The Cloud’s TSMC Moment: Why Neoclouds Won't Own Hardware

In 1987, the founding of Taiwan Semiconductor Manufacturing Company (TSMC) began a fundamental shift in the semiconductor industry by pioneering a “pure-play” foundry model. Prior to this, industry heavyweights like Intel, Texas Instruments, and NEC Semiconductor were vertically integrated device manufacturers (IDMs), handling both integrated circuit (IC) design and manufacturing. TSMC changed that by specializing exclusively in manufacturing, enabling a new generation of fabless startups like NVIDIA to outsource this capability and specialize solely in IC design. 

The decoupling of the semiconductor business has had enormously disruptive consequences. Today, NVIDIA is the largest semiconductor firm by market capitalization and stands at 30x the capitalization of Intel. TSMC itself is 7x the size of Intel and has become one of the most valuable and critical companies in the world, responsible for supplying virtually all of the cutting-edge chips for giants such as NVIDIA and Apple. 

This specialization has facilitated enormous progress on both sides of the industry, which directly led to the advancements facilitating the current AI boom. Now, because of that same boom, similar economics are facilitating another decoupling and disruption further up the value chain. 

Today, cloud computing is experiencing the same critical threshold of complexity and increasingly divergent capital requirements that drove the decoupling of IC design and manufacturing in the late 80s and 90s. In the future, most clouds, and eventually even the largest, won’t own any hardware at all. 

Understanding the Decoupling: Moore’s First and Second Laws 

Most are familiar with Moore’s Law: the number of transistors on integrated circuits doubles roughly every two years. However, Moore’s second law, less widely known but equally important, states that the cost of building state-of-the-art semiconductor foundries doubles approximately every four years. 

This exponential growth in capital requirements, alongside the parallel growth of design sophistication, naturally set the stage for decoupling and specialization. Companies found it increasingly difficult to master both sophisticated chip design and the complexity needed to manufacture them cost-effectively at scale. 

Today, an almost identical dynamic is emerging in cloud computing. Cloud giants like AWS, Google Cloud, and Azure remain vertically integrated, managing both massive server infrastructure and sophisticated software stacks. These big three are even designing their own in-house AI chips (like TPUs) to further verticalize their stack. 

The first generation of AI-centric clouds ("neoclouds"), like Lambda Labs, follow a similar vertically integrated model by building their own hardware while developing software platforms. Yet, due to economic pressures, further vertical integration faces significant headwinds. The first companies to realize and execute on this effectively will dominate the next generation, just as TSMC and NVIDIA do today. 

The Emergence of “Pure-Play” AI Factories 

The accelerating sophistication and cost of GPU-powered data centers (referred to as AI Factories by NVIDIA’s Jensen Huang) parallel semiconductor manufacturing facilities. NVIDIA’s A100 GPU launched around $20,000, while the forthcoming B200 GPU is near $40,000. Power consumption has quadrupled from 250W to 1000W, mirroring the rapid escalation characterized by Moore’s second law.

As HPC infrastructure complexity escalates, so does the pace of AI software innovation. Rapid software iteration cycles demand frequent updates and adaptations. Recent advancements like Cursor, Replit, and GitHub's Copilot, which reportedly boosts developer productivity by 50%, illustrate that software iteration cycles may be doubling roughly every two years, echoing Moore’s first law. 

Cloud providers will increasingly recognize that AI hardware investment and cloud software development require fundamentally different expertise, capital, and focus. Just as AMD eventually went fabless, cloud service providers will similarly spin out hardware ownership. 

A New Generation of Neoclouds 

Future neoclouds will specialize exclusively in software and user experience by offloading hardware investment and management to dedicated AI Factory providers. AI Factories, similarly to TSMC, will absorb intensive capital expenditures, achieving economies of scale, and offer flexible bare metal APIs. This allows neoclouds to concentrate resources entirely on differentiated software innovation. 

Early examples of this emerging model include TogetherAI, Lepton, and Prime Intellect, which emphasize software innovation while leveraging specialized infrastructure partners, mirroring the fabless-foundry relationship. 

Conversely, Fly.io exemplifies the risks of hardware investment for software-focused platforms, having been burned by GPU ownership. The demand for their initial system of GPU delivery did not materialize, highlighting the risk of a CapEx rather than an OpEx strategy when focused on software iteration. Their experience reinforces the need for cloud companies to focus on software innovation and validation and avoid hardware risks. 

The emergence of dedicated AI Factories will dramatically lower barriers to entry, further enabling specialized “fabless” cloud companies. Over time, this landscape could evolve niches similar to the semiconductor industry, with different AI Factories serving specific neocloud segments—akin to Vanguard Semi and UMC serving specific semiconductor niches today. 

TSMC's choice to specialize transformed the semiconductor industry and, consequently, the world. The cloud industry's TSMC moment is occurring now, driven by semiconductor specialization. Companies like Vast AI, Runpod, Salad, SF Compute, and Hydra Host, which transforms independent data centers into specialized AI Factories, exemplify this shift. 

The decoupling is underway. The only questions are how quickly it will play out and whether hyperscalers can adapt—or if they'll be disrupted by the one-two punch of "fabless" neoclouds and pure-play AI Factories.