Hermann Hauser, co-founder of Acorn Computers and a central figure in the creation of Arm Holdings, has spent four decades observing how technology waves emerge, peak, and reshape industries. In a recent interview, he offered a carefully balanced view of the current artificial intelligence boom: it is a genuine revolution, but it will also be a rollercoaster. His perspective carries unusual weight, given that he helped design the chip architecture now found in most of the world's smartphones and has spent the last two decades backing European deep-tech startups through his venture firm Amadeus Capital Partners.
Hauser sat down with CNBC's The Tech Download podcast, and the full conversation ran to 40 minutes as a video episode. The discussion spanned AI, semiconductors, quantum computing, and European sovereignty. The written version of that conversation, published on 14 August, distills his arguments into a series of pointed observations. Rather than simply calling AI a bubble or dismissing the skeptics, Hauser separates the technology question from the pricing question. He argues that AI will create more value than probably any other technology revolution ever seen, but that some valuations have clearly gotten ahead of themselves.
Not a bubble call, and not a defense either
Hauser's starting point is far from the typical tech skeptic. He describes AI as a revolution that will generate enormous economic value, perhaps more than any prior technological shift. Yet he immediately adds a qualifier: the journey will be a rollercoaster. That single word captures his view that while the long-term trajectory is upward, the path will be marked by sharp drops, overcorrections, and painful resets.
On the subject of prices, Hauser is specific rather than sweeping. He does not claim that the entire AI sector is overvalued. Instead, he points to structural risks that could trigger a correction. His primary concern is a recent wave of circular financing deals, where chipmakers take equity stakes in AI labs that then use the proceeds to buy chips from those same investors. This self-reinforcing loop can inflate revenue figures and create the appearance of robust demand, but it also concentrates risk in a way that could amplify a downturn.
Why circular financing matters
The Bank for International Settlements flagged this pattern in June, warning that an AI bust could hit credit markets as hard as the 2008 financial crisis. The mechanism is straightforward: if AI labs are buying chips with money that comes from chipmakers, the revenue is not entirely exogenous. A slowdown in AI investment could cause a cascade of losses that ripple through both the tech sector and the financial system. Hauser is not predicting that outcome, but he argues that the largest players are well positioned to survive a reset. OpenAI and Anthropic, in his view, hold significant capital reserves and should withstand turbulence even if expectations are recalibrated around them.
That leaves an obvious question that Hauser does not directly answer. If the leading AI labs survive a reset, someone else absorbs the pain. The circular deals are the most likely point of impact. Chipmakers that have invested heavily in AI startups could see the value of those stakes decline or the expected purchase orders evaporate. Smaller players without the same capital buffers would be more vulnerable. Hauser's position, therefore, is narrow and unusual: the revolution is real, prices are not all real, and the biggest names are not the fragile part of the system.
Set against other tech leaders' views
His comments contrast sharply with Masayoshi Son, the chief executive of SoftBank Group. Son told shareholders in June that calling AI a bubble is an insult, and he has described such commentary as blasphemy. Hauser occupies a more nuanced middle ground. He agrees that the underlying technology is transformative but insists that the market's pricing of specific companies can become disconnected from fundamentals. This dual stance is harder to compress into a headline but offers more practical guidance for investors and policymakers trying to separate signal from noise.
The distinction between the technology question and the pricing question is central to Hauser's thinking. He does not believe that AI is overhyped in terms of its potential impact. He does believe that some companies have been valued as if that potential has already been realized, which creates room for disappointment. This is not a contrarian position for its own sake; it is a disciplined attempt to evaluate the industry on multiple timescales.
The chip architecture argument is the freshest part
Hauser co-created the architecture that powers most of the world's mobile devices, so his views on the next wave of chip design carry particular weight. He argues that AI is forcing a fundamental rethink of computer architecture itself. The pressures are practical and urgent. AI models are expensive to run, chips generate enormous amounts of heat, memory costs are rising, and bottlenecks have appeared throughout the industry. These constraints are not merely engineering problems; they are economic and environmental limits that will shape the next decade of computing.
Hauser points to two responses that aim to address these pressures: in-memory computing and photonic computing. Both approaches try to reduce the energy and time spent moving data between processors and memory. In traditional computer systems, data shuttles back and forth between the processor and memory, a journey that consumes power and creates latency. In-memory computing moves the arithmetic into the memory cells themselves, eliminating much of that travel. Photonic computing uses light instead of electrical signals to transmit data, offering the possibility of much higher bandwidth and lower energy consumption.
The significance of these changes, in Hauser's words, could match the architectural breakthroughs that allowed Arm to challenge the established chipmakers decades ago. Arm succeeded by focusing on power efficiency at a time when the industry was optimizing for raw speed. The constraint of battery life created an opening that Arm exploited, and the same kind of constraint-driven opportunity is now emerging in the AI era. Hauser admitted that he never thought AI would produce such a fundamental shift in computer architecture, yet that is precisely what he sees happening.
