Artificial Intelligence is certainly, a major technological breakthrough of this century, with the potential to basically reshape how humans live, work, and innovate. Shipping a software feature that once took months of development is now, just a matter of a weekend, a well-thought-out plan and the AI tools’ leverage. Cinematic scenes that once required cameras, actors, and production crews can now be generated from just a thoughtful prompt. Mathematical problems that remained unsolved for decades are now being cracked by AI.
Today, ideas are what matter more than anything, specially in software development. For any SaaS startup that plans to acquire users gradually, launching an MVP is matter of some months after the leverage of AI tools. For software engineers with prior experience, AI simply amplifies their efficiency by multiple times. The applications of AI are virtually limitless spanning across almost every field imaginable.
With such exceptional capabilities, AI has captured the attention of entrepreneurs, corporations, and investors worldwide. Companies are investing hundreds of billions of dollars in developing AI models, building data centers, manufacturing advanced chips, and expanding the infrastructure needed to stay ahead in this technological race. This has been clearly reflected in the capital markets, where several prominent AI companies, and infrastructure providers have experienced soaring valuations, along with the massive capital investment commitments. But this raises an important question:
There are several reasons based on which I believe, the current AI boom deserves closer look. Here are some of the key concerns that I think are worth examining.
1. The Valuation–Fundamentals Disconnect: Valuing the future, Ignoring the present
AI is certainly capable of transforming industries, but the question is whether its future revenues can justify today’s sky-high valuations. As capital spending continues to rise, are investors betting on sustainable growth, or are market expectations getting ahead of financial reality?
Revenue growth, profitability, and capital expenditures provide a clearer picture of actual financial performance. As expectations continue to rise, investors should remain concerned of the gap between present fundamentals and future speculations, and the risks that may emerge if growth is lower than expected.
1.1 The Valuation-Revenue Gap
The core problem is not that AI lacks potential., but if the soaring valuations can be justified by actual revenues and profits. Michael Burry, the investor who predicted the 2008 housing crash, has questioned the sustainability of the AI boom, warning that massive infrastructure spending and understated depreciation costs could make earnings appear stronger than they are. As he observed, “This massive spending is running ahead of the market’s understanding of how to account for it.”
The gap between AI companies’ valuations and their actual revenues remains a major concern. Investors are betting not just on growth, but on whether that growth can be fast, large, and profitable enough to justify massive investments. The question is no longer just how big AI can become, but whether it can earn enough to justify what it is worth today.
As of October 15, 2026, NVIDIA traded at 17.3 times its TTM revenue, while Meta (8.2×), Amazon (3.5×), and Oracle (5.7×) were valued at lower multiples. However, growing investments in AI infrastructure and rising capital expenditures raise questions about future profitability and returns.
Based solely on completed pre-IPO private funding valuations, OpenAI’s $852 billion valuation represents approximately 17× annualized revenue, while Anthropic’s $965 billion valuation implies 14.4×. Looking ahead, Anthropic’s potential 2026 reported IPO valuation of $1.5–2 trillion would imply 22.3–29.8× annualized revenue.
NVIDIA, OpenAI, and Anthropic have considerably high valuation-to-revenue ratios compared with established technology companies such as Meta and Amazon, which have proven earnings and stronger financial track records. In an increasingly competitive AI market, failure to meet revenue growth and profitability expectations could put significant pressure on these valuations.
Are the skyrocketing AI valuations justified by actual financial performance?
1.2 CapEx vs Revenue: Rising Investments, Uncertain Returns
The rapid expansion of AI infrastructures require enormous capital investment, in data centers, GPUs and supporting infrastructures, often before these investments generate returns. Comparing the capital expenditures with revenue provides context for evaluating their potential long-term returns.
In its September 24, 2026 report, Goldman Sachs estimated that five major U.S. technology companies would spend approximately $800 billion in capital expenditures in 2026, rising to $1.2 trillion in 2027. As of Q2 2026, their annualized cloud revenue was only about $70 billion above pre-AI trends, compared with the estimated $300 billion in annual AI revenue needed to break even.
Gartner estimates that AI data center owners would need $8.5 trillion in cumulative revenue by 2030 to cover spending, financing costs, and a 12% return on investment. Yet AI revenue attributable to these data centers is projected to reach only $3.6 trillion, leaving a potential $4.9 trillion gap.
CoreWeave illustrates the financial risks: despite a $104 billion revenue backlog, the company faces substantial debt, rising capital expenditures, and continued losses. Contracted future revenue does not guarantee profitability, particularly when infrastructure investments must be made years before returns materialize.
AI companies are valued very high than their actual revenue, so slower growth could sharply lower their valuations. Huge data-center may not earn enough revenue to cover its costs. If growth is slower than expected, investors could lose a lot of money.
2. Circular Financing: the AI money loop
Circular financing occurs when capital moves among a group of companies and returns as revenue through internal purchases of goods or services within that same group.
How does Circular financing work?
The primary objective of the circular financing is to accelerate mutual growth. AI infrastructure requires substantial upfront capital expenditure (Capex), and these partnerships help companies secure capital, customers, and computing capacity.
