What is computational scale beyond Moore’s Law?

It is the shift from device-level transistor scaling to systems-level computational progress driven by specialized accelerators, datacenter parallelism, algorithmic efficiency, data, energy, and capital allocation.

Why does AI compute still grow if Moore’s Law has slowed?

AI compute can grow through larger clusters, better accelerators, improved networking, more efficient algorithms, larger budgets, and better data pipelines, even when transistor-density doubling slows.

What are the main constraints on frontier AI scaling?

The main constraints are electrical power, data availability, capital intensity, semiconductor supply chains, and interconnect latency across very large training systems.

Why does this matter for executives?

It matters because AI capability increasingly depends on infrastructure economics and operating model design. Strategy teams need to understand compute access, energy exposure, vendor dependence, data readiness, and where advanced AI can change unit economics.

Computational Scale Beyond Moore’s Law

By Juan Beltrán · 2026-04-23

Computational progress has not ended with Moore’s Law. It has moved from device-level scaling to systems-level scaling across accelerators, datacenter fabric, algorithms, energy, data, and capital. For executives, this means AI capability should be treated as an infrastructure and operating model question, not only a software adoption question.

Why this paper matters

For many executives, Moore’s Law still functions as a shorthand for technological progress. Faster chips. Lower cost. More capability. Better software follows.

That mental model is now incomplete.

The classical Moore-Dennard regime has slowed, but the practical computational capability deployed into frontier AI systems has continued to compound. The reason is not that the old law secretly survived. The reason is that the locus of progress moved. The relevant unit is no longer the individual transistor. It is the full production system that combines accelerators, datacenter fabric, algorithms, energy, data, capital, and organizational intent.

That distinction matters for AI transformation. If AI capability is treated as a software feature, leaders will underestimate the infrastructure, economics, and operating model behind it. If it is treated as a systems-level production function, strategic choices become clearer: where to rely on frontier platforms, where to build proprietary advantage, where energy and data constraints matter, and where algorithmic efficiency can change the business case.

The executive mistake I see is treating compute as an invisible supplier decision. In AI, compute behaves more like energy, capacity, and balance sheet: it shapes what is strategically possible before the business case is even written.

Core argument

The end of classical scaling does not mean the end of computational acceleration. It means the end of one dominant mechanism.

For roughly five decades, computing improved because transistor density increased and power density stayed manageable. Moore’s Law and Dennard scaling together gave the industry a reliable roadmap. Designers could expect more transistors, higher clock rates, better performance per watt, and lower cost per operation.

Dennard scaling broke first. Leakage current and threshold-voltage limits made it impossible to keep raising clock frequencies without unacceptable heat. Moore’s Law then slowed materially as lithography, manufacturing complexity, and economics made each new node harder and more expensive.

Yet frontier AI moved in the opposite direction. Since 2010, training compute for notable AI models has grown at approximately 4 to 5x per year. Frontier language models have followed even steeper trajectories. At the same time, algorithmic efficiency improved independently, meaning that the same amount of compute can produce more capability over time.

The new regime is therefore not transistor-led. It is systems-led.

Key findings

The working paper identifies six findings that are most relevant for strategy:

Classical Moore-Dennard scaling has attenuated, but effective AI compute has accelerated.
Training compute for notable AI models has grown by approximately 4 to 5x per year since 2010.
Algorithmic efficiency has improved independently by approximately 3x per year.
Compute-intensive AI has already enabled domain-level breakthroughs in structural biology, materials discovery, weather forecasting, and formal mathematical reasoning.
The binding constraints are shifting from lithography to power, data availability, capital intensity, and interconnect latency.
Through 2030, the most plausible constraint is thermodynamic and economic, not transistor density.

These findings should not be read as technology optimism. They should be read as operating reality. Capability growth is real, but it is increasingly dependent on scarce physical and economic inputs.

Scientific discovery as evidence

The strongest evidence for the new computational regime is not a benchmark chart. It is the concentration of scientific results that would have been difficult or impossible under older computational assumptions.