Money is already moving toward photonic computing
The photonics half of this argument is not theoretical. Nvidia, the dominant supplier of AI chips, has spent $6.5 billion across photonics companies in just three months. The goal is to replace copper wires with light inside AI data centers, where the sheer volume of data being moved between thousands of chips has become a critical bottleneck. This is a case of a constraint being bought out rather than engineered around. The problem that Hauser describes already has a price attached to it, and major industry players are placing large bets on the solution.
In-memory computing is further behind in terms of commercial deployment, but the concept is gaining traction. Instead of moving data to a processor and waiting for the result to come back, the arithmetic is performed where the data is stored. This approach has the potential to dramatically reduce the energy cost of AI inference and training, which has become a growing concern as model sizes expand. Both photonic and in-memory computing are attempts to address the same underlying target: the cost of moving data, rather than the cost of the mathematical operations themselves.
Arm's history provides a useful precedent. The company won on power efficiency at a moment when the industry was focused on raw performance. The constraint of mobile device battery life created the opening that allowed Arm to become the dominant architecture in phones. Hauser sees a similar dynamic at play in AI. The constraints of energy consumption, heat dissipation, and memory bandwidth are reshaping the industry, and the companies that can solve those problems will be the next major winners.
Europe can build, but cannot scale
When the conversation turns to Europe's position in the global tech landscape, Hauser is direct in both directions. He believes that European companies have the innovation and skill to compete with the United States and China in deep technology. The failure, he argues, comes at the next stage. European startups struggle to grow from small, innovative firms into genuinely global competitors. This is not a new diagnosis, but coming from an investor who has spent years funding that attempted jump, it carries particular weight. Amadeus Capital Partners exists to help European companies make that leap, so Hauser's frustration reflects personal experience as well as structural analysis.
He connects this scaling problem to a larger concern that he keeps returning to throughout the interview: technological sovereignty. Hauser worries that Europe depends on foreign suppliers for critical technologies, ranging from AI models to semiconductor design software. That second item is often overlooked. Chip design software is a small market with only a handful of suppliers, but it sits upstream of everything else in the semiconductor industry. Without access to that software, it is nearly impossible to design advanced chips, regardless of how innovative a European company might be.
The recent moves in that sector demonstrate the risk. Synopsys, one of the leading electronic design automation companies, pulled back from chip fabrication software in July to focus on higher-margin AI design tools. This shift means that even the limited options available for chip design may become more constrained over time. Hauser's warning is that dependency on a few suppliers, especially those outside Europe, leaves the continent vulnerable in an era of geopolitical tension and export controls. Export controls turn a simple supplier relationship into a chokepoint. A tool that can be purchased today is not just a purchase; it is a dependency that can be weaponized tomorrow.
Hauser's caveat is as firm as his warning. He believes that close cooperation with allies is essential, and he is not arguing for a break with the United States. But he insists that partnership and dependence are different arrangements, and Europe has been treating them as the same. The distinction matters because dependence carries risks that partnership does not, especially when trade policies, export restrictions, and political tensions can shift quickly.
The line that captures a broader fear
His conclusion on this topic was blunt and quotable. Europe should preserve its partnership with the U.S., but it should not become a technology colony of the U.S. That phrase encapsulates the fear that Europe's technological future is being shaped by decisions made in Washington and Silicon Valley rather than in European capitals. Brussels has already begun to act on that fear. The European Commission proposed a sovereignty package in June that would restrict U.S. cloud providers from handling sensitive government data, alongside a Chips Act 2.0 aimed at boosting domestic semiconductor production. These are steps in the direction Hauser advocates, but he would likely argue that they are not enough.
His position is not one of simple anti-Americanism. He acknowledges the importance of the transatlantic alliance and the value of cooperation on technology and security. But he draws a clear line between partnership, which is a relationship between equals, and dependence, which leaves one side vulnerable to the other's decisions. Europe, in his view, has been treating these two arrangements as interchangeable, and that is a dangerous mistake.
The broader lesson from Hauser's interview is that the AI revolution must be evaluated on multiple dimensions. The technology itself is transformative and will create enormous value, but the market around it is subject to the same cycles of overvaluation and correction that have hit every previous technology wave. The physical infrastructure supporting AI is changing in fundamental ways, driven by energy and bandwidth constraints. And the geopolitical landscape is being reshaped by the concentration of critical technologies in a few players. These are not separate issues; they are interconnected forces that will determine who benefits from the AI era and who is left behind.
What would settle the questions Hauser raises? First, the architecture claim is testable. In-memory and photonic computing are already being developed and deployed, and Nvidia's investment in photonics shows that the market is treating these approaches seriously. Over the next few years, it will become clear whether these technologies can deliver the dramatic efficiency gains that their proponents promise. Second, the question of whether a European company can make the jump from startup to global competitor is an open one. That is a test of growth capital, management talent, and market access, not just research capability. Third, the rollercoaster that Hauser expects will inevitably arrive, and the circular financing deals he flagged are where the evidence of trouble will show up first. If a reset occurs, the resilience of the largest AI labs and the vulnerability of smaller players will become apparent, and the market will have a clearer picture of who was building real value and who was merely participating in the hype.