Rethinking the future of computing infrastructure
The growing demand for compute is driving the industry toward increasingly powerful hardware and infrastructure at unprecedented scale. Yet the scale of a data center does not necessarily reflect how much useful computation it can deliver. Its performance depends not only on the hardware it houses or the power available to it, but also on how efficiently it utilizes and coordinates its resources.
As the physical constraints of energy, cooling, and infrastructure become increasingly important, we believe the next advances in computing will depend not only on building more powerful machines, but also on making the infrastructure that supports them fundamentally more intelligent.
The limits of fragmented infrastructure
The evolution of computing is progressively changing the scale at which computational systems are designed and operated. From individual processors, we have moved toward interconnected accelerators, multi-GPU systems, and rack-scale architectures, where dozens of GPUs operate as a tightly coupled computational system. As a consequence, performance increasingly depends on the interactions between components rather than on their individual capabilities alone. We see this evolution pointing toward a more fundamental shift: the data center itself becoming the next unit of computation and optimization, operating as one integrated machine rather than a collection of independent systems.
A data center, however, is far more than a computational system. It is a complex cyber-physical ensemble in which computing, networking, thermal and fluid dynamics and electrical power distribution continuously interact. Despite these interdependencies, its subsystems are often designed, managed, and optimized through separate tools, models, and objectives. This specialization helps manage complexity, but optimizing each subsystem according to its own objectives can leave the overall infrastructure operating below its full potential.
This complexity is compounded by the fact that data centers are constantly evolving. The physical properties of components change throughout their lifecycle, operating conditions fluctuate, and infrastructure is progressively expanded and reconfigured. At the same time, workloads, model architectures, and computational requirements continue to evolve.
Our thesis
We believe the next major leap in computing infrastructure will come from treating the data center itself as the fundamental unit of optimization. Just as computing has evolved from individual processors toward increasingly interconnected systems, we envision extending this approach to the entire facility, where computational and physical resources can be understood and coordinated together. By optimizing the infrastructure as a whole, rather than its subsystems independently, we see an opportunity to unlock new levels of efficiency, performance, and resource utilization.
The complexity and dynamic nature of these infrastructures also call for a fundamentally different approach. A system governed by thousands of interdependent variables, whose conditions and requirements continuously evolve, cannot rely exclusively on static configurations and predefined strategies if it is to operate at its full potential. It must be able to understand its own state, anticipate the consequences of its decisions, and continuously adapt its behavior to changing conditions. Intelligence, in our view, is not simply about making individual components smarter, but about enabling the entire system to understand and govern its own behavior.
This intelligence, moreover, cannot be confined to the computational domain. Computing is inseparable from the physical infrastructure that makes it possible. Power, heat, fluid flows, mechanical components, and computational systems jointly determine what a data center can achieve. We believe that truly intelligent infrastructure must therefore understand and coordinate both the digital and physical worlds, accounting for their interdependencies within a unified understanding of the system.
We do not see this transformation as something reserved for the distant future. Modern data centers already generate enormous amounts of information about their own operation, while advances in mathematical modeling, machine learning, optimization, and control systems are making increasingly complex problems computationally tractable. Much of the potential of these technologies remains unexplored, particularly in combining information across different domains and translating it into coordinated decisions. We believe many of the foundations required to build truly intelligent computing infrastructure already exist today.
Towards autonomous infrastructure
We envision a future where data centers operate as intelligent, increasingly autonomous organisms, capable of understanding their own state, anticipating changes, and coordinating computational and physical resources as a whole. These systems would continuously adapt to evolving conditions, learn from experience, and improve their efficiency and performance over time. Resilient to failures and scalable as infrastructure grows, they would progressively make and execute decisions autonomously, while operating within defined safety and operational constraints.
Building this future starts with understanding the infrastructure as it operates today. We are engaging with the people who design and run data centers, studying how computational and physical systems interact, and identifying where existing approaches leave the greatest potential unrealized. The solutions may begin with software and control, but could ultimately require new hardware or entirely different architectures. We intend to work from the underlying problems outward, rather than defining the limits of our ambition by the first product we build.
Antemyth was founded by four researchers and engineers who began collaborating at MIT, bringing together backgrounds in mathematics, physics, neuroscience, computational fluid dynamics, and mechanical engineering. We believe this interdisciplinary perspective is particularly suited to a challenge in which the computational and physical worlds are inseparable.
AntemythSan Francisco, October 2026