This article first appeared in Forum, The Edge Malaysia Weekly on July 13, 2026 - July 19, 2026

The geoeconomics of artificial intelligence and its financialisation are key to understanding how to survive the AI bubble cycle.

The internet changed traditional economic thinking because it brought together the phenomenon of near-zero cost of digital transactions with network economics of scale. Prior to the internet, businesses operated on crude average pricing to generate profit. Transactions were paper-based and slow, and markets were limited. Profit is generated when average prices of goods sold (revenue) are higher than average costs. You seldom sell at a loss and you only invest when you expect a higher profit. Digitisation changed all that in terms of speed, scale and scope.

The idea of loss-leaders came with American supermarkets that sold selected popular items at a discount to attract customers who would then buy other items at regular prices. This loss-leader concept took off when digital commerce could reach global customers. If you reach hyperscale (billions of customers), you can monetise your network through advertising, subscription fees and getting high valuations from listing on the stock market. That is the winner-take-all effect of networks.

In the run-up to the 2000 dotcom bubble and bust, global tech platforms developed the freemium+ infrastructure or cloud business model that generated huge profits despite making large losses until they reached profitable status. It took Amazon nearly a decade to reach full-year profit status. The stock market therefore accepted that tech platforms can make losses in order to reach scale. The expectation of huge tech profits gave rise to the stock market euphoria that created today’s Magnificent Seven or Ten listed companies valued at more than US$1 trillion each in market cap. The freemium+ model was a Silicon Valley/Wall Street innovation that reinforced American hegemony in finance through techfin. Fintech is the application of technology to finance, whereas techfin is the monetisation of technology through stock markets.

Fintech models essentially reduce intermediation costs through digitisation and, today, tokenisation. However, current regulations protect institutions such as banks from fintech platforms that could invade their territory through digital banking and investing. The digital competition in fintech basically drove intermediation margins thinner, which led banks to take on more leverage and derivative risks. That was the fundamental cause of the 2007/08 global financial crisis. The fintech revolution exploded after the Covid-19 pandemic pushed e-commerce to the level where consumers switched from cash and cards to use e-payments and e-commerce virtually seamlessly.

The freemium+ model, which Google refined, provided free service to as many customers as possible, extracted revenue from advertising and used the proceeds to develop more business at an ecosystem level. The Amazon.com model latched onto the cloud business by selling Prime and Cloud subscriptions, thus locking in premier customers to their ecosystem. The stock market and private equity began to value these hyperscale platforms based on the scale of customers reached, the speed of customer acquisition and the ability to monetise (derive profitable revenue from such customers).

Technology thus created a business model where a platform can “pay for customer acquisition”, by being willing to spend money to give free service to attract customers to reach scale. They can afford to burn cash because they are funded by private equity investors who believe in their business model through the concept-proof of concept-initial public offering (IPO) process. However, to maintain their customers and their market valuation, the start-up or listed company must prove that they can generate revenue in the future from their acquired customer base. This creates a huge cash flow gap because there is a gap between current losses from heavy investment in capital expenditure (capex), costs in software/model building resulting in negative current cash flow and future revenue cash inflow.

The phase when the company is going through negative cash flow is called the Valley of Death. You survive when you start breaking even and show that you move from the J-curve of negative cash flow to the S-curve of reaping profits from premium customers.

Monetisation of the freemium model arrives when you ask your good customers to subscribe and pay monthly fees for the convenience of shopping. Amazon has Prime customers who not only pay monthly fees but also get better service that generates far more revenue than occasional browsers. Spotify and Netflix models work well with subscriptions that give customers the convenience of music or movies when they want. The “lock-in” effect comes from customers who feel that they must use the system to get maximum value from their subscriptions.

OpenAI (ChatGPT), Anthropic and other AI model builders are all frantically trying to get their customer base and revenue up so that they can benefit from their expected IPO public listing with high valuations. Their whole business model depends on them burning cash until they reach the scale where their subscribers pay enough to escape the Valley of Death. Most AI users stay at the free level of testing and only a handful begin to pay serious fees for more usage.

In essence, the AI boom reflects an upfront infrastructure overbuild, amplified by a financial architecture that detaches market valuations from underlying economic productivity. This creates an investment bubble with a growing gap between massive capex up front and end-user monetisation in the unknown future. In simple language, you must deliver real revenue and not just dreams. If you don’t, there will be an almighty bust in valuation.

How big the cashflow gap is can be seen from the latest estimates of the global AI capex of between US$600 billion (RM2.4 trillion) and US$757 billion annually, mostly into data centres, chips and energy grids. To justify this, based on a yield of 10% annually, the implied revenue generation from such capex is an estimated gross revenue of US$600 billion to US$650 billion annually, depending on the profit margin. The higher the profit margin, the lower the gross revenue needs.