AlphaFold 2 changed protein structure prediction by reaching accuracy comparable to experimental methods for many targets. AlphaFold 3 extended that pattern to biomolecular interactions involving proteins, nucleic acids, ligands, ions, and modified residues. GNoME expanded the known space of stable inorganic materials by orders of magnitude relative to prior computational materials pipelines. GraphCast and GenCast showed that learned models can compete with or outperform traditional weather-forecasting systems on key medium-range tasks, while producing forecasts at radically lower marginal compute during inference.

These are not isolated demonstrations. They represent a pattern: when a scientific domain has reliable training data, a computable ground truth, and a large combinatorial search space, learned models can compress the search process.

Constraints through 2030

The paper evaluates four constraints that will shape the next phase of frontier AI scaling.

First, electrical power. The largest training systems are moving toward power requirements that make energy availability a strategic input. This is no longer a procurement detail. It is a constraint on where, how, and by whom frontier systems can be trained.

Second, data availability. Public high-quality text, code, scientific data, and multimodal corpora are not infinite. Synthetic data, better curation, domain-specific datasets, and interaction data may extend the runway, but they do not remove the constraint.

Third, capital intensity. Frontier training runs are becoming expensive enough that only a small number of firms, states, and partnerships can fund them directly. This increases platform concentration and raises the importance of access strategy for enterprises.

Fourth, interconnect latency. Scaling beyond a single accelerator requires communication across clusters. At very large scales, the fabric that connects compute becomes as important as the chips themselves.

Four scenarios

The paper presents four scenarios for frontier compute through 2030.

Scenario A is baseline continuation. Physical training compute continues to grow rapidly, supported by accelerator improvements, larger clusters, power buildout, and continued capital commitment. This scenario implies several more generations of scientific-discovery systems comparable in significance to AlphaFold, GNoME, and GraphCast.

Scenario B is efficiency-dominated scaling. Physical compute growth slows earlier, but algorithmic efficiency continues to improve. Capability still advances, but competitive advantage shifts toward research talent, architecture, data quality, and training methods.

Scenario C is a constrained plateau. Power, data, capital, and supply-chain constraints bind simultaneously. Progress continues, but at a slower rate, and frontier capability in 2030 resembles an extension of the 2025 to 2026 regime rather than a step change.

Scenario D is paradigm shift. A new architecture, training method, or computing substrate changes the effective compute equation. This is less probable on a short timeline, but it carries the highest upside variance.

Executive implications

The strategic implication is straightforward: AI transformation is becoming an infrastructure-aware discipline.

Executives do not need to manage chip roadmaps directly. But they do need to understand how compute economics affect capability access, vendor concentration, model choice, data strategy, and operating model design.

For most organizations, the right question is not whether to train frontier models. The right question is where frontier capability changes business economics, where proprietary context matters, and where the organization can create advantage without owning the full compute stack.

This creates a practical agenda:

Treat compute access as part of AI strategy, not as a technical afterthought.
Evaluate model choices against workflow value, latency, cost, security, and governance requirements.
Build proprietary advantage around data, process knowledge, distribution, integration, and change execution.
Track energy and platform concentration as strategic risks.
Use algorithmic efficiency as a planning variable, not only raw model size.

Read the full working paper

The web version is intentionally concise. The full paper includes the detailed empirical review, scientific examples, constraints analysis, scenario logic, and references.

Download the complete PDF below for the full working paper.

Research summary

Abstract

For five decades, computing progress was governed by Moore’s Law and Dennard scaling. Both regimes have now attenuated, yet effective AI compute has accelerated through specialized hardware, datacenter parallelism, algorithmic efficiency, and unprecedented capital allocation. This working paper reviews empirical evidence on training compute growth, examines compute-enabled scientific breakthroughs, and evaluates the constraints likely to shape frontier AI scaling through 2030.

Research question

If classical Moore-Dennard scaling has slowed, what now governs the growth of effective computational capability, and what does that imply for scientific discovery and AI strategy through 2030?

Methodology

Desk-based synthesis of public empirical datasets, peer-reviewed AI and scientific-discovery literature, industry roadmaps, and quantitative forecasts from Epoch AI and adjacent research. The analysis connects historical device-level scaling with contemporary systems-level scaling across compute, data, algorithms, energy, and capital.

