This article was first published in the forum, The Edge Malaysia Weekly, from 13 July 2026 to 19 July 2026.
The geoeconomics of artificial intelligence and its financialization is key to understanding how to survive the AI bubble cycle.
The Internet has changed traditional economic thinking by combining the near-zero costs of digital transactions with the economies of scale of networks. Before the Internet, companies operated and generated profits based on crude average pricing. Trading was paper-based, time-consuming, and the market was limited. Profit occurs when the average price (revenue) of goods sold is higher than the average cost. I rarely sell at a loss and only invest when I can expect a higher return. Digitalization has changed everything in terms of speed, scale and scope.
The idea for loss leaders originated in American supermarkets, which sell selected popular items at discounted prices to attract customers who would buy other items at regular prices. This loss leader concept began when digital commerce became able to reach customers around the world. Once you reach hyperscale (billions of customers), you can monetize your network through advertising, subscription fees, and strong valuation through stock market listings. That’s the winner-takes-all effect of networks.
In the run-up to the dot-com bubble and its burst in 2000, global technology platforms developed freemium + infrastructure or cloud business models that generated huge profits while incurring heavy losses until they became profitable. It took Amazon nearly a decade to reach full-year profitability. Therefore, the stock market has accepted that technology platforms may incur losses in order to gain scale. Expectations of huge technology profits created a stock market euphoria that gave rise to today’s spectacular seven or ten publicly traded companies with market capitalizations in excess of US$1 trillion. The Freemium+ model was a Silicon Valley/Wall Street innovation that strengthened American hegemony in finance through Techfin. Fintech is the application of technology to finance, while techfin is the monetization of technology through the stock market.
Fintech models essentially reduce intermediary costs through digitization and, today, tokenization. However, current regulations protect institutions such as banks from fintech platforms that may infiltrate their territory through digital banking and investments. Digital competition in fintech has essentially reduced intermediation margins, resulting in banks taking on more leverage and derivative risks. That was the root cause of the 2007/08 global financial crisis. The fintech revolution exploded after the COVID-19 pandemic pushed e-commerce to the point where consumers could use electronic payments and e-commerce from cash and cards virtually seamlessly.
Google’s improved Freemium+ model offered free services to as many customers as possible, extracted revenue from advertising, and used the revenue to develop further business at the ecosystem level. Amazon.com’s model was to get into the cloud business by selling Prime and Cloud subscriptions, locking in Premier customers into its ecosystem. Stock markets and private equity have begun evaluating these hyperscale platforms based on the size of customers they reach, their speed of customer acquisition, and their ability to monetize (extract profitable returns from such customers).
Technology has thus created a business model where platforms can “pay for customer acquisition” by being willing to spend money to provide free services to attract customers to achieve scale. They can afford to burn cash because they are funded by private equity investors who believe in their business model through a proof-of-concept initial public offering (IPO) process. However, to maintain customers and their market reputation, a start-up or publicly traded company must prove that it can generate future revenue from the acquired customer base. This creates a huge cash flow gap as there is a gap between current losses due to heavy investments in capital expenditures (capex), software/model building costs resulting in negative current cash flows, and future revenue cash ins.
The stage when a company’s cash flow becomes negative is called the “valley of death.” If you can show that you can break even and move from a negative J-curve to an S-curve with cash flows that benefit from premium customers, you can survive.
Freemium model monetization is achieved by asking your loyal customers to subscribe and pay a monthly fee for shopping convenience. Amazon has Prime customers, and in addition to paying a monthly fee, you also get a great service that generates far more revenue than using your browser every once in a while. The Spotify and Netflix models work well with subscriptions that provide convenient music and movies when customers need them. A “lock-in” effect arises from customers who feel they have to use the system to get the most value from their subscription.
OpenAI (ChatGPT), Anthropic, and other AI model builders are all working hard to increase their customer bases and revenues so they can benefit from expected IPO public offerings at high valuations. Their entire business model relies on burning cash until they reach a scale where subscribers pay enough to escape the Valley of Death. Most AI users stay at the free level of testing, and only a handful start paying higher fees for further use.
In essence, the AI boom reflects an overbuilding of upfront infrastructure, amplified by a financial architecture that decouples market valuations from underlying economic productivity. This creates an investment bubble that widens the gap between large upfront capital investments and end-user monetization in an unknown future. Simply put, you need to provide real returns, not just a dream. Otherwise, valuations will collapse across the board.
The magnitude of the cash flow gap is illustrated by the latest estimates that global AI capital spending will reach between US$600 billion (RM2.4 trillion) and US$757 billion annually, with most of it going to data centres, chips and energy grids. To justify this, based on a yield of 10% per year, the implied return from such capital investment is estimated to be between US$600 billion and US$650 billion per year in total returns, depending on the rate of return. The higher the profit margin, the lower the total revenue required.
