Algorithmic Squeeze-Out: How Computing Infrastructure Stifles SMEs in the AI Era

Building an Artificial Intelligence is no longer a software challenge, but a brutal race for hardware procurement. In 2026, the innovation ecosystem is a victim

Silicon Valley rhetoric loves to portray innovation as a garage where two brilliant students invent the future with a laptop and an internet connection. In 2026, this narrative is technically and economically obsolete. Developing, training, and scaling an Artificial Intelligence model requires not just mathematical intuition, but immense and extremely costly physical infrastructure: accelerated chips, boundless data centers, and gigawatts of electricity.

It is at this physical juncture that what we can define as the "Algorithmic Squeeze Out" is taking place. The fundamental problem is not simply that Artificial Intelligence is expensive, but that the necessary infrastructure is firmly controlled by a small number of operators (the hyperscalers) who can decide who gains access to it, at what price, and under what conditions .

In this in-depth analysis for the AI Business Lab, we will explore how infrastructural barriers are crushing Small and Medium Enterprises (SMEs) and independent startups, transforming the free technology market into an oligopolistic fiefdom.

1. The Supply Chain Oligopoly: From Chips to Cloud

To understand the scope of the squeeze out (exclusion from the market), we must look at the numbers of infrastructural concentration. We are not merely witnessing a monopoly on software models, but total control over the entire supply chain: datasets, computing power, cloud infrastructure, and app stores ".

Data from late 2025 compiled by Epoch AI estimates that the five largest global players (Amazon, Google, Meta, Microsoft, and Oracle) hold approximately 71% of cumulative global AI computing capacity . The traditional cloud services market, essential for hosting these models, also reflects this trend: a UK CMA survey confirms that, as early as 2024, Microsoft and AWS each hold a share between 30% and 40%, raising extremely high barriers to entry .

The US Federal Reserve has also highlighted how the chronic shortage of computing capacity in advanced economies makes it extremely difficult for SMEs to compete with large companies for access to computational resources (compute) . If *compute* remains a scarce resource, Big Tech wins by controlling the limited supply, while smaller players remain dependent and vulnerable to unfavorable contractual terms .

2. The Four Gauntlets of AI Startups

How does infrastructure block innovation? The European Commission and the OECD have mapped a series of obstacles that translate into a veritable competitive "squeeze out." We must distinguish four precise dynamics:

  • Barriers to entry: The exorbitant costs of accelerated chips and cloud capacity radically limit competition from the outset . The exponential costs of training models force new tech companies to seek partnerships (often unbalanced) with Big Tech just to access servers .
  • Barriers to expansion: Companies that manage to overcome the startup phase face enormous fixed costs and a severe lack of interoperability between systems, which prevents them from growing agilely ".
  • Infrastructural lock-in (Dependency): Smaller customers remain trapped within a single cloud provider's ecosystem due to rigid contracts, vertical effects, and above all, egress fees (the punitive charges imposed for transferring their data to a competing server) .
  • Competitive squeeze out: This is the final phase, where infrastructure costs exceed available profit margins, excluding SMEs from the market. A study by the Competition Commission of India (CCI) finds that 61% of businesses report cloud costs as a significant obstacle , highlighting how information asymmetries and cost barriers limit the startup ecosystem .

3. The Physical Limit: Concrete and Electricity

Artificial Intelligence lives in the cloud, but the cloud is made of concrete, steel, and copper. The shortage of accelerated computing clashes with the very long lead times for building new data centers ".

According to a McKinsey analysis, hyperscalers will capture approximately 70% of new data center capacity in the US . Furthermore, grid connection times have become the primary constraint for anyone wishing to expand . In this scenario, as ACT points out, small tech companies are the weakest players, most exposed to rising energy and permitting costs, and structurally less capable of absorbing delays in the construction of new infrastructure ".

The result, as industry analysts emphasize, is that infrastructure costs are turning the race for Artificial Intelligence into an elitist game reserved for very few ".

