Hermann Hauser: From Microprocessors to Modern AI and How the I.E.C.T. Supports Deep Tech Entrepreneurs

“Little did my co-founders and I know at the time that we would develop one of the world’s most successful processors.” 

Today, Arm’s processors are part of billions of smartphones and countless other devices, ranging from consumer electronics to enterprise servers. Helping to spin Arm out of Acorn Computers in the early 1990s and thereby starting one of the biggest success stories in European deep tech is what Hermann Hauser has become best known for.

In addition, he co-founded Amadeus Capital Partners, a venture capital firm that invests in European high-technology companies, and the Institute for Entrepreneurship Cambridge-Tirol – Hermann Hauser (I.E.C.T. – Hermann Hauser), which has been supporting deep tech entrepreneurs since 2015 through various programs.

The I.E.C.T. summer school is celebrating its 10th anniversary this year and will take place from August 22-28, 2024, in Tirol, Austria, offering workshops, mentoring, and networking for 30 international participants. If you’re a deep tech entrepreneur eager to participate, you can apply here until May 3.

For this interview, we spoke with Herman Hauser about why he became a deep tech entrepreneur, why analog computing might see a comeback thanks to AI, and how chip startups can succeed today:

What Inspired You to Become a Deep Tech Entrepreneur? 

Entrepreneurship seems to run in my family. My father owned and operated a wine business, and I learned a lot from him about entrepreneurship and the entrepreneurial lifestyle.  

When I was studying physics in Cambridge, UK, there was a lot of excitement around microprocessors in the late 1970s, just like today’s excitement around large language models and generative AI. The feeling was similar. All the students were talking about it; there was even a student group called the microprocessor group.

Before, computers were giant, clunky machines that wouldn’t fit into a box and required a clean room. The microprocessor changed everything; suddenly, computers could become small, cheap, and still powerful enough for people to own and play with. 

Like many entrepreneurs, I wanted to make a difference and have a big impact, so I set out to found Acorn, a microprocessor kit company. Little did my co-founders and I know at the time that we would develop one of the world’s most successful processors.

But some of our key considerations were: Does it matter what we are building? Can it become big enough? Could it, at least potentially, change the course of history? For microprocessors, the answer was a clear yes. And it’s still the same if you start out in deep tech today, regardless of whether it’s in specialized AI chips, quantum computing, or synthetic biology.

What Excites You About the Future of Computing?

On a hardware level, quantum computing is the most exciting trend we see today. That’s not just because it’s appealing to me as a physicist but also because it promises to increase the performance of computers beyond what classical computers can achieve and continue Moore’s law in a way we never expected. 

With the rise of AI, the basic architecture of AI processors is also up for grabs again; neither CPUs nor GPUs are ideal for training and operating machine learning models, and there’s huge potential in designing coprocessors specifically for machine learning and speeding up matrix multiplication. In addition to large language models, semantic or propositional knowledge representations seem promising for making AI systems more accurate and robust. 

And then there’s analog computing, which I have a soft spot for, as it’s a really elegant way to represent real-life problems more directly without translating them to zeros and ones and abstract algorithms. Instead, you can program them more directly. 

The reason why analog computing didn’t make it in the past is that analog computers are error-prone and can’t produce results with high precision. But this has changed with AI, which is inherently probabilistic and can work with much lower precision. People have gone from 32-bit to 16-bit, 8-bit, and down to 1-bit precision to represent the parameters of neural networks. Precision doesn’t matter as much as neural networks have billions, or even trillions, of parameters, i.e., the search space for the ideal parameters is so huge that even with lower precision, you can find sets of parameters that yield great performance. 

Low precision was the main disadvantage of analog computers in the past, but it doesn’t matter as much anymore for large AI models. At the same time, analog computers, such as photonic computers, can offer incredible performance, especially for Fourier transforms and matrix multiplications, i.e., the essential math for modern AI applications. 

How Can Chip Startups Succeed Commercially?

Arm has proven that thinking carefully about your business can create as much value as your technology. It has become the largest tech company in the UK, valued now at around $100B, because of the combination of a clever, low-power chip architecture used today by almost all smartphones and its business model around licensing IP. Arm today sells 20x the number of processors compared to Intel, thanks to its licensing model, enabling thousands of companies to use Arm’s technology for their chips. 

Most chip startups today are fabless, outsourcing manufacturing to fabs, but as Arm has shown, you can even be chipless and license IP instead of selling chips. Now, because of Arm’s success, I often get asked about the IP licensing model, and I always tell founders: Don’t try this yourself. 

Arm is the exception to the rule, and it’s tough to make an IP strategy work: You first still need to produce a chip to prove to people that the technology works. Then, you need to attract customers who like to use it in high-volume production, so getting into an IP licensing business takes a lot of time and money.

Going one step further and integrating vertically to offer a better product to customers is even harder. Some people excel at developing new chip architectures, some at implementation, and some at building a complete product for end customers. These are three different teams! You’ll need plenty of people and another magnitude of money to use your chips in-house to develop a superb product others can use.

Most chip startups will be better off selling an excellent chip at a premium. But that’s why I generally say plan your business carefully—it will pay off! 

Who is the I.E.C.T. Summer School for?

It has always been important to me to give something back and support others in building their start-ups. When my cousin Josef Hauser suggested setting up an institute to support entrepreneurs in Tyrol with links to high-tech ecosystems like Cambridge, I was excited. 

Fast-forward to today. The Institute for Entrepreneurship Cambridge-Tirol – Hermann Hauser (I.E.C.T. – Hermann Hauser) has supported 250 alumni who raised more than 200M in funding, and the institute has become a much larger success than I had expected. It’s a great pleasure to spend a few weeks each year with young and bright people in Tirol. 

Each year, we host a summer school designed for spinouts from universities or research labs, as well as for researchers who are committed to developing a deep tech idea into a startup. While we don’t facilitate the pairing of co-founders, participants must already have a project in place. The program unites people from various disciplines, including research and business, to found a deep tech startup.

For us, deep tech refers to technologies that are challenging to develop. These are not merely apps but complex innovations that demand substantial know-how, often involve patents, and are technologically defensible.

When I look at deep tech startups, I always consider three factors—in that order:

  1. The size and growth rate of the market – how big could this become?
  2. The founding team – is it a great team with all the necessary know-how?
  3. The defensibility of the technology – how hard will it be for others to copy it? 

I have often seen A-grade teams with C-grade technology beat C-grade teams with A-grade technology. The I.E.C.T. summer school will help teams become investment-ready, refining their startup idea and product definition and developing their go-to-market and sales strategy.

What Advice Would You Give to a Fellow Deep Tech Entrepreneur?

It’s much easier to start companies today, just given the tremendous support you get from the internet. You can look up most things I’ve just mentioned, especially now that large language models can digest information and answer your questions directly. 

My best advice is to pick one or two great mentors who have done it before. Not necessarily industry veterans but founders whose startup is two to three years ahead of yours, who have just gone through the process and dealt with all the challenges like setting up processes, dealing with lawyers and accountants, and hiring employees. 

Today is the most exciting time I have lived through, especially with AI giving people new superpowers. This is so powerful because it can directly increase productivity in the economy and thereby make everyone better off. Don’t just let time just pass by—do something!

Comments are closed.