E152: Timo Valtonen: Rethinking the Future of Computing

Hi Timo.

Hey, good to see you. Timo Aaltonen, CEO, co-founder of Flow Computing.

How did you start such a company?

Yeah, it's coming around great. So we started the company two years ago, first of March 24. So now been operational for two years, a couple of months and a couple of weeks. So it's coming around really good. Great progress and it's been really exciting ride during those couple of years and looking forward to the future. Everything is looking quite interesting out in the marketplace for us.

What made you to start the company and how do you see in two years?

We started the company with the notion that the CPUs are the weakest link in computing, central processing units in in the computational stack so at that time there wasn't that much of a market interest but what has happened during the last couple of months has sort of turned around the whole notion about the central processing unit so HNTK-AI is driving that forward so because what happens with the HNTK-AI so it's more complex it's more dynamic so that essentially means that And there's more CPU workloads compared to the GPU workloads. So most of the time it's now GPUs are waiting for the CPU to release the computing and that's agentic. AI is really changing that you have several agents running in parallel. It's really then underlining what we actually are doing. So now the whole market has realized in the AI hardware side of things that CPU actually is the weakest link computation and something needs to be done. And there's kind of a tremendous interest towards what we are building on that side.

Wow, perfect timing. What did change most since your investment? What kind of milestone you're happy to share?

Yeah, the key things that we've been doing is really being able to support the legacy code and showing while doing that also demonstrating CPU performance. So at the moment, we have a development version so we can do 19 exos. So 19 times faster computation when you are comparing a four-core vanilla CPU against one-core, even one-core flow BPU or parallel processing units with this main product. So they're not 19% faster, but 19 times faster.

That's very impressive. So what do your clients say about that?

Proud on that one. So it's really, really showing the power of the CPU. But also what we are looking after is to be able to support the existing workloads, the legacy workloads, so working with benchmarks on that side that are really so promising because as well that we are able to support all the existing codes really deep on the customer conversations, looking to lead customer commercially and pushing ahead with the commercial implementation of the product.

What's your ideal customer?

Yeah, the ideal customers, what we are really looking for are the CPU companies. So the CPU manufacturers, Intel, Qualcomm, IBM, Nvidia, and so on. So that's what we are looking forward. And also what we do is instruction set independence. We're not stuck on any particular instruction set. So we want to be doing so that we'll have a customer in each of the instruction set. And also what is the case with the And with the FlowPPU, Flow Parallel Processing Unit, it is also parametric so we can cater it for