For example, a GPU manufacturer invests in an AI company. That company uses the capital to purchase computing services from cloud providers, which in turn expand their data centers using GPUs from the same manufacturer. This creates a reinforcing cycle: the AI company gains computing power to train its models, the cloud providers secure contracts, and the GPU manufacturer benefits from growing hardware demand. Ultimately, this can help maintain healthy balance sheets and revenue and support higher market valuations as stock prices rise.
When Investors Become Customers: Is AI Funding Its Own Boom?
However, this raises a fundamental question: Is the growth driven by genuine end-user demand, or sustained to a significant extent by capital circulating among the same companies? While the transactions are real, circular financing raises concerns about if such arrangements are inflating reported growth and market valuations beyond what underlying demand and profitability can sustain.
The figure below illustrates circular financing among major AI companies.
Sources & Disclaimer: SEC filings, company disclosures, Reuters, and Financial Times (as of October 19, 2026). Figures include investments, contractual commitments, backlogs, and estimates, not necessarily realized cash flows. The diagram illustrates financial interdependencies; amounts are not directly comparable or additive.
What happens if the cycle breaks??
The biggest risk arises when AI demand fails to meet the expectations. The warning signs are already visible.
If AI companies reduce spending, the chain reaction unfolds in sequence. First, training runs slow down, GPU demand falls, and neo-cloud utilization drops. Cloud providers are left with expensive data centers, fixed debt, rapidly depreciating physical assets, and leases that do not shrink when cloud rental income falls. A September 2026 VentureBeat survey found that 81% of surveyed enterprises that reported GPU utilization operated at 50% capacity or less (based on survey of 71 enterprises running AI in production), while 31% reported utilization of 25% or less. The capacity has already been built. However, the revenue is not guaranteed. This raises serious questions about the financial sustainability of the current AI infrastructure boom.
When Growth Was Financed: Lessons from the Dot-Com Boom
In The 2000 Dot-Com boom, Cisco loaned money to telecom companies to buy its equipment, with around 1.9 billion in outstanding loan commitments by 2001. When the market collapsed, some customer failed to pay the loan, compelling Cisco to prepare for potential losses. Cisco also recorded a $2.25 billion inventory charge as demand fell.
The internet continued to evolve, but investors paid the price.
The internet technology didn’t fail. The issue was that the same capital was moving between the vendor and the customer until credit tightened and the loop couldn’t continue. Today’s AI deals use equity stakes and cloud credits instead of direct loans, but the logic is the same.
Circular financing can certainly support business expansion, but it also creates financial dependencies among the participating companies. If one company runs into trouble, the effects can ripple through the entire chain. The real test is whether this growth is sustained by genuine end-customer demand and profitability, rather than by continued investment within the same ecosystem.
3. The AI Price War: China cutting Costs, reducing Dependency
For complex tasks such as large-scale software development, advanced reasoning, and autonomous programming, American AI models like OpenAI’s ChatGPT and Athropic’s Claude, remain among the strongest options. However, the competition is narrowing. On July 17, 2026, Chinese AI company Moonshot AI launched Kimi K3, a 2.8-trillion-parameter open-weight model designed for advanced reasoning, coding, and long-running autonomous tasks. The model performed competitively against leading American systems and surpassed them on selected benchmarks, including programming evaluations.
However, not every business requires such powerful and costly models. Routine tasks such as customer support, chatbot automation, document processing, information extraction, email classification, and basic workflow automation can often be handled by smaller, less expensive models. This is where open-weight AI models are becoming a serious competitive threat. Unlike proprietary models, open-weight models make their trained parameters publicly available, allowing businesses to download, customize, and operate them on their own infrastructure instead of relying entirely on paid APIs.
How Open-Weight AI Is Reshaping Margins
The benefits are already visible in the businesses. AirBnB has plugged in Alibaba’s Qwen models for the customer support system. More importantly, Shopify replaced a proprietary frontier model with a custom version of Alibaba's Qwen3-32B for generating merchant workflow automations. According to Shopify's April 2026 engineering report, the new system was 68% cheaper, 2.2 times faster, and more accurate than the model it replaced.
The US-based coding platform Cursor used Moonshot AI's Kimi K2.5 to develop its Composer 2 model, while AI startup Lindy switched most of its model traffic from Claude and Gemini to DeepSeek V4 Flash, reporting around a 90% reduction in inference costs on migrated workloads. Likewise, Coinbase reduced its overall AI spending by nearly 50% through the use of Chinese AI models and other optimizations.
Legal AI startup Harvey, valued at $15.5 billion, reportedly saw its gross margin fall from 50% to minus 50% by June 2026, largely due to rising AI usage costs. In August, it introduced Harvey Tenet, built on Moonshot AI's open-weight Kimi K3, helping restore positive margins alongside other cost-saving measures.
