Rebuilding AI from the foundations
Real world, real time
AI is blocked
The next generation of AI is blocked by hardware.
We still don't have robots in everyday life.
Not because the world is too hard, but because today's AI is built on the wrong foundations.
It was not designed to learn in the real world, in real time.
Today's AI is matrix models on matrix machines. They're trained upfront on vast datasets, at enormous cost, in gigawatt datacentres with thousands of chips. Then they're frozen and deployed for inference.
Real-world AI has to learn continuously, from its own experience. In real time. On a power budget. On device
The path forwards
New models and
a new machine.
Every leap in AI has needed the right hardware to make it possible. Learning in real time is no different. It needs a new kind of machine, not just new models.
A vertically-integrated
AI & hardware company.
Algorithms and chips, co-designed — with our own toolchain and mathematical proofs. Together, they form a single end-to-end stack for real-world AI.
WHY A NEW PARADIGM
The real world creates inescapable constraints.
Intelligence is bound by them.
Learning from the physical world in real time, on a limited power budget, is unforgiving.
Every joule and every millisecond must count.
This rules out the matrix paradigm.
Biology is the existence proof. The only intelligence that keeps pace with the real world is found in nature, and it evolved under hard limits of time and energy — not despite these constraints, but because of them.
From these constraints emerge the design principles for any real world intelligence.
Three principles that today's matrix paradigm breaks:
Sparse
Only a small fraction of the network is active, connected in asynchronous shifting patterns.
But matrices are dense, synchronised and fixed in structure.
Low-energy
Computes only the active parts, mostly locally, so it runs on minimal power.
But matrix machines compute giant matrices, and spend more energy moving data than computing.
Real-time
A continuous loop of sensing,
acting and learning. All from its own experience.
But matrix models are frozen. Trained once upfront, then fixed for inference.
You don't reach real world robotics by scaling the matrix mismatch.You have to build for how intelligence actually works
Our technology
So we're building a new stack,
co-designed.
We work on a single problem that can only be solved end-to-end.
Intelligence for the open world.
Sparse, dynamic models
Structures that change and grow as they learn. Built for continual learning, on device. With no matrix multiplication.
Ultrascalable hardware
Massively parallel, general-purpose. Computes only when there's something to compute. Real-time performance at scale.
Verified design tools
Building on decades of formal methods research. Designed for correctness by proof, not testing alone.
What it unlocks
The capability that defines the next decade.
AI systems that learn in the real world, in real time, will be a foundational technology: the critical enabler of autonomous robotics.
This will transform sectors across the economy, including transport, manufacturing, medicine, and the home.
You don't get there by iterating the existing matrix technologies. You get there by building something new
Built to be trusted
Governed for good
Logicians exists to develop transformative technology for the long-term benefit of humanity.
That purpose is written into the company's constitution and guarded by a special class of share, so it can't be gradually eroded.
Secure by design
We design for security at every layer. It's essential for the critical systems we all want to rely on.
Our toolchain uses mathematical proofs for verification, and our hardware architecture removes whole classes of vulnerability found in current chips.
The team
Our team:
one goal,
many fields.
Deeply interdisciplinary, our team's expertise spans computer architecture, algorithms, neuroscience, formal methods and commercialisation.
With experience from Cambridge, Oxford, Bristol, The Advanced Research and Invention Agency (ARIA), XMOS and STMicroelectronics.
CONTACT US
We're growing our team carefully.
This is the defining challenge of our era, and the most rewarding work of our lives.
Could it be yours, too?
If you answer yes to any of these, we'd love to hear from you.
Do you suspect AI is built on the wrong foundations for the open world?
Do you think current programming languages hold us back?
Do you think computer architecture took a wrong turn?
Curious?
If what we're building resonates, we'd like to talk.
hello@logicians.com