What Future for Neoclouds?

Investors arguably have lots of reasons for wondering how long the neocloud market opportunity lasts, and which parts of the business case are more durable, assuming demand and revenue growth hold up. 

The simple “we have graphic processor units available when the hyperscalers do not” opportunity is narrowing, but is not gone. And the market appears to be shifting from simple GPU “scarcity” to other dimensions. Scarcity created the market opportunity, but enduring businesses need more than bare-metal GPU rental, according to analysts at McKinsey. 

source: McKinsey

Opportunity	What creates the opening	Assessment
Scarcity-based GPU rental	Customers cannot obtain suitable GPUs quickly enough from incumbent clouds	The most vulnerable proposition as supply expands and customers gain alternatives
Wholesale capacity provision	Hyperscalers and AI labs purchase dedicated clusters from third parties instead of developing everything themselves	Still a substantial opportunity, but highly dependent on financing, delivery, and contract economics
Specialized AI cloud services	Customers value performance, software, support, sovereignty, or deployment flexibility beyond hardware availability	Potentially durable, but requires capabilities that are harder to build than a GPU fleet

Concern about a potential “AI bubble bursting” aside, observers do not agree about the demand-supply balance. Gartner argues “we have more compute capacity than we have demand,” arguing that commercial returns are not yet sufficient to justify the build-out. 

Analysts at UBS argue that compute remains scarce and points to rising rental rates for older GPUs as evidence.

Actual procurement commitments also show continued demand for third-party infrastructure. CoreWeave announced an approximately $21 billion expanded Meta agreement extending through December 2032, while Nebius reported in August that it had delivered all capacity tranches under its Microsoft contract and was progressing capacity supporting another Meta agreement, for example.

The bigger concern seems to be whether neocloud service providers can earn a sustainable profit, given capital investment requirements, the cost of financing and the revenue growth rates. 

 “Is compute still scarce?” Yes. But the issues are more “which compute, available when, for which workload, at what economically sustainable price and profit margin?”

A neocloud compute services provider  can have real customers, a growing backlog, and strong revenue growth while still struggling to earn an adequate return on capital.

Perhaps generally speaking, some models seem safer than others. Bare metal services built using borrowed money seem most vulnerable. Providers with guaranteed and adequate energy supply, value-added features or regulatory advantages might be better positioned. 

Provider profile	Relative exposure	Why
Undifferentiated GPU renter relying on scarcity pricing	High	Few defenses against repricing, lower utilization, or customer migration
Highly leveraged operator building before securing dependable demand	High	Financing and deployment risks compound one another
Wholesale supplier with strong counterparties and well-matched financing	More defensible, but contract-dependent	Demand visibility helps; margins and residual asset risk still matter
Operator controlling attractive power, sites, and delivery capabilities	Potentially durable	Can solve infrastructure bottlenecks that chip availability alone does not resolve
Specialized AI cloud with demonstrable software or operating advantages	Potentially durable	Can compete on customer outcomes rather than inventory
Sovereign or regulated-market specialist	Potentially durable	Location, governance, and independence can matter beyond price
Provider assuming old fleets will automatically find enterprise customers	Unproven	Residual demand is possible, but neither utilization nor pricing is guaranteed

All that noted, Gartner’s prediction that 80 percent of operating neoclouds will ultimately disappear simply recognizes the arguably-inevitable consolidation of the market. That happens with every new market. 

Computing market	Early proliferation	the shakeout	Later market structure or outcome
Commercial mainframes	1950s–1960s: IBM competed with Burroughs, UNIVAC, NCR, Control Data, Honeywell, GE, and RCA—the “Seven Dwarfs.”	IBM’s System/360 strengthened its platform position. GE sold its computer business to Honeywell in 1970; RCA exited in 1971. Computer History	An IBM-led market with a smaller set of competitors—not eight equally sustainable platforms. This is a case of concentration around an already powerful incumbent, rather than an initially level playing field. Computer History
Minicomputers	1960s–1980s: DEC, Data General, Scientific Data Systems, HP, and other suppliers developed smaller, less expensive alternatives to mainframes. ETHW	Competition from newer computing architectures eroded the category. Compaq acquired DEC in 1998; EMC acquired Data General in 1999; HP acquired Compaq in 2002. ethw	Many independent suppliers disappeared, but not because the original market reached a stable end state. The category itself was largely superseded. ethw
Personal-computer hardware	Late 1970s–1980s: Apple, Commodore, Tandy, Atari, Osborne, IBM, Compaq, and numerous compatible-system suppliers.  Doncio, U.S. Navy	IBM-compatible systems became the dominant standard; mergers and exits reduced the number of major independent manufacturers. HP acquired Compaq in 2002, and Lenovo acquired IBM’s PC business in 2005.	In 2025, Lenovo, HP, and Dell accounted for approximately 63.8% of worldwide PC shipments; the top six accounted for approximately 86.1%. Smaller suppliers remained, but leadership was concentrated. gartner
Smartphone operating systems	2000s–early 2010s: Symbian, BlackBerry, Windows Mobile/Windows Phone, Android, and iOS competed for users and developers. Wikipedia	Earlier platforms lost relevance as Android and iOS gained adoption. Symbian lost its leadership position; BlackBerry and Microsoft’s mobile platforms became marginal. weforum	Android and iOS accounted for 99.9% of worldwide smartphone sales to end users in 2017. This describes operating-system concentration, not concentration among handset manufacturers. weforum
Cloud infrastructure services	From the 2000s onward, the market developed a broad field of global, regional, and specialized providers. It subsequently became strongly led by AWS, Microsoft, and Google. srgresearch	The observable result is concentration at the top rather than disappearance of all smaller suppliers. Synergy continues to identify a second tier, including rapidly growing providers. srgresearch	In Q2 2026, AWS held 28%, Microsoft 20%, and Google 15% of worldwide cloud infrastructure services revenue—a combined 63%. A substantial remainder still belonged to other providers. srgresearch

