100% Utilization: A Talk with Andrew Lison

Abstract: Exponential growth has become a central computing industry concept since the formulation of Intel Corporation cofounder Gordon Moore’s “law” in 1965. Predicting—and in so doing, instantiating—the regular halving of semiconductor component size (and corresponding performance benefits) without increasing manufacturing costs, its internal logic has ushered in significant economic transformations. 

 

From the centrality of computing for contemporary life amidst its “dematerialization” of commodities like vinyl records, videocassettes, and print media to the rise of computer-aided financial trading and ongoing improvements in factory automation, we are everywhere beholden to expectations that computational processes will continue encompassing ever more aspects of the material world. Yet this dynamic is increasingly coming up against the physical limitations of the hardware enabling it, namely the impossibility of shrinking transistors and other microchip elements much further than their present-day size. 

 

An end to Moore’s law thus threatens to impede the replacement of human labor with automation that is a key component of not simply the postindustrial turn but our present mode of political economy more broadly. One response to this has been increased reliance on parallel computing—that is, using more processors rather than faster ones. Yet this type of hardware, most famously associated with Graphics Processing Units and, more recently, the dedicated artificial intelligence chips that descend from them, is subject to the same physical constraints. 

 

Recent efforts to promote “Huang’s law,” a GPU-/AI-specific narrative of continuing growth named after NVIDIA CEO Jensen Huang, must, then, reckon with the same factors endangering Moore’s law.