Last night, during dinner with one of our AI infra portfolio companies, we had an interesting discussion whether we’re in an AI bubble or not. With everything that’s happening these days - the attention, the hype, the money flowing into AI, the circular economy between Nvidia, OpenAI, Oracle, AMD, etc. - it’s easy to see parallels to the dotcom era in the late 90s.

At the Italian Tech Week in Turin last week, Goldman Sachs CEO David Solomon even predicted a stock market reset would be coming in the next 12 to 24 months following a “big investment cycle” in AI.

On the flip side, there are lots of reasons to believe that it’s different this time, that we’re observing the largest infrastructure build-out of our time that will transform all aspects of our lives. While the dotcom boom was built on potential, today’s AI wave is already creating tangible value - from productivity gains in enterprises to entirely new business models emerging at record speed.

So, are we witnessing another speculative mania or the early innings of a long-term structural transformation? In today’s episode, I’ll share learnings and perspectives from dozens of discussions with AI leaders, from researchers to engineers, managers, investors, and politicians.

Here’s my take of what’s really going on.

First off, let’s dissect what’s really happening behind “the circular economy of AI”. The AI boom is increasingly being powered by an intricate loop of investments, partnerships, and dependencies between a handful of dominant players.

Nvidia, the world’s most valuable semiconductor company, and OpenAI, the most hyped AI startup, sit at the center of this circular economy. Nvidia recently invested up to $100 billion to fund OpenAI’s massive data-center expansion, facilities that will, in turn, be filled with Nvidia’s own chips.

The math behind it? Take $100bn cash from the balance sheet and invest it into any AI company (here OpenAI) that sooner or later needs compute. Put in place an agreement to get a right of first look/refusal on any GPU orders by the AI company. Parts or all of the $100bn come back in form of revenue to Nvidia. Nvidia EV/Revenue multiple hovers around 25x, so $100bn revenue multiplied by 25x drives a theoretical $2.5bn enterprise value increase for Nvidia.

And it’s not only this deal with OpenAI. Nvidia has been doing it for years all across the world - some examples:

One of the smartest strategies and financial engineering I’ve seen out there. And it comes in different flavours and colours, yet always follows the same rational.

Getting back to recent news, not long after the Nvidia investment in OpenAI, OpenAI struck a similar multibillion-dollar deal with AMD, becoming one of its largest shareholders while agreeing to deploy billions of dollars’ worth of AMD GPUs. AMD Share price? Went about 50% up following the announcement.

These reciprocal relationships are incredibly smart strategic moves on the one side, but also blurring the lines between buyer, supplier, and investor on the other side; creating interwoven dependencies. Each deal sends capital spinning around the same closed system: Nvidia and AMD invest in OpenAI; OpenAI buys their chips; cloud providers like Oracle finance and host the infrastructure, then buy more Nvidia hardware to power it all.

Money moves, valuations rise, but little external value is created yet.

The AI ecosystem’s major players are forming what looks increasingly like a self-sustaining financial flywheel. Nvidia invests in companies that buy its chips; those firms, in turn, use Nvidia-powered services to attract new capital, which then flows back to Nvidia through chip purchases. OpenAI’s own network of deals, now reportedly exceeding $1 trillion, exemplifies how capital and capacity are locked together in an expanding loop.

This circular structure has echoes of the late-1990s dot-com bubble, when startups bought each other’s services to inflate growth. The difference is that today’s AI firms do sell tangible products - chips, models, compute - but their spending vastly outpaces monetization. As a result, analysts warn that the market’s perceived momentum may rely more on financial engineering than on sustainable profits.

While skeptics see a speculative bubble fueled by circular investments, there’s another way to interpret what’s happening: the largest coordinated industrial buildout since the internet, and one with tangible, long-term value.

Unlike the dot-com era, today’s AI investments aren’t simply propping up virtual traffic numbers, they’re financing physical and computational infrastructure that will underpin nearly every industry. Data centers, chips, and energy capacity are real assets with measurable utility, not paper valuations.

AI isn’t just a consumer fad, it’s becoming a foundational technology layer. The demand for generative and analytical intelligence is expanding across healthcare, manufacturing, logistics, and defense.

Every major corporation is integrating AI models into operations, automating white-collar tasks, and creating new digital products. The massive capital expenditure from Nvidia, AMD, and OpenAI is less about short-term revenue and more about securing strategic control over computational supply, the scarce resource of the digital age.

In that sense, the circular nature of these deals may be an adaptive feature, not a flaw. The AI economy requires tight integration between hardware, software, and compute infrastructure. Nvidia’s investments in startups that use its chips or OpenAI’s equity-linked partnerships with chipmakers can be seen as vertical alignment, ensuring reliability and speed in a market where shortages could cripple innovation. It’s the same logic that led Apple to control its silicon or Tesla to lock in its battery supply chain.

Moreover, these relationships are creating massive economies of scale. Every new data center lowers per-model training costs and improves efficiency through network effects in compute utilization. The capital recycling inside the AI ecosystem helps accelerate innovation, compress time-to-market, and stabilize prices across the supply chain. Far from inflating artificial growth, this reinvestment loop could be building the rails of the next computing paradigm.

Critics often point to the absence of profit, but transformational technologies rarely start with strong margins. The early internet, smartphones, and cloud computing all went through decade-long gestation phases before monetization caught up.

Already, the productivity impacts of AI are beginning to show: code generation, drug discovery, and enterprise automation are reducing costs and expanding capabilities faster than traditional adoption cycles predict. Case in point, McKinsey cutting headcount from about 45,000 people in 2023 to 40,000 while rolling out roughly 12,000 AI agents.

If these trends continue, the AI buildout could prove anticlimactically rational, a deliberate overinvestment phase designed to front-load infrastructure for a century-defining technology. As with railroads, electricity, and broadband, the early overspend may look excessive in the moment but inevitable in hindsight.

Share this article with others who might benefit.

In short, what looks like circular financing might instead be a high-conviction bet on exponential demand. The same interdependencies that make the AI economy appear fragile also make it resilient: the stakeholders have every incentive to sustain and optimize the system they’re building. The investments by Nvidia, AMD, Oracle, and OpenAI are binding the ecosystem together, ensuring that the hardware, cloud capacity, and model innovation evolve in lockstep.

If AI truly becomes the new infrastructure for human knowledge and productivity, this will not be remembered as a bubble but instead as the great acceleration, when capital, compute, and ambition aligned to build the next industrial platform.

Both sides of the table have valid arguments pro and against an AI bubble. I personally believe both arguments can be true at the same time. Yes, we see financial engineering and a circular economy of AI where money/commitments are handed around, driving valuations through the roof while not creating the equivalent value short-term.

However, we should look beyond the short-term. In the mid- and long-term, I’m convinced that we’re observing the largest infrastructure build-out we’ve ever seen. Critical players in the circular AI system seem to have robust fundamentals; longterm contracts, revenue certainty, healthy margins. With that in mind, I personally see a limited risk of key building blocks defaulting and remain positive.

Will public markets see a reset? Maybe. Yet, I do not expect this to impact the underlying technological progression. Keep in mind: value and valuation are something different, and I’m convinced that the value created will exceed our expectations by far.

Stay driven,
Andre