In the quiet corridors of Silicon Valley venture firms and the bustling trading floors of Lower Manhattan, a single chart has begun to circulate with the same fervor once reserved for pandemic infection curves. It depicts a vertical ascent so steep it renders the traditional logarithmic scales of technological progress obsolete. According to new data from research organization Metr, the capacity of artificial intelligence models is not merely improving; it is doubling every seven months. This pace obliterates the decades-old benchmark of Moore’s Law, which dictated a doubling of computing power roughly every two years, and suggests that the current frenzy of capital expenditure by hyperscalers is not a bubble, but the down payment on a fundamental economic restructuring.

The implications of this compression in development timelines are staggering. If the seven-month doubling rate holds true, the industry is not approaching a plateau but is rather in the nascent stages of an exponential climb that could persist for at least another half-decade. As reported by Sky News, this trajectory suggests the disruption caused by AI will be "bigger than COVID," a comparison that refers not to viral transmission, but to the sheer magnitude of the shock to global supply chains, labor markets, and productivity metrics. For industry insiders, the signal is clear: the infrastructure currently being laid—the gigawatt-scale data centers and the clusters of H100 and Blackwell GPUs—is likely insufficient for the compute demands arriving in 2027, let alone 2030.

The divergence between historical hardware cycles and the current algorithmic velocity suggests that capital expenditure estimates for the remainder of the decade are likely conservative rather than exuberant.

The skepticism surrounding the current AI boom often hinges on the comparison to the dot-com era, where fiber-optic cable was laid for bandwidth demand that took a decade to materialize. However, the Metr data indicates a fundamental inversion of that dynamic: demand for compute is outstripping supply by a factor that hardware manufacturers are struggling to physically accommodate. While Moore’s Law was a hardware-driven phenomenon, the current expansion is a compound effect of hardware density and algorithmic efficiency. We are witnessing a "compute super-cycle" where the utility of the next generated token is higher than the last, driving a voracious appetite for inference capacity that existing grids cannot easily support.

This seven-month cycle creates a unique pressure cooker for the semiconductor supply chain. It implies that a chip architecture released today is effectively legacy technology within three quarters. For companies like NVIDIA and AMD, this necessitates a product roadmap that is not just aggressive but relentless. The recent volatility in tech stocks, driven by fears of return on investment (ROI), clashes with the engineering reality that stopping the investment now would mean ceding the future. As the Sky News report highlights, there is likely "at least another five years" of this exponential runway remaining. If the models continue to scale at this velocity, the primary bottleneck shifts from silicon availability to energy generation, forcing Big Tech to become de facto energy utilities.

The physical constraints of power generation and thermal dynamics are emerging as the only true governors on a pace of innovation that has otherwise decoupled from traditional economic friction.

The narrative of "Bigger than COVID" is instructive when analyzing the downstream effects on enterprise operations. During the pandemic, the world saw a forced digitization of work; the AI shift represents the automation of the cognition behind that work. The graph produced by Metr does not just show speed; it shows the accumulation of capability. When capabilities double twice a year, the window for regulatory oversight or corporate adaptation shrinks to near zero. Corporations that are currently piloting distinct AI projects are finding that by the time a proof-of-concept is validated, the underlying model has been superseded by a version twice as capable and half as expensive to run.

This rapid obsolescence creates a paradox for CIOs and CTOs: invest heavily now in infrastructure that depreciates instantly, or wait and risk total irrelevance. The smart money is currently betting on the former, viewing the depreciation as the cost of admission to the new economy. We are seeing a shift where "compute" is treated less like IT equipment and more like a commodity resource, akin to oil or electricity. The volatility in the AI sector is not a sign of weakness but a symptom of a market trying to price a resource that is expanding in supply and demand simultaneously at rates never before seen in industrial history.

While the focus remains on Large Language Models, the seven-month doubling rate is silently revolutionizing adjacent fields like materials science and synthetic biology, creating value that current market caps fail to capture.

The shockwaves of this growth are perhaps most visible in the recent moves by sovereign entities and hyperscalers to secure sovereign AI clouds. If capacity doubles every seven months, the strategic advantage of possessing the most advanced cluster is fleeting but decisive. A nation or corporation that falls six months behind is effectively a generation behind. This explains the frantic acquisition of GPUs by nations in the Middle East and Asia, and the frantic pace of datacenter construction in North America. They are not merely buying chips; they are buying a hedge against the exponential curve described in the Metr research.

Furthermore, the "Bigger than COVID" analogy extends to the inflationary pressures on specific sectors. Just as the pandemic caused inflation in durable goods due to supply chain snarls, the AI boom is causing hyper-inflation in the cost of technical talent and specific high-voltage power equipment. Transformers, cooling systems, and grid interconnects are the new N95 masks—essential, scarce, and priced accordingly. The industry is effectively pulling forward a decade of energy infrastructure investment into a three-year window, driven by the terror of the graph that shows no sign of bending.

The integration of reasoning models and agentic workflows marks the transition from passive information retrieval to active economic participation by software systems.

Recent developments in "reasoning" models, such as those pioneered by OpenAI and DeepSeek, suggest that the next phase of this doubling curve will focus on the depth of thought rather than just the speed of token generation. This shifts the compute load significantly. "Thinking" models require vast amounts of inference-time compute, meaning that even if the training runs were to plateau (which they are not), the demand for GPUs to run the models would continue to explode. This validates the thesis that we are in the early innings of the infrastructure build-out. The seven-month doubling refers to capacity, but the utility of that capacity is compounding as models learn to check their own work and execute complex, multi-step tasks.

Ultimately, the graph from Metr serves as a warning label for the global economy. It suggests that linear projections for economic growth, energy consumption, and productivity are fundamentally flawed because they do not account for a variable that is compounding at a rate of roughly 300% per year. Investors and executives looking for a return to "normalcy" or a plateau in AI hype are likely to be disappointed. The math dictates that the disruption is accelerating, not stabilizing. In this environment, the only dangerous move is assuming that the future will resemble the recent past.