How Long Will Neocloud Scarcity Last?

The commercialization of any era-defining computing technology typically features a race between suppliers: how long can new suppliers live off of scarcity before abundance eliminates the profit margins?

In other words, how long can business strategies based on scarcity flourish until abundance is reached, and how many of the early suppliers can make the jump from “commodity” product supply to some form of new value add?

That is one question investors have to ask themselves about the rise of third-party high-performance computing facilities operated by the likes of CoreWeave, Nebius and others. 

A related question for investors in artificial intelligence firms is whether the capital intensity of AI changes the traditional sources of value for computing products, which has tended to migrate “up the stack” over time. 

The capital intensity of the AI model training and inference operations might suggest there is a new and possibly somewhat lasting new “moat” value based on infrastructure ownership and control, which has tended not to be the case for most other computing innovations. 

This tension between scarcity-driven value and abundance-driven commoditization arguably has occurred in every new major wave of technology.

A new resource (compute, bandwidth, or power) is initially scarce, allowing early movers to capture "bottleneck rents."

However, incumbents and new entrants eventually respond by flooding the market with supply, destroying those rents and forcing value to migrate to a different layer of the stack.

Google (TPU), Amazon (Trainium), and Microsoft (Maia) have responded to Nvidia’s dominance by building their own silicon, starting to alleviate the physical scarcity of graphics processor untis for their internal workloads.

Open source has in the past functioned as a method of alleviating scarcity. The issue is how well this strategy will work in a capital-intensive AI industry. 

This has played out in past computing technology eras. 

In the late 1970s, computing was defined by integrated, proprietary systems like the Apple II or the Commodore PET. If you wanted the software, you had to buy specific hardware.

For a brief moment, hardware manufacturers enjoyed high margins because they controlled the entire "stack."

In 1981, IBM entered the market using off-the-shelf parts (Intel CPUs, Microsoft OS). That eliminated “scarcity” as other firms flooded the market with "IBM-compatible" PCs, turning the hardware into a low-margin commodity.

As the hardware became a commodity, the value migrated to the two components that remained proprietary and scarce: the Intel microprocessor and the Microsoft operating system (the "Wintel" monopoly).

In the era of cloud computing, server ownership ceased to be a barrier to compute hardware costs. As a practical matter, that meant software or app startups did not need to invest in their own compute facilities, but could rent them from AWS, for example. 

Tech Wave	Initial Scarcity (Rents)	Mechanism of Abundance
Electricity	Localized Generation	The Centralized Utility Grid
PC Era	The Integrated "Box"	Open Architecture & Clones
Internet	Bandwidth / Connectivity	Massive Fiber Overbuild
Cloud	Physical Server Racks	Virtualization & Hyper-scale
AI (Current)	GPU Compute / Tokens	Neoclouds & Model Distillation

The point is that, over time, scarcity is eliminated in the value chain, shifting value creation elsewhere in the product stack. 

So the issue is how neocloud providers such as CoreWeave and Nebius, which essentially have democratized access to high-performance clusters, will flourish going forward, as high-performance computing becomes less scarce and more commodity-like. 

Stage	Commodity Model (The "Plumbing")	Value-Added Model (The "Experience")
Example	Raw Token API, Basic GPU Hosting	AI Agents, Industrial Robotics, Sovereign AI
Revenue Driver	Volume and Scale (Low Margin)	Outcomes and Reliability (High Margin)
Competitive Edge	Lowest Price per FLOP	Trust, Performance, and Workflow Integration
Incumbent Threat	Massive CapEx from Hyperscalers	Regulation and “Reasoning” Moats

The rise of "Neoclouds" (specialized AI infrastructure providers like CoreWeave, Nebius, and Lambda Labs) challenged the assumption that only the "Big Three" hyperscalers (AWS, Azure, Google) owned the “AI as a service” infrastructure business. 

But history suggests the "compute moat" goes away over time, forcing suppliers to find new sources of value. 

Aspect	The "Old" Moat (Pre-2024)	The "New" Moat (Late 2025)
Primary Barrier	Ownership of H100 GPU Clusters.	Proprietary "Reasoning" and reinforcement learning
Strategy	Vertical Integration (Own the DC).	Architecture Efficiency (Train for less).
Infrastructure	Proprietary Cloud (Azure/AWS).	Multi-Cloud/Agnostic (Rent where available).
Value Capture	Selling Compute / Tokens.	Selling Outcomes / Agentic Actions

So investors must make decisions about where scarcity is in the value chain, how long that scarcity will last, and where value shifts within the value chain as a result.

