The archive · Hardware & Devices · Strategic decision · 2023–2026
Flow Computing's 100x-CPU bet: VTT spinout, €4M seed, still hunting a chipmaker
Finnish VTT spinout Flow Computing claims an on-die 'parallel processing unit' can boost any CPU up to 100x; €4M pre-seed, still hunting its first chipmaker.
Flow Computing
What the business is
Flow Computing sells licensable chip IP: a Parallel Processing Unit (PPU) that a CPU vendor integrates into its design to speed up parallel workloads, claiming 2x out-of-the-box gains and up to 100x with recompiled software.
Starting capital：€4M (~$4.3M) pre-seed announced June 2024, led by Butterfly Ventures with FOV Ventures, Sarsia, Stephen Industries, Superhero Capital and Business Finland.
How it started
Flow Computing is a 2023 spinout of VTT, Finland's state-backed technical research centre, where CTO Martti Forsell had spent years on parallel architecture research. CEO Timo Valtonen brought it out of stealth in June 2024 with €4M in pre-seed and a claim that stunned the industry: integrate a Parallel Processing Unit on-die and any CPU can double its performance immediately, reaching up to 100x with recompiled software.
What happened
The proof of concept was an FPGA board and simulations — no silicon had been taped out. Flow presented at IEEE Hot Chips in August 2024, published a white paper, and said it would license the design like Arm, targeting AMD, Apple, Intel, Nvidia and Qualcomm. Independent researchers called the timing right (energy-efficient mobile compute plus AI demand) but warned that chipmakers have long roadmaps and may resist unproven IP. Flow then focused on a PPU compiler and finding a co-design partner; as of mid-2026 it had not announced a design win or a new round.
No ending yet — it is still running.
Background
Flow Computing is a 2023 spinout of VTT, Finland's national technical research centre, commercializing years of parallel-architecture research by CTO Martti Forsell. CEO Timo Valtonen left stealth in June 2024 with €4M in pre-seed funding and a claim at the edge of credibility: an on-die Parallel Processing Unit that any CPU vendor could integrate, doubling performance overnight and reaching up to 100x on recompiled code.
The technology is positioned as a fix for the serial CPU itself. Flow argues CPUs waste cycles because they execute one task at a time, while GPUs are specialized and costly; a PPU adds parallel-execution cores next to the standard cores, handling memory latency, synchronization and thread scheduling so the existing CPU does more per cycle without raising clocks or power.
The proof was an FPGA board and simulations, not silicon — Flow said it would license the design like Arm to AMD, Apple, Intel, Nvidia and Qualcomm, with the first PPU-equipped chip still years away. It presented at IEEE Hot Chips in August 2024 and published a white paper and FAQ, drawing skeptical but engaged coverage: TechCrunch called it 'one of the wildest claims ever heard in silicon engineering.'
Two years later the company is still in the building phase: it has continued work on the PPU core, a compiler and benchmarks, but has not announced a design win, a chip in production, or a new funding round. The entire bet rests on one question Flow cannot answer by itself: whether a major chipmaker will redesign around unproven IP.
What has to be true
- The spinout kept VTT's parallel-processing IP, but Flow's Arm-like licensing model means the company lives or dies on one customer: a chipmaker willing to redesign around unproven silicon.
- The provocative 100x claim earned launch-day coverage from TechCrunch, The Verge and Tom's Hardware, but it also amplified engineering skepticism about any-code, any-architecture speedups.
- FPGA and simulation results assumed a same-speed silicon implementation, a gap between benchmark and product that chipmakers must close at their own expense.
- Timing looked favorable — AI compute demand and mobile power limits — yet analysts warned that vendors with five-year roadmaps rarely rock the boat for unproven IP.
What can be applied
A spectacular claim gets press, but a licensing chip startup lives or dies on one chipmaker adoption. Without a design win, an FPGA demo and a white paper are a research project, not a business.
Aftermath
As of 2026-09-02 Flow Computing is a building-stage, pre-revenue startup in Helsinki. Public funding records list only the €4M pre-seed from June 2024; the company says work on the PPU core, its compiler and benchmarks has continued, announcing new technical benchmarks in May 2025. No commercial chip containing a Flow PPU has been announced and no design win disclosed, so the central bet remains unproven. The stated plan is unchanged: license the design to CPU makers and let them integrate it, keeping Flow a small IP shop rather than a chip manufacturer.
Sources
- Flow claims it can 100x any CPU's power with its companion chip and some elbow grease
- Did startup Flow Computing just make CPUs 100x faster? Here's the white paper and FAQs
- Finnish Startup Wants to Build 100x Faster CPUs
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card