AI Deals: Financial Bubble?
The question of whether the current AI boom represents a technological paradigm shift or is developing into a financial bubble is increasingly being discussed. Investments in data centers, GPUs, and AI startups are reaching historical dimensions, drawing parallels to the Dotcom era. In particular, so-called 'circular AI deals', where tech giants and AI companies finance each other and purchase services from one another, are intensifying this debate.
Source: Internal Representation
AI Market Overview
The current AI boom is a gigantic infrastructure project and, at the same time, a bet on future demand. This leads to discussions about the formation of an AI bubble or whether valuations are backed by real investments and profits. Anyone searching for 'circular AI deals possible financial bubble explanation' will find themselves in the midst of this debate.
The global market for data center infrastructure will reach more than $1 trillion annually by 2030, according to IoT Analytics In 2024 alone, around $290 billion was spent, with Alphabet, Microsoft, Amazon, and Meta contributing nearly $200 billion in CapEx. The Guardian-Analyse '3 trillion dollar datacenter spending spree' adds that banks like Morgan Stanley expect cumulative data center investments of about $3 trillion by 2028, of which around $1.5 trillion could be financed through debt.
At the same time, valuations are spectacular. According to The Guardian, Nvidia became the first company with a market capitalization of $5 trillion, while Microsoft and Apple are around $4 trillion. OpenAI is valued at around $500 billion after a restructuring, including a Microsoft stake of over $100 billion.
Circular AI Deals
Bloomberg describes a network of deals between OpenAI, Nvidia, and AMD. Nvidia has committed to investing up to $100 billion in OpenAI to finance a new generation of data centers. OpenAI, in turn, promises to fill these data centers with millions of Nvidia chips.
Shortly thereafter, a similar agreement followed with AMD: OpenAI intends to purchase billions of dollars worth of AI hardware from AMD over several years, while OpenAI is simultaneously set to become one of AMD's largest shareholders. This is discussed, among other places, in the Harvard Business Review and at VanEck .
Source: seekingalpha.com
Reuters reports in parallel on the cloud company CoreWeave, whose deals illustrate the logic of these 'circular AI deals'. The partnership between CoreWeave and OpenAI now has a volume of up to $22.4 billion, including a new $6.5 billion contract and previous agreements of $11.9 and $4 billion.
Nvidia has acquired more than 5% of CoreWeave. CoreWeave, in turn, buys billions of dollars worth of Nvidia hardware – and Nvidia contractually agrees to purchase unused cloud capacity from CoreWeave.
- Nvidia invests equity in customers like OpenAI or CoreWeave.
- These customers use the capital to place large chip orders with Nvidia.
- The orders appear as revenue for Nvidia, strengthening its stock market value and balance sheet.
- At the same time, the valuations of AI partners increase, boosting the value of Nvidia's stakes.
Bloomberg summarizes this network as a 'web of circular deals' that drives a market of around $1 trillion in AI infrastructure while simultaneously fueling fears that part of this boom is driven more by mutual deals than by genuine end-customer demand.
'Circular AI deals' are thus a concrete pattern where money and revenues circulate within a relatively small circle of companies, thereby supporting both operational figures and valuations. This raises the question of a potential financial bubble.
Comparison with Dotcom Bubble
The comparison between the Dotcom bubble and the current AI hype focuses on the structure: who finances whom, how solid are business models, and how heavily are valuations dependent on stories rather than cash flows.
The Dotcom bubble of the late 1990s was characterized by internet startups with minimal revenues, valued at hardly justifiable revenue multiples. The NASDAQ Composite rose massively until March 2000 and then lost about 78% of its value by 2002, as Investopedia and Goldman Sachs report.
World Economic Forum points out that the current AI phase differs in that enormous sums are now flowing into physical infrastructure such as data centers, power grids, and semiconductor manufacturing – not just stock prices.
At the same time, the sentiment in parts of the market is strongly reminiscent of well-known manias. The WEF points to parallels with the Dotcom era and even the 17th-century tulip mania, driven by the idea that prices can continue to rise as long as a buyer can be found who will pay even more.
Prominent investors such as Ray Dalio, founder of Bridgewater, have explicitly compared the AI euphoria to the Dotcom phase. In 2024, Baidu CEO Robin Li stated that the current situation reminds him of the internet bubble, as the World Economic Forum and the Financial Times report.
