Why We’re Only Using a Fraction of Available Compute — Elon Musk on Efficiency and the Role of Copper in Powering the AI Era
Share
Introduction: The Musk Insight on Underused Compute Capacity
In a recent Moonshots with Peter Diamandis podcast conversation, Elon Musk laid out a perspective that challenges common assumptions about the state of modern computing: we are currently using only a small fraction of the theoretical capacity inherent in today’s compute infrastructure and energy systems. While raw digital computing power continues to grow exponentially, Musk noted that the potential computational leverage — the combination of raw hardware, energy availability, and architectural design — remains vastly underutilized by current AI and software workloads.
This insight has major implications both for how we think about optimizing AI systems and for the real-world infrastructure — the physical compute, power delivery, cooling networks, and copper-rich electrical systems — that must evolve to support more effective utilization.
1. What Musk Is Talking About: Underutilized Compute Potential
While there isn’t a single soundbite on record claiming an explicit percentage of total potential compute in use, Musk’s comments throughout the podcast convey a clear theme: we are far from hitting the limits of what our compute infrastructure could achieve. The conversation described how current experimental AI, energy storage, and digital capacity are orders of magnitude behind both the theoretical and practical potential that exists when you factor in hardware, cloud systems, and energy scalability.
In essence, Musk is highlighting two key realities:
-
Hardware is not the sole bottleneck. Many data center clusters sit idle part of the time because workloads are bursty, or because system designs and software cannot saturate the hardware fully.
-
Energy supply and distribution are underleveraged. Even where power infrastructure exists, it’s rarely fed into compute at maximum continuous capacity due to cooling limits, cost optimization, or energy pricing structures.
This combination means that although we have trillions of FLOPs (floating-point operations per second) available globally, a large percentage is not consistently put to use.
2. Why This Matters for AI and Advanced Compute
In practice, the compute landscape looks like this:
a. AI Workloads Are Spiky
Large language model training, simulation workloads, and inference tasks often use massive bursts of compute, but then return to idle states while data is preprocessed, pipelines are queued, or energy costs fluctuate.
b. Data Centers Are Designed Conservatively
Operators often size clusters and power provisioning to avoid peak thermal or electrical stresses — ensuring reliability rather than absolute utilization. This is sensible for uptime but means compute hardware may sit idle much of the time.
c. Software Isn’t Fully Extracting Hardware Capability
Modern CPUs, GPUs, and AI accelerators have enormous theoretical throughput, but actual utilization percentages for many workloads (especially mixed or irregular ones) can be a fraction of peak capability.
Elon’s broader argument — that we have orders of magnitude more potential if we organize software, energy, and hardware more efficiently — is a reminder of how far current systems are from theoretical limits.
3. Copper: The Hidden Material Enabler of Compute Utilization
All of this is deeply physical. Hardware potential exists not just in abstract silicon or firmware — it is grounded in metal, power delivery, cooling, and the data center infrastructure that supports these workloads.
Here’s where copper becomes central:
a. Power Distribution and Delivery
To fully exploit compute potential — whether AI training clusters or general-purpose HPC — you need reliable, high-capacity power feeds. Copper is the dominant material in:
-
Electrical busbars
-
High-current power cabling
-
Transformer windings
-
Grounding systems
Copper’s exceptional electrical conductivity means it can carry large currents with lower losses compared to alternatives like aluminum. This reduces energy loss and supports higher stable utilization of compute hardware.
b. Thermal Management and Cooling Integration
Running hardware near its maximum capacity generates heat. Efficient cooling systems — whether air, liquid, or immersion — rely on copper for:
-
Heat exchangers
-
Cold plates
-
Chilled water loops
Copper’s superior thermal conductivity allows heat from dense rack installations to be efficiently transferred away from critical components, enabling compute hardware to sustain high utilization without throttling. Without copper-rich cooling infrastructure, chips will quickly hit thermal limits that force underutilization.
c. Short-Range Networking and Rack Interconnects
Even with fiber optics for long-haul links, within racks and cabinets, copper still plays a crucial role in:
-
Power distribution units (PDUs)
-
Board-level interconnects
-
High-current connectors
These components ensure that communications and power delivery inside compute clusters are synchronized and loss-minimized — which in turn supports higher utilization of the silicon available.
d. Fabrication and Lifecycle Support
Beyond data centers themselves, copper is used in:
-
Fabrication facility (fab) utilities
-
Test and burn-in environments
-
High reliability connector plating
-
Backup power systems
Copper’s reliability makes it a machine-builder favorite when designing for uptime, redundancy, and scalability — all of which are prerequisites for high sustained compute utilization.
4. The Feedback Loop: Smarter Software Meets Hardware Limits
As Musk suggests, it’s not enough to have raw wood-chipper levels of silicon capacity: without software models and scheduling systems that can effectively saturate the hardware, much of that potential remains idle. And these systems rely on the physical substrate — copper-rich infrastructure — to deliver power and cooling where and when it’s needed.
The compute stack from metal to machine to software breaks down like this:
-
Metal & Power (Copper): The foundation enabling energy to reach every transistor reliably
-
Hardware & Chips: The silicon that can compute at high FLOP rates
-
Operating Systems & Orchestration: Platforms that schedule, batch, and optimize workloads
-
AI Architectures: Models — from Transformers to Grok and beyond — that use that compute with purpose
Optimize only at layers 3 or 4 and you still hit limitations if layers 1 and 2 cannot supply and sustain continuous power and heat dissipation.
5. Implications for Infrastructure, Energy, and Metals Markets
Musk’s observation — that we are only using a fraction of available compute capacity — highlights a systems inefficiency that cuts across hardware and infrastructure. The response of the industry involves:
-
Building denser, high-capacity data centers
-
Increasing power delivery capacity (often requiring grid upgrades)
-
Deploying advanced cooling technologies
-
Reserving copper-rich electrical and mechanical systems for sustained utilization
As more companies push toward full utilization — especially for AI training and inference — the demand for copper grows. Copper will continue to be the core material that allows physical compute stacks to scale alongside software advancements.
Conclusion: Utilization, Infrastructure, and the Material Base of Compute
Elon Musk’s recent remarks with Peter Diamandis make a compelling point: today’s compute installations represent a fraction of the potential we could achieve if energy, scheduling, and architecture were aligned with hardware capacity. That’s a call not just for smarter software, but for stronger underlying physical infrastructure.
Modern computing doesn’t run on code alone.
It runs on power, cooling, and metal — and copper is the material that binds these together.
From power busbars and transformers to cooling loops and rack infrastructure, copper’s role is not optional — it’s foundational. As the world accelerates toward more intense utilization of computational resources, copper demand isn’t just preserved — it expands.
In the era where software meets the full potential of compute, copper is the metal that makes high utilization possible.