U.S. Productivity is Up Sharply Over Two Quarters, But AI is Not the Reason

Economies are extraordinarily intricate machines, which explains the humorous references to “one-handed economists” (“on the one hand, on the other hand”). 

So we probably should not be surprised that U.S. data reveals a possibly-unexpected boost to nonfarm productivity growth at an annualized rate of 4.9 percent, a second consecutive quarter of gains, nor that we aren’t sure what caused the change. 

Morgan Stanley chief economist Michael Gapen said “it remains an open question as to what is driving the productivity acceleration.”

Some will undoubtedly want to point to the impact of artificial intelligence. Some of us likely doubt that. It seems unlikely we can measure such impact, so soon, and, in any case, there are other likely drivers. 

Many would note the U.S. labor market has been in a “low-hire-low-fire” mode for much of 2025.

When companies maintain or increase their output while hiring fewer workers, the mathematical result is a jump in productivity, while a “K-shaped economy” that has wealthier households propping up aggregate spending that less-wealthy households might be trimming, also is at work. 

And some will suggest that slower hiring now is a reaction to a period of over-hiring that happened after the Covid pandemic. 

The Great Resignation (or Big Quit) was a massive, pandemic-era trend starting in 2021 where millions of U.S. workers voluntarily quit their jobs, driven by burnout, low wages, poor benefits, and a desire for flexibility (especially remote work).

The Great Resignation began in early 2021, peaking with record quit rates around 4.5 million people in late 2021, according to the Bureau of Labor Statistics. 

The hiring pattern after the Great Resignation shifted to what some call the "Great Reshuffling," where workers didn't just leave but moved to better roles, demanding higher pay, flexibility (remote/hybrid), improved benefits, and better work-life balance, forcing companies to focus on employee retention, culture, and creating more attractive, purpose-driven environments to compete for talent.

The net result was a big wave of hiring where employers arguably “over-hired.”

The point is that AI probably does not explain the change. Businesses have not deployed widely enough, for long enough, in high-leverage use cases, to explain such a change. 

Some economists suggest AI could raise productivity around an extra 0.01 to 0.3 percentage points, with the primary effect so far being an indirect boost to overall gross domestic product growth through heavy investment in AI infrastructure itself. 

Over a decade or so, economists at Goldman Sachs suggest AI could boost productivity growth by between 0.3 and 3.0 percentage points a year over the decade following its widespread adoption, with a median estimate of 1.5 percentage points.

But that is not a reality, yet. So, no, AI does not explain the recent upsurge in productivity.

Which Future for Neoclouds: Rational Consolidation or Collapse?

Technology market structures tend to change as they age. Small upstart companies get acquired; bigger firms merge; a few dominant leaders emerge, taking a “winner takes most” structure. 

Any market researcher, studying any particular capital-intensive market, will tend to find something like a Pareto distribution often applies: up to 80 percent of results are produced by 20 percent of actors. Some might call that the rule of three. 

Market share structures in computing, connectivity and software tend to be fairly similar: leadership by three firms, corresponding to the rule of three. 

“A stable competitive market never has more than three significant competitors, the largest of which has no more than four times the market share of the smallest,” BCG founder Bruce Henderson said in 1976.  

Codified as the rule of three, the observations explains the stable competitive market structure that develops over time, in many industries. 

Others might call this winner take all economics. 

So a logical question is what happens in the high-performance computing market, including the market space for neocloud providers such as CoreWeave, Nebius and others changing their business models from bitcoin mining to focus on artificial intelligence model training and inference operations. 

Some might argue we are shifting from a focus on training capabilities and towards inference operations. It’s hard to argue with that observation, as models become routine apps used by businesses and consumers. 

So some might argue we could see less need for highest-performance compute capabilities of the sort neocloud providers offer. Others might argue more of the computational load will be handled by edge devices, and there is some truth to that position as well. 

But inference operation ubiquity does not necessarily mean less power; less powerful chips; fewer operations inside massive data center complexes; less physical real estate or water consumption. 

Although pre-training growth is slowing, and compute is shifting from training to inference, the compute demands from post-training scaling and test-time scaling, and increased usage suggest that the world likely needs a lot more AI-focused data centers, and the ramp from US$300 billion to US$400 billion in 2025 to roughly US$1 trillion in 2028 is directionally realistic, according to one Deloitte estimate. 

