPT Gov to streamline the US government’s adoption of its AI models and the company is also partnering with the Estonian government to integrate its ChatGPT Edu model into a national education system. Such developments indicate that while market conditions may be unstable, a private and public sector interest in AI uptake is certainly present. Although the rate of AI adoption may not be meaningfully adding to wider economic growth yet, it should be remembered that ChatGPT was released just over two years ago.

Painful lessons were learnt during financial crises like the dot-com crash and the 2008 recession. The soaring demand for AI-anything certainly echoes the investment frenzies of the past. However, amidst the AI hype cycle, it should be remembered that network technologies in the dot-com era did transform the global economy eventually. The crash happened because it took longer for the internet to take root than investors and developers initially expected. The possibility of AI-driven progress occurring on a longer timescale is real but the question of timescale should not be conflated with the question of likelihood. From a governance perspective, the longitudinal priority of policymakers and regulators in emerging markets should be on promoting AI maturity and consolidating their position in segments of the AI value chain rather than chasing the next frontier models. Countries like the UAE and India have already begun taking steps in this direction by investing in up-skilling initiatives and building a local data centre and chip fabrication capacities. The presence of multifarious destabilising factors like disruption in the AI value chain, geoeconomics rivalries, pivoting policies and slow uptake may likely cause a market readjustment in the short to medium term. Nevertheless, mandates by governments across the world to prioritise investment in frontier technologies are illustrative of AI’s entrenched relevance in the great power competition of this era.

Siddharth Yadav is a Fellow in Technology at ORF Middle East