China's Leap for Chip Independence
The US export ban on advanced NVIDIA chips has pushed China to invest massively in domestic semiconductor development. China's $47 billion Big Fund 3.0 and a reported $295 billion AI infrastructure plan aim to reduce dependency on foreign technology, with a goal of producing 80% of AI chips domestically. Chinese chipmakers are also increasing research spending, while China's data center capacity is projected to grow from 32 GW in 2025 to 40 GW in 2026, potentially exceeding 60 GW by 2030. Recently, China also launched a 24 MW wind-powered underwater data center near Shanghai, which uses seawater for cooling and reportedly reduces electricity consumption by 22.8%.
Restricting the Chips, Reshuffling the Deck?
Meanwhile, Huawei is escalating its Ascend AI chip development, with new chips expected in 2027 and systems designed to connect up to 4,096 processors. Its Ascend 910C reportedly delivers around 80% of Nvidia's H100 computing performance in certain comparisons. According to IDC, Nvidia's share of China's data centre AI chip market fell from 55% in 2025 to 34% in the first half of 2026, while Huawei captured 28%. Chinese companies also expect to increase domestic chip purchases from 30% to 46%, while Morgan Stanley forecasts China's AI chip self-sufficiency to reach 70% by 2029, up from 42% in 2025. This growing domestic competition poses a significant challenge to Nvidia's position in China.
In short, Chinese open-weight models are driving down AI costs and squeezing margins, while US export restrictions are accelerating China’s domestic chip push. Together, cheaper models and growing chip self-sufficiency threaten Nvidia’s China position and pressure American AI providers’ pricing power.
4. The infrastructure bottleneck: Billions Invested, Constraints Unresolved
The AI boom is now also a race to build physical infrastructure. Microsoft, Alphabet, Meta, and Amazon spent around $400 billion on capex in 2025, which is more than entire global telecommunication industry spends annual. This figure is projected to rise to $700 to $745 billion in 2026. However, capital only doesn’t guarantee these facilities can be built on schedule.
America is facing growing electricity and infrastructure constraints, with data centres accounting for roughly 50% of electricity demand growth in 2025. Projects in Arizona, Northern Virginia, and Memphis have faced public opposition of over water consumption, noise, air pollution. Europe faces similar challenges: data centers consumed 23% of Ireland’s metered electricity in 2025, while Amsterdam has restricted new large data centers until at least 2030.
Oracle’s Project Jupiter: When infrastructure Meets reality
Oracle's CapEx stands at $75.5 billion, compared with its Trailing Twelve month revenue of $71.78 billion. In other words, Oracle is spending more on infrastructure than it earns in a year, its CapEx equivalent to 105.5% of its revenue.
Oracle’s ambitious “once-in-a-generation” project, Jupiter is a $18 billion AI data center in New Mexico, planned to deliver 2.4 GW of power capacity to OpenAI. However, this project has faced setbacks involving energy supply constraints, regulatory delays, and local protests over water consumption and environmental issues, which forced Oracle to issue a force majeure notice in September 2026, seeking the contractual protection against delay.
The project uses its own gas-powered fuel cells instead of waiting years for a grid connection, but that created a new problem: it now depends on a gas pipeline that keeps getting delayed. Without gas, the fuel cells cannot run and the huge campus has no backup power, so the project is blocked by a pipeline, not by chips or money.
The credit market has already noticed. Oracle’s 2065 bonds crossed an 8% yield in August 2026, higher than the average for B-rated debt and its default insurance (CDS) spread widened sharply after the OpenAI deal. This ties to Oracle’s July 2026 downgrade to BBB-,one notch above junk and a roughly 60% stock drop from its high to a 52-week low.
Oracle’s experience shows the financial and operational risks of AI infrastructure investments. If AI demand and revenue growth fail to meet expectations, companies could face severe financial losses, mounting debt, and billions in investments that could vanish into thin air.
The AI infrastructure boom faces challenges beyond capital investment. Electricity shortages, regulatory delays, environmental concerns, opposition from environmental activists, and rising borrowing costs could slow down projects and increase expenses. Meanwhile, in this era of rapid technological development, depreciation and technological obsolescence of expensive GPUs could further threaten returns. The primary question is whether these facilities can become operational on time and generate enough revenue to justify their massive costs.
“The biggest issue we are now having is not a compute glut, but it’s power” - Satya Nadella
5. The Conclusion: Technological Success, Financial Uncertainty
There is no doubt that AI will continue to transform the industry and improve efficiency and productivity. However, technological success does not necessarily guarantee financial success for every company or investor involved. The growing gap between valuations and financial fundamentals, circular financing among specific companies, growing competition from cheaper AI models, and increasing infrastructure constraints all raise serious questions about the sustainability of the current investments made in AI.
When the music stops, the models could still dance, but valuations may not.
In the end, considering all these above mentioned factors, AI could follow a similar path to the internet: real and lasting change, but a painful adjustment for investors. The technology can succeed while many investments still fail to earn back their cost. The main risk is timing. If valuations and spending stay far ahead of profits, the market may correct sharply first. That correction would not stop AI. It would decide which companies and investors can afford to stay in the game.
Thank you for reading!