Key Operational Takeaways (for SMEs and CTOs)

  • Multi-Cloud Architecture from the Start: Avoiding vendor lock-in is a matter of survival. Design agnostic architectures using containers (e.g., Kubernetes) and open-source standards that allow you to migrate workloads from one cloud to another based on current economic convenience.
  • Consider Specialized Cloud Providers (Tier 2): Hyperscalers offer complete ecosystems, but often at premium prices. Evaluate alternative or specialized low-cost GPU cloud providers (GPU-as-a-Service) that offer computing power without the burdensome egress costs for data export.
  • Extreme Model Optimization (SLMs): Instead of training or fine-tuning expensive Large Language Models (LLMs), invest in developing hyper-specialized Small Language Models (SLMs) tailored to your business data, which require an infinitesimal fraction of the computing power (and cloud cost) for daily inference.

FAQ: Understanding the Algorithmic "Squeeze Out"

1. What exactly does "Squeeze out" mean? In economic and financial language, squeeze out (squeezing, expulsion) refers to an operation or market condition that forces minority shareholders or smaller companies to exit the market itself. In the AI context, it occurs when the operational costs of maintaining infrastructure (cloud and APIs) become so high that they wipe out startups' profit margins.

2. Why is Cloud so expensive for AI compared to traditional software? Traditional software (e.g., an e-commerce site) primarily requires CPUs (standard processors) that consume relatively little. Generative AI requires high-performance GPUs (graphics processors) that consume an enormous amount of electricity and require very complex liquid cooling systems. These physical costs are passed on to the end user's cloud bill.

3. Are antitrust authorities intervening? Yes. Both the European Union, the British authority (CMA), and US agencies are conducting in-depth market investigations into the Cloud and AI markets, focusing particularly on unfair practices such as egress fees (data exit costs) and exclusive partnerships that tie startups to the big tech giants.

Conclusions: The End of the Free Digital Market?

The Artificial Intelligence revolution was promised to us as the greatest wave of technological democratization in history. Yet, the harsh hardware reality tells a story of extreme concentration of power.

The analysis of the current ecosystem leaves no room for illusions: having a brilliant idea or the most efficient algorithm is not enough if you lack the financial means to rent the servers needed to run it. If access to fundamental infrastructure remains a bottleneck controlled by those who set the rules of the game, a troubling systemic question arises. If only those who control the infrastructure can afford to run the AI race, does it still make sense to talk about a competitive market, or are we silently sliding towards the formation of an impregnable infrastructural oligopoly?

Bibliographic References and Sources

  1. Market Dynamics and Regulation (OECD, EU, CMA):
    • OECD – Competition in Artificial Intelligence Infrastructure. "
    • OECD – Cloud Computing Services and Competition. "
    • European Commission – Competition Policy Brief on AI. "
    • UK CMA – IaaS Market Investigation. "
  2. Impact on SMEs, Startups, and Local Studies:
    • Competition Commission of India (CCI) – Market Study on AI and Competition. "
    • Moneycontrol – CCI Study Flags Steep Barriers for Indian AI Startups. "
    • Federal Reserve – The State of AI Competition in Advanced Economies. "
    • ACT – Comments on Singapore Digital Infrastructure. "
    • Ticker News – "Very Difficult": Why Small AI Companies Are Being Squeezed Out. "
  3. Analysis of Hyperscaler Power and Physical Infrastructure:
    • Bruegel – Why AI Is Creating Fundamental Challenges for Competition Policy. "
    • Mozilla – Is Big Tech Stifling AI Innovation? "
    • Epoch AI – AI Data Centers. "
    • McKinsey – The Next Big Shifts in AI Workloads and Hyperscaler Strategies. "
    • CERRE – A Competition Policy for Cloud and AI. "
    • AI Now Institute – Heads I Win, Tails You Lose. "

Article by the Editorial Team of La Bussola dell'IA – AI Business Lab Column.