The hidden risks of AI capital: Can a money-burning race bring future growth?

At the beginning of 2026, the global tech giants' race in capital expenditure in the field of artificial intelligence (AI) continued to intensify. Google's parent company Alphabet announced that its 2026 capital expenditure will double to $185 billion, Amazon set a new record with a $200 billion plan, and the combined investment scale of companies like Microsoft and Meta surpassed $650 billion. This capital frenzy, centered on a "computing power arms race," is raising deep concerns in the market about "burning money without achieving growth."

The explosive development of AI technology is driving tech giants into an unprecedented cycle of capital expenditure. As the world's leading chip foundry, TSMC's 2026 capital expenditure plan is as high as $56 billion, with 70%-80% invested in advanced processes to meet the explosive demand for AI chips from clients such as NVIDIA and Google. Of Google's $185 billion expenditure, 60% is for server purchases and 40% goes to data centers and network equipment, with the core goal of consolidating AI technology advantages by monopolizing computing power.

Behind this "heavy investment, light return" strategy is the tech giants' belief in an AI market that rewards the "winner takes all." Google has reached a cooperation with Apple to embed its Gemini model into future iPhones; Microsoft binds OpenAI technology through its Azure cloud service; Amazon AWS competes in the AI inference market with a low-price strategy. The giants believe that only by making massive early-stage investments to build a computing power barrier can they dominate the next generation of AI-driven technological revolution.

However, the patience of the capital market is being tested. A Goldman Sachs report pointed out that from 2025 to 2027, U.S. leading tech giants are expected to invest $1.4 trillion in AI infrastructure, but their average return is far below market expectations. OpenAI, as a benchmark company in the AI field, posted a single-quarter net loss of $11.5 billion in 2025, and its major shareholder Microsoft suffered an investment loss of over $3 billion in the same period. Even more severe, a McKinsey survey showed that nearly 80% of companies deploying AI failed to achieve net profit growth, and 95% of generative AI pilot projects did not bring direct financial returns.

Market concerns about an AI bubble have already been reflected in stock price fluctuations. After Google announced a $185 billion expenditure plan, its stock price once plummeted 7% in after-hours trading; Microsoft and Amazon both experienced single-day drops exceeding 5% after releasing their earnings reports. Investors are beginning to question: when tech giants use large amounts of cash for data center construction instead of shareholder returns, is AI investment turning into a 'Ponzi scheme'?

The core contradiction in the AI field lies in the mismatch between rapid technological iteration and difficult commercial implementation: model training costs are rising exponentially, with each generation increasing costs by 3-5 times; chip depreciation far exceeds that of traditional assets, with NVIDIA's H100 dropping 60% in price one year after release. High investment and fast depreciation force giants to continuously seek financing.

Even more severe is the lack of commercial scenarios: while global AI spending is expected to reach $2.52 trillion in 2026, 60% will flow into infrastructure, and software services will account for less than 40%; end consumers mostly use free functions, and on the enterprise side, aside from cloud services and ad optimization, there is a lack of scalable monetization paths. Demand lagging behind supply exacerbates return concerns.

To break the deadlock, the giants are adjusting their strategies: Google is reducing costs through self-developed TPUs, and Gemini 2.0 Flash has improved inference efficiency by 40%; Microsoft is embedding AI into Office 365 to drive subscription growth, shifting value creation to deep scenario cultivation. At the same time, regulatory pressure is increasing: the European Union is conducting antitrust investigations into giants' AI spending, and the U.S. SEC is requiring transparent disclosure of return data, potentially forcing giants to slow expansion and focus on technological implementation efficiency.

AI technology is experiencing a critical turning point from 'technological explosion' to 'value realization.' The capital expenditure frenzy in 2026 is both a competition among tech giants for future dominance and a stress test of the economic viability of technology. When the tide recedes, those companies that can prove AI investment returns and build sustainable business models will truly win this revolution; while the blind followers may repeat the mistakes of the late 1990s internet bubble. The future of AI will ultimately belong to those who balance rationality and innovation.