Currently, the stock market seems to cheer companies that invest heavily in data centres, bringing the market cap value of Nvidia to US$4.3 trillion and Oracle to US$869 billion. Nvidia is using its cheap capital to invest US$10 billion in integrated chip maker Intel, as well as US$100 billion in OpenAI partnership. Do these investments make sense or are they signs of investor hype that would further fuel the AI bubble?
In its latest second-quarter earnings report, Intel’s revenue rose US$0.1 billion to US$12.9 billion, compared with the same period last year. The company is still unable to generate profit, making its sixth consecutive quarterly loss and the longest losing streak in 35 years. The US government acquired a 9.9% stake in Intel for about US$8.9 billion, overtaking BlackRock as the largest shareholder. Just as Microsoft’s timely investment in Apple in 1997 provided the cash it needed to revive and reinvent its business to become one of the most valuable companies in the world, these new investments may provide sufficient new capital to enable its new CEO Tan Lip-Bu to execute its innovation and process turnaround, including customising its general-purpose processors to work with Nvidia’s graphics processing units inside data centres.
As more data centres are built and the volume of data movement for AI training grows, investor interest is focusing on networking optics, which are the fibre cables and optical transceivers that move data between servers. Broadcom and Marvell are the major suppliers of optical interface products, both supplying to hyperscalers (Amazon Web Services or AWS, Microsoft Azure, Google Cloud and Meta). About 70% of the AI workloads will be hosted by these hyperscalers. Broadcom has a market cap of US$1.6 trillion with a price-earnings ratio (PER) of 124 times, while Marvell is valued at US$64 billion but is still unprofitable. Most AI companies train models largely on hyperscalers’ public clouds, but private clouds are going to rise as more companies want control over their own data. The cloud industry is expected to generate between US$1.6 trillion and US$3.4 trillion by 2040.
Faster data transfer speed requires more complex and costly digital infrastructure. This creates an opportunity for hyperscalers and cloud builders to expand their supplier base instead of buying finished servers from large original equipment manufacturers. Buying different subcomponents from large and small niche vendors and using original design manufacturers or electronic manufacturing services to assemble and package modules can reduce costs. Since semiconductors power this entire digital infrastructure, growth in the sector will continue to accelerate.
See also: Bank of Singapore uses AI agents to cut source of wealth report time to one hour
While the US accounts for 40% of the global data centre market, Europe contributes around 20% due to high energy costs to operate 24/7 European data centres. AI workload increases are driving expansion in Asian data centre markets, with India, Malaysia, the Philippines and Thailand growing at triple-digit rates. These locations are attractive because co-location facilities allow foreign companies to share infrastructure such as space and IT equipment, making it easier to scale without building their own data centres. However, scaling data centres requires highly efficient power infrastructure, such as energy and cooling systems that are increasingly constrained by limits in grid capacity and water availability.
In developing countries, foreign companies sponsor about half of all data centre projects but contribute only 45% of the total investment value. The largest funding comes from domestic players and sovereign wealth funds (Temasek and Government of Singapore Investment Corporation), particularly in India and Malaysia. Data centres are not just physical infrastructure; they have become strategic assets that attract significant cross-border investments.
The AI bubble is premised on exponential growth of consumer, business and government demand for AI services, such as increases in productivity and competitive edge. The AI industry also hopes that consumers will ultimately pay more for its services. However, it is also taking up more and more costly investments in hardware and software infrastructure, as well as energy consumption, not including carbon emission costs.
See also: The babble about a looming AI bubble
Goldman Sachs has just issued warnings that value investors are increasingly worried that such heavy investments in data centres may have a short-term impact on earnings, which will reward some companies but not all companies in the AI sphere. In other words, even if AI delivers earnings and productivity in the economy as a whole, not all companies will win, and if the majority of companies suffer from the obsolescence of their legacy assets, the markets could react accordingly.
In an earlier review, Goldman Sachs surveyed historical asset bubbles, such as the 1630s tulip mania in Holland, the 1720 South Sea bubble in Great Britain and the Mississippi bubble in France, the 1840s–1873 railway bubbles, the 1920s US stock market boom, and the 2000 Nasdaq boom and bust. The bubbles followed a common pattern. A breakthrough in science or technology emerges and reaches commercial scale. New companies emerge and capital floods into the new area, causing higher prices. Speculation builds and valuations of companies rise, often resulting in a bubble. The bubble bursts, but the technology tends to re-emerge as a principal driver in the economy and stock market.
However, the technology/industry becomes dominated by a few large players, and secondary innovations emerge, creating new companies and products that are enabled by the initial technology and its increased adoption. Companies that do not adapt could disappear. These secondary innovations create new employment opportunities and, with them, new sources of demand. Productivity tends to rise, but usually only after the full adoption of this new technology and network effects are realised.
The speed of innovation creates significant changes in society, resulting in shifting social attitudes, consumer behaviour, government policy and business practices and operations. In recent years, there is an assumption that science and technology can solve all ills, which may not necessarily be true. The bursting of bubbles reminds everyone that in all markets, there are winners and losers.
Based on mainstream valuations, such as the Shiller cyclically adjusted price-earnings (CAPE) ratio, the US market does seem high. But many investors think that it may burst, but not just yet. The problem with all technology is that new technologies may emerge that may completely disrupt the older technology. As investments get more expensive, the bets become higher and the returns to higher investments become more and more difficult to achieve.
The ultimate winner is the one who cashes in near the top, stays cool and liquid and then buys up the stocks when the valuations become ludicrously cheap. To be fair, no one knows when the market will turn. But what goes up will come down, just as what the J-curve tells us, that we have to invest before our returns pay off. Next month, we will know whether there will be an October bull again, or an October bust. Only time will reveal the ultimate truth.
Andrew Sheng writes on global issues from an Asian perspective. Loh Peixin is a research associate at the George Town Institute of Open and Advanced Studies, Wawasan Open University. The authors are engaged in a major study of the tech industry in Penang