Key findings

Classical Moore-Dennard scaling has attenuated, but effective AI compute has accelerated through system-level scaling.
Training compute for notable AI models has grown by approximately 4 to 5x per year since 2010.
Algorithmic efficiency has improved independently by approximately 3x per year, compounding the effect of physical compute growth.
Scientific breakthroughs in protein structure prediction, materials discovery, weather forecasting, and formal reasoning show how compute-intensive AI is changing discovery.
The binding constraints are shifting from lithography to power, data, capital intensity, and interconnect latency.
Through 2030, the most plausible constraint is thermodynamic and economic, not transistor density.

Executive implications

AI strategy can no longer be separated from compute economics, energy access, and infrastructure planning.
Executives should treat frontier AI capability as a systems-level production function, not a software feature.
The next competitive divide will come from organizations that understand how compute, data, algorithms, capital, and operating models compound together.
AI transformation roadmaps should include infrastructure realism, not only use case prioritization.

Limitations

This is a working paper based on public evidence available as of April 2026. The 2030 scenarios are directional, not deterministic forecasts, and should be updated as new empirical data on compute supply, algorithmic efficiency, energy availability, and model capability becomes available.

References

Moore, G. E. (1965). Cramming more components onto integrated circuits. Electronics.
Dennard, R. H. et al. (1974). Design of ion-implanted MOSFETs with very small physical dimensions. IEEE Journal of Solid-State Circuits.
Sevilla, J. et al. (2022). Compute trends across three eras of machine learning. International Joint Conference on Neural Networks.
Epoch AI. (2026). Trends in AI compute and hardware performance.
Jumper, J. et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature.
Merchant, A. et al. (2023). Scaling deep learning for materials discovery. Nature.
Lam, R. et al. (2023). Learning skillful medium-range global weather forecasting. Science.

Key takeaways

The post-Moore era is not a slowdown of computational capability. It is a shift from transistor scaling to systems-level scaling.
AI training compute and algorithmic efficiency have compounded faster than classical Moore’s Law, changing the economics of scientific discovery.
The next bottlenecks are power, data, capital intensity, and interconnect latency rather than lithography alone.
Compute-intensive AI has already produced domain-level results in structural biology, materials discovery, weather forecasting, and formal mathematical reasoning.
Executive AI strategy needs infrastructure realism: compute access, energy exposure, vendor concentration, data strategy, and operating model design.

Decision brief

More compute does not guarantee proportional business value. Leaders should translate infrastructure progress into task economics: what becomes reliable, affordable, fast, and governable enough to change a workflow that matters.

Decision criteria

The performance gain changes an economically important task threshold.
Energy, latency, data movement, and integration costs are included.
The organization can absorb the capability without creating control debt.

Evidence to check

Cost and quality per completed task at production volume.
Energy, latency, and infrastructure utilization under realistic load.
Sensitivity of the business case to model and hardware price changes.

What I would do Monday

For one compute-heavy AI proposal, remove model parameters and benchmark headlines from the first page. Replace them with cost per useful task, acceptable error rate, response time, and business owner.

Questions executives ask

What is computational scale beyond Moore’s Law?: It is the shift from device-level transistor scaling to systems-level computational progress driven by specialized accelerators, datacenter parallelism, algorithmic efficiency, data, energy, and capital allocation.
Why does AI compute still grow if Moore’s Law has slowed?: AI compute can grow through larger clusters, better accelerators, improved networking, more efficient algorithms, larger budgets, and better data pipelines, even when transistor-density doubling slows.
What are the main constraints on frontier AI scaling?: The main constraints are electrical power, data availability, capital intensity, semiconductor supply chains, and interconnect latency across very large training systems.
Why does this matter for executives?: It matters because AI capability increasingly depends on infrastructure economics and operating model design. Strategy teams need to understand compute access, energy exposure, vendor dependence, data readiness, and where advanced AI can change unit economics.

Canonical URL: https://juanbeltran.ch/blog/computational-scale-beyond-moores-law