Unfortunately, the AI industry’s current end-user revenue from software subscription fees and cloud fees is between USD 50 billion and USD 150 billion annually, meaning that AI platforms will have to face a “revenue gap” of as much as USD 450 billion to justify their current market valuations.
In any bubble cycle, a peak of inflated expectations can lead to a trough of disillusionment before productivity gains eventually ripple through the economy. This can take anywhere from 10 to 30 years.
In other words, in the euphoria of a gold rush, not everyone can find gold, and when the limited profits are shared among many people, the dream of getting rich quickly is shattered, so there is no justification for overvaluation in the market.
This “financialization” of technology may be structurally unsustainable for three reasons.
First, there is a mismatch between capital investment and asset life. Infrastructure such as traditional data centers such as buildings, cooling systems, and servers are depreciated over 5 to 15 years. But if new technologies make AI cheaper and without the need for vast amounts of computing power, asset values could depreciate, say, one to three years faster. Technology obsolescence occurs faster than traditional accounting depreciation. Technology obsolescence turns assets into liabilities overnight.
Second, there is the unsustainable circular revenue story. Currently, software companies like OpenAI must purchase or rent vast amounts of computing power to develop, train, and make models available to customers. They don’t have the cash to invest in capital expenditures, so they call chip manufacturers to invest, then buy chips and fund capital expenditures for their computing power needs.
From the perspective of Nvidia, Broadcom, and Oracle, whose oligopolistic market power in the production of AI chips and equipment has earned them high market valuations, there is a seemingly magical business model of generating revenue from equity. Chip companies invest, say, $10 billion in model builders, which then use the money to rent or buy computing power from chip companies. The chip manufacturing company has US$10 billion in revenue while owning US$10 billion in equity assets. This circular financing allows model builders to expect higher equity returns if they succeed, but if they fail, they will incur write-downs, offset by profits already generated from chip and equipment sales.
Third, the risks of technology financialization become higher and deeper through leverage, as large hyperscalar technology companies with huge stock market valuations start taking on debt to finance investments in AI model builders and their own chip foundries and data centers.
Historically, Big Tech companies have been underutilized because they have strong cash flows and can leverage high stock valuations to acquire startups or companies that align with their business strategies. Additionally, a significant portion of current AI infrastructure is financed through off-balance sheet special purpose vehicles (SPVs) and private credit markets. Financial regulators are already sounding the alarm as leverage levels rise in the system. Because of the concentration of revenues, large companies, and debt, if valuations were to collapse, the entire system could go down.
On the positive side, the AI bubble may now be funding massive physical capacity that is likely to drive the digital economy for decades to come, but it remains to be seen whether this expensive financial architecture that monetizes the construction phase through stock market valuations can withstand revenue and cash flow shocks.
In summary, while large countries and major hyperscaler platforms are investing heavily in AI, the rest of the world, which cannot afford such large capital investments, will focus on the widespread application of AI tools to improve productivity. Nobel Prize-winning economist Robert Solow famously said in 1987, “The computer age is everywhere except in productivity statistics.” Spread into general productivity and gross domestic product growth could take decades to materialize.
My own experience applying AI tools to small and medium-sized enterprises (SMEs) has shown me that while everyone seems excited about the promise of AI, the actual application can be much more tedious and complex than most people think. First, while AI promises a lot, it is difficult to find a flexible workforce that is adept at using AI and understands how to adapt these tools to on-the-ground situations where legacy software, hardware, and mindsets resist change.
Second, new AI technologies require an entire ecosystem of complementary assets, processes, procedures, and operators that take time to build to function properly. It typically takes a lot of effort to understand how AI can be incorporated into current business processes and workflows, retrain employees to use new tools effectively, and develop new management structures.
The reality is that small businesses lack the capital to experiment early and the awareness of the benefits of AI. They wait until the technology is cheap, standardized, and user-friendly. Meanwhile, as early adopters improve productivity and products, the gap between large enterprises and leading AI adopters and laggards will only widen.
In reality, it will take a lot of effort for small and medium-sized enterprises to transform their companies and their economies into AI-driven, productive and innovative economies. Government documents, certifications, and processes get in the way of every digital transaction. AI cannot handle non-digital documents. Therefore, governments need to improve their digital public infrastructure to take advantage of the AI revolution. Small businesses need support from chambers of commerce, universities, and government agencies to rapidly transition into the AI era.
Malaysia is facing this major change management stage where it is easy to fall into thousands of bureaucratic red tapes. If Malaysia wants to create the next unicorn, supporting more SMEs in their AI journey must be a top priority.
Tan Sri Andrew Shen writes about global issues from an Asian perspective
Save by subscribing to us for your print and/or digital copy.
P.S.: The Edge is also available on Apple’s App Store and Android’s Google Play.