Most capital-intensive businesses ultimately settle into a market structure that is highly concentrated. That might suggest eventual consolidation. But not all capital-intensive industries wind up concentrated. Sometimes, fragmentation persists, as in the cement and oil well businesses. 

Industry	Capital-intensive assets	Leadership structure	Illustrative leaders	Evidence
Semiconductor foundry services, global	Wafer fabrication plants, advanced process equipment, packaging facilities	Highly concentrated; dominant leader	TSMC; Samsung Foundry a distant second	Counterpoint reports Q2 2026 revenue shares of 73% for TSMC and 7% for Samsung Foundry. This concerns foundry services, not the entire semiconductor industry or all internally manufactured chips. counterpointresearch
Large commercial jet aircraft, global	Aircraft development programs, tooling, production facilities, certification investment	Highly concentrated; effective duopoly	Airbus, Boeing	The large jet airliner market has been characterized as a duopoly since the 1990s. The definition excludes treating regional aircraft, business jets, and every emerging competitor as equivalent markets. wikipedia
Cloud infrastructure services, global	Data centers, servers, accelerators, networking and power infrastructure	Concentrated leadership with a substantial competitive fringe	AWS, Microsoft Azure, Google Cloud	The three accounted for 63% of worldwide cloud infrastructure services revenue in Q2 2026: 28%, 20%, and 15%, respectively. Other providers collectively retained 37%. srgresearch
Container shipping, global liner capacity	Container ships, container fleets and supporting logistics infrastructure	Concentrated oligopoly	MSC, Maersk, CMA CGM, COSCO, Hapag-Lloyd	A September 2025 industry compilation puts the top five at approximately 65% of global container capacity. Capacity share is not the same as revenue share or concentration on an individual trade lane. freightender
Crude steel production, global	Blast furnaces, electric-arc furnaces, rolling mills and related facilities	Relatively fragmented globally	China Baowu, ArcelorMittal, Nippon Steel, Ansteel and numerous others	In 2025, the largest producer, China Baowu, produced 124.76 million tonnes against approximately 1.85 billion tonnes worldwide—roughly 6.7%. Very large plants and companies therefore coexist with low global single-firm concentration .worldsteel
Cement manufacturing, global	Kilns, grinding plants, quarries and distribution facilities	Fragmented globally	CNBM, Anhui Conch, Heidelberg Materials, Holcim and other international and domestic groups	Industry analysis explicitly distinguishes global fragmentation from substantially more concentrated local markets. A global classification should not be interpreted as evidence of intense competition around every plant. mckinsey
Cement manufacturing, local markets	The same production assets, serving an economically reachable customer area	Often locally concentrated	A mixture of multinational suppliers and local leaders	McKinsey estimates an average of only three to four competitors in local markets. This is an important example of the same capital-intensive industry being fragmented globally but concentrated where customers actually purchase. mckinsey
Oil and natural gas production, U.S. Lower 48	Wells, drilling and completion investment, gathering systems and processing assets	Numerous producers, with production weighted toward larger firms	Publicly traded producers plus a large private-operator population	EIA identified approximately 12,000 producers in 2025. Public companies represented only 2% of that population but produced 68% of oil and natural gas. The evidence demonstrates an extensive producer base, although it does not by itself establish a top-four or top-ten concentration ratio. eia

And sometimes local concentration coexists with global fragmentation. 

Structural consideration	What the comparison shows
Market boundaries	Cement looks fragmented globally but concentrated locally. Defining the market correctly can reverse the classification. mckinsey
Technology and development barriers	Aircraft and foundries show that capital intensity can accompany sharply concentrated technological leadership.
Scale without complete dominance	Cloud infrastructure has three clear leaders, but their combined 63% share still leaves a meaningful market for other suppliers. srgresearch
Many firms versus evenly distributed output	U.S. oil and gas has thousands of producers, yet public companies account for most production. A high firm count does not imply equal economic importance. eia
Large physical assets versus global company share	Steel demonstrates that enormous industrial facilities do not necessarily translate into a dominant global producer. worldsteel

Still, the neocloud opportunity has been based on compute capacity scarcity, which should ease over time, despite some concerns in the U.S. markets about the possible impact of political forces seeking to slow down or halt some construction of data centers. 

Whether the neocloud industry persists is a reasonable question.