Are We Back in the Tulip Growing Business?

AI’s newest startups are attracting billions before proving they can build a business. Is this the next technological revolution or another speculative mania?

In seventeenth-century Holland, tulip bulbs became objects of extraordinary speculation, with prices soaring not because the flowers had become more beautiful, but because buyers believed someone else would pay more tomorrow. Four centuries later, Silicon Valley appears to have rediscovered horticulture. The bulbs are artificial intelligence companies, the gardeners are venture capitalists, and billions of dollars are being planted in the expectation of spectacular returns.

The latest crop is particularly exotic. Known as neolabs, these startups typically consist of small teams of elite AI researchers backed by substantial venture capital, pursuing breakthroughs that might allow them to challenge established frontier laboratories such as OpenAI, Google DeepMind and Anthropic. Twenty-nine such ventures have reportedly raised more than $23 billion, although fewer than 35% have released a commercial product and only slightly more than half have developed technology that other businesses can use.

In most industries, such a disparity between funding and commercial achievement would invite uncomfortable questions about valuation. In Silicon Valley, it appears to have become part of the investment narrative.

Welcome back to the tulip-growing business.

The Price of a Promise

Consider Safe Superintelligence, which has reportedly raised $8 billion to pursue the development of safe superintelligence, a mission the company describes as its entire product roadmap. Investors are effectively financing an extraordinarily ambitious scientific objective without the conventional assurances of revenue projections, a customer acquisition strategy or a demonstrated path to profitability.

Venture capital has always financed uncertain futures, and some of the greatest technological advances have emerged from projects that conventional financiers considered speculative. The semiconductor industry, biotechnology and the internet all benefited from investors prepared to tolerate extraordinary uncertainty, while OpenAI itself could scarcely have anticipated the commercial success of ChatGPT. Yet there remains an important distinction between financing uncertainty and abandoning economic discipline: the former enables innovation, while the latter creates the conditions for speculative excess.

The central investment thesis behind the neolabs is that a handful of exceptional researchers can outperform organisations possessing vastly greater computing power, capital and distribution. History offers ample examples of small teams overturning established industries, but artificial intelligence presents a formidable challenge. Training frontier models demands enormous computational resources, sophisticated infrastructure and scarce technical talent, giving the largest laboratories advantages that may strengthen as the industry matures.

The neolabs are therefore betting that intellectual ingenuity can overcome industrial scale. It is a plausible proposition, but plausibility alone cannot establish what investors should pay for the possibility.

When Valuation Becomes a Popularity Contest

According to PitchBook figures, AI companies attracted 83% of nearly $520 billion in venture funding directed towards American startups during the period reported. Such extraordinary concentration raises the question of whether valuations are being determined by prospective earnings or by competition among investors anxious not to miss the next technological giant.

Speculative markets have a tendency to develop their own logic. Investors justify increasingly ambitious valuations by pointing to the participation of prestigious funds, while those funds anticipate that subsequent financing rounds will establish still higher prices. The reasoning becomes circular: companies are considered valuable because sophisticated investors back them, and sophisticated investors back them because they are considered valuable.

This is the greater-fool theory dressed in the language of technological inevitability. The danger lies not in overestimating AI’s commercial potential, which may indeed be immense, but in confusing the scale of a technological opportunity with the economic value of individual companies pursuing it.

The railway and internet revolutions provide useful reminders that transformative innovation and speculative excess can coexist. Both reshaped economic life while destroying substantial amounts of investor capital along the way. A revolution can be genuine even when the valuations attached to it are wildly optimistic.

When the Tulips Are Bought With Borrowed Money

The speculative enthusiasm is no longer confined to venture capital; it has spread into the debt markets financing AI’s enormous infrastructure requirements, introducing risks that extend well beyond inflated equity valuations. CoreWeave, with a reported $72 billion in liabilities, has seen its bond yields exceed 15% as investors question its ability to meet future obligations, while borrowing costs and credit concerns have also intensified around Oracle and SoftBank. With risk premiums on AI-related debt rising more than 50% over six months, lenders are beginning to confront a potentially dangerous mismatch between the enormous sums committed today and revenues that may take years to materialise, if they materialise at all.

Unlike equity investment, where losses are generally limited to the capital committed, debt creates contractual repayment obligations regardless of whether the anticipated technological and commercial breakthroughs occur. What begins as speculative enthusiasm can therefore become a broader financial problem when rising borrowing costs, refinancing pressures and disappointing revenues reinforce one another. The tulip trade becomes considerably more dangerous when the bulbs are purchased with borrowed money.

