Emilio Billi

One causal system

Emilio Billi does not view computing infrastructure as a collection of processors, accelerators, networks and cooling systems. He treats it as one causal system in which every design choice changes the behaviour of the whole. The starting question is not how many peak operations a component declares, but how much of that potential can become useful, sustainable and verifiable computation.

For more than twenty years, across Europe and the United States, his work has examined system architecture, data movement, memory, interconnects and physical constraints. It spans HPC, PCIe scale-up, shared memory, hardware-software co-design and accelerated-system engineering, with engagement in international technology ecosystems including IEEE Hot Interconnects, PCI-SIG and HyperTransport.

Architecture begins with the path of data

In AI Infrastructure senza nebbia, Billi presents a quantitative method: MODEL → NUMBERS → MEASUREMENT → DECISION. Before choosing hardware, the architect reconstructs who produces the data, who consumes it, where it resides, which path it crosses, and how much time and energy are available to move it.

The Future of AI Infrastructure expands this into a thesis about computing: useful performance belongs to the whole system. Data distance, memory hierarchy, communication determinism and heat removal are first-order specifications, not secondary infrastructure details.

The Bandwidth Collapse Theorem formalises a related idea: nominal bandwidth is not the bandwidth available to the workload. Each boundary and contention point can reduce the effective resource; the narrowest cut crossed by the workload governs the result.

From limits to machines

Billi’s engineering connects to a broader investigation into the limits of formal systems, knowledge and representation in works including L’eco di Gödel, Il limite intelligente and La nuova ontologia. A limit is not merely a deficiency: it is a structural property that must be made visible before it can be governed.

GRIFO is the industrial translation of that view: enlarge the local machine, keep work close to data, reduce avoidable boundaries and engineer compute, memory, I/O, software, power and cooling as one system. Its international relevance comes from the global scale of the problems addressed — data movement, energy density, determinism, technological sovereignty and sustainable computing.