The Bessemer Process of Compute: How Google's Custom Silicon Could Reshape Cloud Economics

The cloud computing market is the steel industry of our age—concentrated, capital-intensive, and decisive for the digital empires rising upon it. The global market, valued at $781 billion in 2025, is projected to reach as much as $2.9 trillion by 2035 ^{3,12,23,37,54}. Three great combines—Amazon Web Services, Microsoft Azure, and Google Cloud—command between 63% and 72% of global infrastructure ^6,33,38. For Google Cloud, the smallest of the three with a 14% share ^2,13,34,40, the race is not merely for market share but for the strategic chokepoints that will define the next decade of artificial intelligence. This is a contest where capacity, cost curves, and political maneuvering will determine whose platform becomes the “master resource” of the AI economy.

The Oligopoly and Its Discontents

This concentration has drawn scrutiny from UK and European regulators ^49,50, with the EU now considering gatekeeper designations that could reshape the competitive landscape ^27,39. Proposed procurement criteria could exclude US-controlled clouds from the most sensitive state tenders ¹⁴, a direct threat to Google Cloud’s European ambitions. So far, the company’s sovereign cloud offerings—such as its partnership with Thales in Germany ^11,16—represent a pragmatic countermove, but legal tests like the Dutch DICTU assessment ¹¹ reveal the deep vulnerability of any sovereignty solution ultimately tied to a US parent. The wider EU market, where American firms hold 70% share ⁵¹, is under political pressure to shift sensitive data to domestic providers ^8,28,42. For Google Cloud, the question is whether its sovereign cloud strategy can satisfy both regulators and national security hawks, or whether it will be forced to cede ground to local alternatives.

The Rise of Specialised Mills: Neoclouds and Multi-Cloud Realities

In the steel age, integrated giants faced challenges from specialty mills that undercut them on price and flexibility. Today’s neoclouds—CoreWeave, Lambda, Nebius—are the AI-native specialty mills, offering dense GPU capacity and simpler terms ^10,17,18,41. Five such firms now appear among the top 30 cloud suppliers ⁴⁰, and their addressable market is at least $10 billion ^44,45. These players are reducing the dominance of incumbents ⁵⁵. While Google Cloud is not a neocloud, this fragmentation matters: enterprises are adopting multi-cloud strategies at 89% rates ^20,52,53, seeking to avoid hyperscaler premium pricing for AI workloads ¹⁷. This could pressure Google’s margins, but it also opens opportunities—hyperscalers themselves are already customers of neocloud capacity ^9,43. The multi-cloud norm, orchestrated by tools like Terraform and Ansible ⁴⁶, makes switching costs high ⁵⁰, but it also means that no single hyperscaler can afford complacency. Google’s comprehensive suite—from compute to AI accelerators—positions it well to capture sticky spend, but it must continuously invest to maintain feature parity with AWS and Azure, whose service breadth in areas like CDN and serverless computing remains larger ¹¹.

The Bessemer Process of Compute: Custom Silicon and Capital Discipline

The decisive advantage in cloud computing is shifting from software features to the cost and performance of underlying hardware. Hyperscalers are committing hundreds of billions to infrastructure through 2030 ³¹, and capital expenditure now consumes roughly 90% of operating cash flow ²⁹ and equals about 2% of U.S. GDP ⁴⁷. Just as Carnegie’s adoption of the Bessemer process gave him command over steel economics, Google’s custom silicon—its TPUs and Axion CPUs—represents a bid to break free from NVIDIA’s pricing power ^7,35,53. Arm-based architectures are gaining ground across the industry ³⁶, and Google is scaling proprietary infrastructure ³² while offering multi-year committed-use discounts to lock in enterprise customers ⁵². Yet, Google Cloud’s GPU inventory remains the smallest of the three hyperscalers ¹⁵, a near-term bottleneck in serving AI training demand. Google Cloud’s operating margin of 32.9% ²⁶ shows that profitable scaling is possible even as the cycle strains cash flows ²². But the long-term return on this capital hinges on whether custom silicon can bend the cost curve faster than competitors, and whether regulatory and competitive fragmentation can be contained.

Strategic Imperatives for Google Cloud

In this industrial struggle, Google Cloud’s advantages are real but precarious. It is the fastest-growing major hyperscaler ²⁵ and holds the top spot in the Cloud Wars Top 10 ranking ⁴⁸, driven by strength in data analytics, cybersecurity, and AI workloads ⁵⁶. It leads in developer-agent partnerships ³⁰ and competes directly with AWS and Azure on infrastructure, latency, and developer experience ^{1,4,5,19,21,24,56}. Yet its revenue base is roughly one-third that of AWS ²⁵, and in public IaaS/PaaS its share is less than half of AWS’s ⁴⁰. To thrive, management must act with the discipline of capital that built the great industrial fortunes:

• Double down on proprietary accelerators and Arm-based CPUs to reduce hardware dependency and achieve structural cost leadership.
• Use committed-use contracts and platform integration to deepen switching costs, transforming sticky relationships into a moat against neocloud and sovereign incursions.
• Expand sovereign cloud partnerships aggressively, but also invest in legal and technical architectures that can pass the hardest sovereignty tests—perhaps through locally incorporated operational entities.
• Leverage multi-cloud orchestration trends by ensuring that Google Cloud’s AI and data services are the easiest to integrate across environments, turning fragmentation into an advantage via API surface and developer ecosystem gravity.

Regulatory headwinds are the wild card. If EU procurement exclusions materialise broadly, the damage to Google Cloud’s government pipeline would be immediate. Yet, the hyperscaler model—with its massive capex, platform integration, and global scale—remains the most efficient way to deliver advanced cloud services. History suggests that industries subject to heavy regulation do not become less concentrated; they merely shift the basis of advantage from raw scale to regulatory navigation. Google Cloud’s ability to master that navigation, while relentlessly improving its cost curves and platform moats, will determine whether it remains a distant third or emerges as a true industrial peer to AWS and Azure.

This is the new steel: a combination of compute, platform power, and geopolitical alliances. The firm that commands the most efficient production and the widest distribution will, in the end, own the means of computation.