AI investment risks: what Italian SMEs should weigh first

AI investment risks deserve a sober reading before any purchase decision. OpenAI generated USD 3.7 billion in annual revenue in 2024 and closed the year with USD 5 billion in losses, then carried a valuation of USD 157 billion. The case is not isolated. AI investment is on track to absorb USD 3 trillion by 2029, three times the dot-com bubble, while the underlying economics tell a different story. Bill Gates says the situation reminds him of 2000, Goldman Sachs speaks of an imminent correction, and the European Central Bank flags systemic risks. For an Italian SME, the question is how to invest without being swept away when the market cools.

The cracks in the triumphant narrative

Bill Gates has been direct: the current state of AI reminds him of the dot-com bubble. Bret Taylor, chief executive of OpenAI, concedes that many companies will fail. The chief executive of Goldman Sachs prefers the word correction and avoids the word bubble, and the substance does not change: even at the top of the industry, there is awareness that current valuations fail a serious analysis.

Important differences separate this cycle from 2000. The so-called Magnificent Seven trade at 26.8 times earnings, against 52 times during the dot-com bubble and 67 times for Japanese banks in the 1980s. Companies such as Nvidia generate real profits rather than slide projections, and the practical use of AI in many sectors is verifiable.

The warning signals keep accumulating. Around 40% of the total capitalisation of the S&P 500 sits in eight technology stocks. Active funds have cut their exposure to AI equities to a five-year low, a sign that professional investors are taking profits. OpenAI, despite a valuation above USD 500 billion, loses money on every subscription it sells and burns cash at a rate it cannot sustain. The ECB warns that the concentration of AI investment in technology stocks is draining the liquidity reserves of the system. That is arithmetic rather than alarmism.

What vendors leave out of the pitch

The AI market carries an evident contradiction: stellar valuations on one side, unsustainable business models on the other. Development costs are enormous. Training advanced language models requires GPU investments measured in hundreds of millions, the energy drawn by dedicated data centres grows exponentially, and specialised staff cost six figures a year. At the end of that costly chain, revenue does not cover expenses.

Security issues add a further layer of risk. Documented breaches, exposure of sensitive data and architectures that are not yet mature leave companies that hold client information exposed to legal and reputational consequences that are hard to quantify. Vendor lock-in completes the picture: proprietary formats, non-standard APIs and closed ecosystems create dependency, and when a supplier fails or rewrites the terms, migration becomes complex and expensive. With many AI companies likely to disappear, as the chief executive of OpenAI himself concedes, this scenario belongs in the risk assessment rather than in the footnotes.

Europe's specific exposure

If the AI bubble deflates in the United States, Europe takes a double shock. The continent owns no proprietary AI giants. It depends on foreign capital, foreign cloud infrastructure and imported GPUs. A withdrawal of global capital would hit both funding and supplier stability at the same time.

The European Union has planned EUR 50 billion of public investment and EUR 150 billion from private sources, through the European Digital Innovation Hubs, AI gigafactories and programmes such as Digital Europe and Horizon Europe. The figures matter, and they remain marginal against the trillions moving across the Atlantic. The gap describes an ecosystem that follows rather than leads.

For Italian SMEs, that structural weakness becomes an operational risk. A manufacturer investing in AI for production optimisation, a logistics company adopting predictive fleet management, a law firm implementing document analysis: each is betting on the stability of vendors operating in an overheated market. The same applies to agri-food, where AI could improve traceability, quality and efficiency, to tourism, with personalisation and revenue management, and to architecture, engineering and real estate, where generative design and predictive analysis are attracting interest. A purchase made with the wrong vendor, at the wrong moment and without sustainability guarantees, leaves systems nobody can maintain, data nobody can access and costs that surface only later.

The signals worth reading

Goldman Sachs, not a group of anti-capitalist pessimists, has published an analysis that raises serious doubts about the return on AI investment. The bank notes that the billions being spent have not yet produced applications that reach the mass market, and that a visible gap separates promises from delivery.

Active funds, managed by professionals paid to beat the market, have reduced their exposure to Nvidia, Apple and the rest of the AI leaders to the lowest level in five years. When professional money steps back, the moment rarely favours new entrants, and the pattern of extreme concentration repeats one already seen in 2000. OpenAI is the clearest case study: valued above USD 150 billion, losing money on every transaction, dependent on continuous funding rounds, carrying documented security vulnerabilities and conceding that the current business model is not sustainable, while still attracting enormous investment. The distance between economic fundamentals and market valuations is the definition of a speculative dynamic.

What it means for a decision today

Italian SMEs sit in a delicate position. AI offers real opportunities in efficiency, automation and competitiveness. The market is heated, valuations are stretched, and many suppliers may not survive a correction. Ignoring AI is not a strategy, and investing blindly on hype is dangerous. The middle path requires method, prudence and the ability to evaluate critically, which means distinguishing a real technology shift from financial speculation and identifying which vendors rest on solid foundations.

Practical tools exist to lower the risk: public incentives that cover part of the investment, qualification frameworks for vendors, and methods to calculate the actual return. The first step is awareness. Gates, the chief executive of OpenAI, Goldman Sachs, professional funds and the ECB are all pointing in the same direction. An investment made today without evaluating scenarios, without quality data and without a plan that includes negative outcomes is a gamble with company resources, and SMEs rarely have the margin to absorb a wrong bet.

A practical reading before investing

For an SME, evaluating AI starts with understanding which processes need automation, which data is actually available, which responsibilities stay internal and which risk comes with a given vendor or platform. A digital and operational assessment reads those elements before the investment: goals, tools already in place, data quality, organisational constraints, risks and priorities. From that reading comes a roadmap that introduces automation where it produces measurable value, instead of experiments disconnected from daily work.

When the analysis confirms a concrete need, the path can move toward software development and automation, or toward public funding programmes that can support the investment. AI will change how businesses operate. The difference between that shift and unsustainable valuations is the capacity to tell them apart and to act on the distinction.

Investment decisions also involve the losses that are insurable, and cyber insurance for SMEs is one of the costs that belongs in the calculation.