So the future might not include less need for high-performance computing facilities. 

On the other hand, what technology market has not evolved over time to patterns with just a handful of market leaders?

So if the independent neocloud provider market follows the historic pattern, market consolidation will happen, with a handful of major, scaled neocloud providers; traditional hyperscalers, plus a long tail of smaller, niche players.

Some argue the process has already begun. 

But there are other possibilities as well.  The neocloud provider market might not consolidate but instead collapse.

The "Big Three" hyperscalers possess massive scale, deep financial resources, and comprehensive service portfolios, allowing them to engage in price wars and continuously innovate at a pace the smaller players cannot match. So some would argue this creates immense and unsustainable pressure on the neoclouds' margins and ability to compete effectively in the long term.

Without a genuinely-unique value proposition or niche, the independent neocloud providers might struggle to retain customers who often prefer the security and breadth of services offered by the large providers.

The hyperscalers also will be better positioned to handle the likely higher regulatory costs, ability to attract talent and risk aversion of enterprise customers, as well. 

A collapse scenario might happen for at least some providers if customers abandon the neoclouds because of longevity fears. The danger cannot be dismissed. 

That happened around the turn of the century to many would-be capacity providers and competitive local exchange carriers. 

In the late 1990s, driven by the Telecommunications Act of 1996, which opened markets to competition, hundreds of new companies rushed to build wide-area optical fiber networks and local access facilities.  

This resulted in a vast oversupply of "dark fiber" (unused capacity), with estimates suggesting 85 percent to 95 percent of constructed fiber went unused after the bust. 

The industry and investors widely believed demand for bandwidth would grow indefinitely, leading to an investment frenzy based on the mentality of "if you build it, they will come". Actual demand and revenue growth, however, did not keep pace with the rapid network construction, creating an unsustainable business model for many.

CLECs and fiber providers were able to secure massive amounts of funding through debt and speculative equity offerings. When the broader stock market began to decline in 2000, this financing dried up, immediately pushing heavily leveraged companies into bankruptcy.

Hypercompetition and Price Wars: The presence of too many competitors in the same markets led to vicious price wars that drove down bandwidth prices (in some cases, by 60 percent per year), making it difficult for many new entrants to become profitable or even cover their costs.

In that case, rational merger activity did not drive the consolidation. Instead, the sectors mostly collapsed into bankruptcy. It’s impossible to tell, today, which of these outcomes develops. Over-investment, over-capacity and inadequate demand have happened with many earlier technologies, including railroads in the nineteenth century; the telecom and internet bubbles of the late 1990s and early 2000 era.

Will AI Propel New "Cognitive Elites?"

Though we might speculate about artificial intelligence impact on social stratification (the notion that “cognitive elites” could form), it might be more accurate to argue that cognitive abilities will still operate as a social mobility enabler into the upper middle class, much as they do now, but not so much a platform for new entrants into the “upper class” (top couple of percent of families and persons). 

Then there is the matter of “celebrity wealth,” which arguably does not confer membership in the “one-percent” ranks of the upper class, which is based on more than mere wealth. In other words, we might argue that wealth alone does not equate to full membership in the upper class. 

Such mobility across class categories will still exist, but the propelling forces might not necessarily be primarily based on cognitive abilities and AI skills. 

Driver	Estimated Importance (Variance Explained)	Key Evidence/Notes
Inherited Wealth/Status	High (40-50%)	Intergenerational transmission; 50% income inheritance; dominates top 1%.
Cognitive Skills	Medium (20-30%)	Strong for education/occupation but plateaus at elites
Post-Graduate Degrees	Medium-High (25-35%)	Gatekeeper for upper middle class
Occupation/Power	Medium (20-25%)	Tied to education; gig economy erodes middle class.
Race/Gender	High (overlaps with others, adds 10-20%)	Amplifies gaps

So claims of a "new" meritocracy driven by cognitive abilities and advanced degrees must be taken in context. Such factors are significant but often secondary to inherited privileges, particularly at the upper echelons.

That does not imply that social mobility into the upper middle class is unaffected. Indeed, cognitive abilities, as “measured” by earning of post-graduate degrees, will remain a key way people experience social mobility into the upper middle class. 

But it arguably overstates the importance of AI skills in creating a new meritocracy. We’ll keep the old one, in all likelihood. That still is important for social mobility into the upper reaches of the middle class.