TL;DR
Consultants from McKinsey argue that GPU-focused 'neocloud' providers face commoditization unless they build higher-level AI services, yet doing so risks direct competition with the hyperscalers that account for much of their revenue. Firms such as CoreWeave, Nebius and Crusoe are cited as examples navigating heavy capital spending, customer concentration and strategic uncertainty.
What happened
Neocloud operators — companies that rent access to GPU-accelerated compute for AI workloads — grew rapidly as customers struggled to secure chips. McKinsey told the market this model is fragile because hardware access alone offers limited differentiation and often devolves into price competition. Investors in names such as CoreWeave, Nebius and Crusoe are banking on a shift toward higher-level offerings: training and inference tooling, sector-specific stacks and managed AI services built on top of their infrastructure. But building those software and services will take time, investment and the willingness to compete with hyperscale cloud providers that today supply much of their revenue. The report highlights CoreWeave’s concentration risk — 77% of its 2024 revenue came from two customers, with Microsoft accounting for 62% — and notes steep recent spending: $1.21 billion in Q2 revenue (up 207% year‑on‑year), a $290.5 million net loss and $2.9 billion of capital expenditure in that quarter. McKinsey sees three mid-to-long-term paths: niche specialization, serving startups as a growth runway, or consolidation via acquisition or failure as GPU supply tightness eases.
Why it matters
- Business models that rely solely on renting GPU hardware may be unsustainable as differentiation narrows and price competition intensifies.
- High customer concentration makes neoclouds financially vulnerable if a major hyperscaler changes strategy or sourcing.
- Shifting up the software stack could create direct competition with the very hyperscalers that are key customers, complicating partnerships.
- Heavy capital expenditure to scale capacity increases financial pressure and raises the stakes for choosing a viable strategic path.
Key facts
- Neoclouds are GPU-as-a-service providers that grew as demand for GPU compute outstripped direct chip availability.
- McKinsey characterizes renting hardware access as an inherently commoditized business with limited differentiation.
- Investors are backing companies such as CoreWeave and Nebius (both publicly traded) and Crusoe (privately held) to move into higher-level AI services.
- CoreWeave reported that 77% of its 2024 revenue came from two customers; Microsoft represented 62% of its revenue.
- In Q2, CoreWeave reported $1.21 billion in revenue, a 207% year-on-year increase, alongside a $290.5 million net loss and $2.9 billion in capital expenditures.
- McKinsey outlines three potential paths for neoclouds: niche specialization, focusing on startup customers, or consolidation/absorption by larger players.
- Some neocloud operators may fade as GPU supply catches up with demand, according to the analysis.
What to watch next
- Whether individual neoclouds succeed in developing differentiated software and managed AI services without alienating hyperscaler customers (not confirmed in the source).
- Significant M&A activity: acquisitions of neocloud operators by hyperscalers, telcos or other buyers (not confirmed in the source).
- How shifts in GPU supply and pricing affect demand for rented GPU capacity and the viability of rent-a-GPU business models (not confirmed in the source).
Quick glossary
- Neocloud: A provider that offers rented access to GPU-accelerated computing resources, often focused on AI workloads.
- GPU-as-a-service: A cloud-like offering where customers pay to use GPUs for computing tasks rather than owning the hardware.
- Hyperscaler: A very large cloud provider that operates massive, scalable data center infrastructure and offers a wide range of cloud services.
- Inference: The stage in AI deployment where a trained model is run to generate predictions or outputs from new data.
Reader FAQ
What is a neocloud?
A company that rents GPU compute capacity, primarily to support AI training and inference workloads.
Why do consultants say the model is fragile?
McKinsey argues that offering only hardware access is easily commoditized, leaving providers exposed to price competition and limited differentiation.
Which companies were cited in the report?
The story references CoreWeave, Nebius and Crusoe; CoreWeave and Nebius are noted as publicly traded and Crusoe as privately held.
Are these neoclouds profitable?
Not confirmed in the source for the sector broadly; the source does report CoreWeave had a net loss of $290.5 million in the cited quarter.
PAAS + IAAS 6 Rent-a-GPU neoclouds need to adapt or die as the AI market evolves McKinsey points out the quandary facing companies like CoreWeave Dan Robinson Tue 25 Nov 2025 // 18:12 UTC…
Sources
- Rent-a-GPU neoclouds need to adapt or die as the AI market evolves
- The evolution of neoclouds and their next moves
- Rent-a-GPU neoclouds need to adapt or die
- Neocloud Evolution & Future Outlook: An In-Depth Report on …
Related posts
- DJ Garman drops the ball instead of the bass in AWS re:Invent keynote
- AWS expands Storage Gateway hybrid cloud support to Nutanix AHV hypervisor
- OpenAI pitches ChatGPT as a gatekeeper for third‑party app integrations