Design & Future Compute Architectures

1.45pm – 2.55pm, 8 July 2026 ‐ 1 hour 10 mins

Speaker Session

While AI dominates headlines, compute innovation is diversifying. Distinct markets are emerging for data centre training, inference, and edge computing, each requiring specialised solutions. Beyond AI, we are seeing a resurgence in CPU innovation and ultra-low latency optical architectures, while quantum computing nears commercial viability. The UK is well-positioned to lead, boasting a rich history of innovation and over 200 semiconductor design teams. The challenge lies in leveraging these domestic strengths within a global landscape currently controlled by a few massive vendors and customers.

Wenmiao Yu, Co-Founder, Director of Business Development, Quantum Dice Ltd.
Enhancing Logistics and Operations with Probabilistic Computing
Operational complexity across logistics, ports, and aviation is increasing at a pace that traditional computing struggles to match. From stochastic demand patterns and network congestion to real-time optimisation under uncertainty, many of today’s hardest problems are inherently probabilistic. This talk will explore how probabilistic approaches can unlock measurable gains in operations.

Phillip Burr, Head of Product, Lumai
The Future of AI Compute Will Be Optical
AI is power-limited, and silicon scaling is slowing at exactly the point inference demand is accelerating. Optical compute is an order of magnitude more energy-efficient than silicon and offers a direct route through the energy wall – this talk sets out the physics, the architecture, and the path to deployment.

Chris Smith, SVP Software, Fractile
Scaling Ultra-Fast Inference
Large-scale AI inference is beset by the famous "memory wall", with inference speed crushed by delays waiting on data. Previous attempts to break through with novel architectures have struggled to deliver at scale and cost. In this talk, Chris sets out why memory bandwidth has become the binding constraint on frontier inference, and what it takes to combine breakthrough speed with the real-world economics that make ultra-fast inference viable at scale.