Unfortunately, the current end-user revenue from software subscription fees and cloud fees for the AI industry is between US$50 billion and US$150 billion annually, meaning that AI platforms must face as much as a US$450 billion “revenue gap” to justify current market valuations.

In every bubble cycle, the peak of inflated expectations may end up in a trough of disillusionment before productivity gains eventually diffuse through the economy. This could take as long as 10 to 30 years.

In short, in any gold rush euphoria, the dream of “getting rich quick” is shattered when not everyone finds gold and limited profits are shared with many, so supervaluations cannot be justified by the market.

This “financialisation” of technology could be structurally unsustainable for three reasons.

First, there is a capex asset-life mismatch. Infrastructure such as traditional data centres in terms of buildings, cooling systems and servers is depreciated over five to 15 years. But if a new technology provides AI cheaper, without requiring huge computing power, then the asset value may have to be depreciated faster to, say, one to three years. Technological obsolescence moves faster than conventional accounting depreciation. Assets become liabilities overnight through tech obsolescence.

Second, there is a circular revenue story that cannot be sustained. Currently, software companies like OpenAI must purchase or rent huge computing power to develop, train and allow customers to use their models. Since they don’t have the cash to invest in capex, they invite the chip manufacturers to invest in them and then they buy the chips and fund the capex for their computing power needs.

From the viewpoint of Nvidia, Broadcom or Oracle, which have high market valuations from oligopolistic market power in producing AI chips or equipment, there is an equity-to-revenue business model that seems magical. The chip company invests in the model builder, say US$10 billion, which the model builder in turn uses to rent or buy the computing power from the chip company. The chip builder books revenue of US$10 billion while having an equity asset of US$10 billion. This circular funding has the promise of a higher equity return if the model builder succeeds but, if it fails, there is a write-off, which is offset by the profits already generated by the chip or equipment sales.

Third, tech financialisation risks rise ever higher and deeper with leverage, where the hyperscaler big tech companies with huge stock market valuations begin to take on debt to finance their investments in the AI model builders and in their own chip foundries or data centres.

Hitherto, BigTech companies are not that leveraged because they have good cash flow or can use their high equity valuations to acquire start-ups or companies that fit their business strategy. Furthermore, a substantial portion of current AI infrastructure is financed via off-balance-sheet special purpose vehicles (SPVs) and private credit markets. Financial regulators are already sounding warning bells when leverage levels rise in the system. It is the high concentration of revenues, big players and debt that could bring down the whole system if and when valuations crash.

On the positive side, the AI bubble may have financed huge physical capacity today that will likely power the digital economy over the next few decades but it remains to be seen if this expensive financial architecture that monetises its construction phase through equity market valuation can withstand any revenue-cashflow shocks.

To sum up, while the great powers and big hyperscaler platforms are investing heavily in AI, the rest of the world which cannot afford such heavy capex will be focused on AI diffusion, which is the broad-based application of AI tools to improve productivity. Nobel laureate economist Robert Solow famously said in 1987: “You can see the computer age everywhere but in the productivity statistics.” That diffusion into general productivity or gross domestic product growth may take decades to realise.

My own experience in applying AI tools to small and medium enterprise (SME) usage showed that even though everyone seems enamoured with AI promises, the actual application may be far more tedious and complicated than most people think. First, AI promises a lot but it is not easy to get flexible workers who are adept at using AI and understand how to apply these tools to fit ground conditions where legacy software, hardware and mindsets are against change.

Second, for the new AI technology to work properly, it requires a whole ecosystem of complementary assets, processes, procedures and operators that takes time to build. Typically, there is a lot of hard work in understanding how AI can be built into current business processes and workflows, retraining the workforce to use the new tools effectively and developing new managerial structures.

Realistically, SMEs lack the capital and awareness of AI benefits to experiment early. They wait until the technology becomes cheap, standardised and user-friendly. In the meantime, the early adopters improve productivity and products, which means that the divide between larger companies and advanced AI adopters and laggards will only widen.

In practice, SMEs need a lot of hand-holding to transform themselves and the economy into an AI-driven productive and innovative economy. Government documentation, certification and processes stand in the way of all digital transactions. AI cannot work on non-digital documentation. Thus, governments must improve the digital public infrastructure in order to take advantage of the AI revolution. SMEs need help from chambers of commerce, universities and government agencies to move rapidly into the AI age.

Malaysia faces this huge change management phase that can easily get stuck in a thousand forms of bureaucratic red tape. If Malaysia wants to have the next unicorn, helping more SMEs on their AI journey must be top priority.

Tan Sri Andrew Sheng writes on 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.