Innovation Is Not Investment Return

Perhaps the most overlooked risk is that artificial intelligence could transform the global economy without generating attractive returns for many of its financiers. As increasingly capable models become more interchangeable, competition may drive down prices and compress margins, benefiting customers while concentrating economic rewards among businesses controlling infrastructure, distribution or specialised applications.

Commercial aviation offers a useful historical parallel. Few technologies have transformed global transportation more profoundly, yet airlines have frequently struggled to earn attractive returns on capital. The economic importance of an innovation does not necessarily translate into shareholder wealth, particularly when competition transfers much of its value to consumers.

For AI investors, the critical questions concern who will capture the resulting economic value, what competitive advantages will protect those profits and how much should reasonably be paid today for uncertain future earnings. These considerations are easily overshadowed by the excitement surrounding technological breakthroughs, particularly when the possibility of extraordinary growth encourages investors to discount the risks of competition, commoditisation and technological obsolescence.

Two Futures, Very Different Returns

The neolabs face two broad possibilities, each carrying radically different implications for investors.

In the first, a renaissance of small-team innovation produces radically more efficient AI architectures, reducing dependence on enormous computing budgets and allowing smaller laboratories to develop proprietary breakthroughs that established players must license or acquire. Under such circumstances, a handful of today’s seemingly speculative ventures could become extraordinarily valuable businesses.

In the second, artificial intelligence undergoes the industrialisation familiar from earlier technological revolutions, with competitive advantage increasingly determined by capital, infrastructure, distribution and scale. Established frontier laboratories consolidate their positions while smaller competitors struggle to differentiate their technology or finance continuing development.

The eventual outcome will probably combine elements of both, producing a few spectacular winners alongside numerous expensive failures. That pattern is familiar to venture capital, but the scale of funding being committed before commercial winners have emerged makes the present boom particularly consequential.

The industry is increasingly valuing scientific talent as though discovery itself guarantees a profitable enterprise, overlooking the possibility that research may produce knowledge of enormous social value without creating a durable commercial advantage for its financiers. Breakthroughs can be replicated, surpassed or commoditised, and even exceptional technology may deliver disappointing investment returns when acquired at excessive valuations.

The Tulip Test

For investors, the challenge is to distinguish genuine technological promise from speculative enthusiasm without dismissing the possibility of transformative innovation.

A disciplined assessment should establish what a company has demonstrably achieved, whether its technology offers a meaningful and defensible competitive advantage, who will pay for its products and how much additional capital commercialisation will require. Above all, the valuation must offer a reasonable prospective return after allowing for failure, dilution, competition and the time required to generate sustainable earnings.

These questions may seem pedestrian in an industry captivated by the prospect of artificial general intelligence, but they remain central to sound investment decisions. However extraordinary the technology, businesses must ultimately contend with customers, competitors, operating costs and the price of capital.

The same discipline applies to the infrastructure boom surrounding AI. Chipmakers, cloud providers and data-centre developers may enjoy more tangible revenues than speculative research startups, but their valuations depend on assumptions about future demand, capacity utilisation and technological efficiency. Improvements that reduce the computing resources required for advanced AI could alter the economics of infrastructure investments long before those assets have generated their expected returns.

Selling shovels during a gold rush can be profitable, but that does not make every shovel factory a good investment at any price.

The Flowers Are Real. The Prices May Not Be.

Comparing the AI boom with tulip mania should not be mistaken for dismissing artificial intelligence itself. The technology is already transforming software development, scientific research and business processes, and its longer-term economic consequences may prove profound. What remains uncertain is how much of that value will accrue to today’s investors, particularly those paying extraordinary prices for businesses whose commercial prospects have yet to be established.

Transformative technologies are especially fertile ground for speculation precisely because their ultimate potential is difficult to measure. When the future appears almost limitless, investors can rationalise valuations that would seem indefensible in more conventional industries. The promise of extraordinary technological progress gradually becomes a substitute for evidence of sustainable economic returns.

Some neolabs may produce breakthroughs that redefine the industry, while others will disappear, be acquired for a fraction of their funding or discover that scientific brilliance does not automatically translate into commercial success. The distinction will matter enormously to investors, particularly as the industry’s dependence on borrowed capital introduces financial obligations that cannot be deferred indefinitely.

Four centuries ago, Dutch speculators discovered that beautiful flowers could be terrible investments. Today’s venture capitalists and lenders may discover that brilliant technology can be equally expensive, especially when the price of admission reflects an assumption that almost everything will go right.

The question is not whether artificial intelligence will change the world; it almost certainly will. It is whether investors are paying tomorrow’s prices for businesses that have yet to prove they can earn today’s revenues, while borrowing against profits that may remain years away.

That is how a technological revolution can become a very expensive tulip-growing business.