Source: YouTube
Source: YouTube
The crucial difference to the Dotcom era: Today, a few dominant corporations with existing profits and cash flows – Microsoft, Alphabet, Amazon, Meta, and Nvidia – dominate instead of hundreds of barely profitable newcomers. This is highlighted by VanEck and The Guardian .
AI Market Valuation
The question of whether the AI market is overvalued cannot be easily answered, as the data is contradictory.
On one hand, there are enormous infrastructure investments. IoT Analytics estimates that the global market for data center infrastructure will reach more than $1 trillion annually by 2030. In 2024 alone, around $290 billion was spent, with Alphabet, Microsoft, Amazon, and Meta contributing nearly $200 billion in CapEx.
Guardian-Analyse '3 trillion dollar datacenter spending spree' adds that banks like Morgan Stanley expect cumulative data center investments of about $3 trillion by 2028, with around $1.5 trillion potentially financed through debt.
Source: bain.com
At the same time, we are seeing valuations that are spectacular even in this context. According to Guardian , Nvidia became the first company with a market capitalization of $5 trillion, while Microsoft and Apple are around $4 trillion. OpenAI is valued at around $500 billion after a restructuring, including a Microsoft stake of over $100 billion.
On the other hand, the same Guardian-Recherche points out that a large part of this infrastructure is based on highly optimistic revenue forecasts for generative AI: Morgan Stanley expects Gen AI revenues to increase from $45 billion in the previous year to $1 trillion by 2028 – growth that is yet to be proven.
Particularly sobering: The Guardian authors refer to an MIT study, according to which 95% of companies have so far realized no financial return from their Gen AI pilot projects.
World Bank-affiliated and other analyses, for example at VanEck , warn of a concentration of risk: if a small group of companies with massive AI investments falters, other investors – such as pension funds – could also see significant losses, which in turn could dampen consumption and growth. The Guardian At the same time, market participants such as
argue that the current expansion is fundamentally different from the Dotcom speculation: the major tech corporations are investing primarily from their current cash flow. Amazon plans CapEx of about $100 billion for 2025, Microsoft around $80 billion, Alphabet about $85 billion, and Meta between $66 and $72 billion – predominantly financed from profitable core businesses. Reuters-Interview The US Federal Reserve reflects this ambivalence: Fed Vice Chair Philip Jefferson stated in a
that the AI-driven stock market rally looks less like the Dotcom bubble because many AI winners are established, profitable companies with solid balance sheets. At the same time, the latest Financial Stability Report indicates that around 30% of surveyed market participants see a bursting of AI optimism as a significant risk.
- Bubble Thesis: Valuations and infrastructure investments are far ahead of realized demand; pilot projects are failing, and some revenues may stem from 'circular AI deals' rather than stable end-customer relationships. (Bloomberg, Reuters)
- Transformation Thesis: The AI boom is an early phase of a long-term structural change, analogous to railroads, electrification, or the internet – with strong fluctuations, but real productivity gains. (World Economic Forum, VanEck)
Sustainable AI Investments
Many investors are looking for 'how to identify sustainable AI startup investments' to have substance in their portfolios. Hard indicators help here, especially against the backdrop of circular deals. A first test is the revenue mix. An AI startup whose core revenues come almost exclusively from one or two major projects with strategic investors moves closer to the 'circular' logic. This is especially true if the same corporation is involved as an investor, places large orders, and purchases infrastructure. Reuters illustrates this with the example of CoreWeave: Nvidia is a shareholder, supplies the company with hardware, and simultaneously commits to buying back excess capacity.
A startup whose recurring revenues are distributed across many customers, industries, and regions – and whose growth does not depend on a single partner – appears more sustainable. Harvard Business Review explicitly warns against structures where revenue and valuation fuel each other without clearly measurable added value for end customers.
Source: user-added
A second test is the cost structure. Reputable AI companies transparently report how much capital flows into GPUs, cloud infrastructure, personnel, and product development – and how gross margins and payback periods result from this. If a startup collects money for GPU costs and little else, but invests hardly anything in product, sales, or integration, it suggests an infrastructure speculation case rather than a productive business model. Analyses like those from IoT Analytics show how expensive GPU-intensive data centers are – and how heavily their profitability depends on high utilization.
Third test: real productivity gains instead of 'demo theater'. The Guardian-Recherche refers to the problem that many Gen AI pilots are purchased but not scaled – according to MIT data, 95% of organizations currently do not achieve measurable returns from their projects. If a startup shows impressive demos but has few solid ROI cases, reference customers, or contract extensions to show, caution is advised.
Harvard Business Review emphasizes that sustainable AI investments are identifiable by whether products are deeply integrated into business processes, reduce error rates, shorten lead times, or demonstrably increase revenues – and whether customers are willing to sign long-term contracts for this, not just experiment short-term with pilot budgets.
A fourth point is financing. The Guardian-Analyse shows how strongly private credit structures and other forms of shadow banking are pushing into AI infrastructure, often with very optimistic assumptions about the value of data centers. An AI startup that relies on high-interest loans and ever-larger funding rounds without cash flows keeping pace carries a different risk than one that grows organically or is only moderately leveraged.
For those who want to see this topic presented compactly, videos like 'Let's Talk About the AI Bubble' offer an illustrative discussion about valuation levels, capital flows, and the question of which business models will be sustainable in the long term.
Implications for CFOs and Regulators
From the perspective of CFOs of large companies, AI is no longer an abstract tech bet but a multi-billion dollar investment program with very concrete balance sheet effects. Studies like the one by VanEck show that hyperscalers are now directing almost all of their free cash flow into AI data centers.
This forces finance chiefs to scrutinize more closely which AI projects are truly strategic and which are predominantly driven by partner deals. A project that only materializes because a major chip provider offers capital or marketing support in return will look different on the P&L than one initiated out of a clear efficiency or revenue logic.
Regulators, in turn, are primarily observing systemic risks. Reuters reports that central banks and supervisory authorities are increasingly paying attention to how much of the AI infrastructure is financed through debt and private credit, and how quickly the underlying assets – particularly specialized GPUs – become economically obsolete. The Guardian-Analyse adds to these concerns.
Das World Economic Forum warns that over-indebted data center projects could, in the worst case, become a kind of 'new shopping mall ruins' – expensive, underutilized infrastructure that no one needs anymore, but whose debts burden the real economy.
For CFOs, this means:
- AI investments should be clearly linked to operational key figures – such as cost per processed ticket, revenue per user, or time savings per process.
- Contracts with AI partners must be structured in a way that not only boosts revenues and valuations in the short term but also generates standalone sustainable cash flows.
- Dependency on individual chip or cloud providers should be actively managed to avoid ending up in a one-sided negotiation situation.
A sober, data-driven perspective is also important because public discourse is heavily influenced by narratives and symbols. Videos like 'Why the A.I. Boom Isn't A Bubble' show how professional investors try to separate the story from the substance – exactly what should be standard practice in controlling.
The AI boom is moving in a field of tension. On the one hand, there are real, gigantic investments: data center CapEx, which could rise to over $1 trillion per year by 2030, multi-billion dollar GPU orders, and physical infrastructure that, in its scale, is reminiscent of railroads, power grids, and the early internet backbone phase. IoT Analytics, VanEck)
On the other hand, there is a growing number of warning signs: circular AI deals that drive revenues and valuations in circles; data center projects based on aggressive debt structures; and studies showing that many companies have not yet achieved a clear ROI from their Gen AI pilots. Bloomberg, Reuters, The Guardian)
The picture resembles the Dotcom episode, but with crucial differences. Back then, a broad, often unprofitable startup landscape collapsed. Today, the risk is concentrated on a few very large, predominantly profitable corporations and specialized infrastructure players. The World Economic Forum puts it concisely: what matters less is whether there is an AI bubble, but what remains of it – in terms of infrastructure, productivity, and genuine, sustainable business models.
Anyone who does not want to get lost in the 'hype fog' as an investor or corporate decision-maker should therefore think along two lines: on the one hand, take the long-term significance of AI as a foundational technology seriously; on the other hand, treat every concrete AI investment like any other major project – with careful risk analysis, clear return expectations, and healthy skepticism towards constructions that look good only because all parties are mutually buying and selling from each other.
The AI hype is real – the question is how much of it is substance and how much is valuation acrobatics. It is precisely this question that decides whether we will be talking about a burst bubble in ten years or an investment wave that laid the foundation for the next productivity surge.