The first time I heard about neoclouds was from a friend suggesting I look at miners that were going to convert to AI/HPC about a year ago. Since by then I had held RIOT for nearly a year with zero performance I dismissed the idea swiftly.

However when Nebius, a reborn Yandex, appeared on my radar I dug deeper and bought a substantial position in Nebius at around $25 per share, largely due to the fact it was run by Volozh, a fiercely talented entrepreneur.

Although it seems crazy now, a year ago, the market was still shaking from the release of DeepSeek that would supposedly kill the need for AI compute through more efficient models that activated only part of processes needed instead of brute force methodology. Since then AI did get better, but the need for AI did not disappear.

This has led to neoclouds completely obliterating all price targets, with Nebius rising over 7x, although I did derisk my stake entirely at $65 per share, only keeping my profits in for the ride to $120. Miners shifting to AI data centers also received 10-15x runs.

The reason behind all that is it is assumed that hyperscalers like Meta and Anthropic will require an ever rising amount of compute and GPUs to function which will require more and more infrastructure. To that end, these neoclouds moved to secure land with access to electricity on GW levels and offer their services to hyperscalers to essentially build data centers to suit to accelerate their time to market.

Since data centers are essentially real estate development projects not unlike industrial real estate, many hyperscalers decided to buy up this capacity from the likes of IREN, Coreweave, Nebius, Wulf and others in deals valued in billions over periods of 5-10 years in order not to deal with the hassle. Some of these platforms like Nebius moved to build software to manage the data centers and virtually integrate.

Over the past year we’ve seen a numerous series of deals from AI hyperscalers as they rushed to secure AI/HPC compute, sending AI HPC stocks soaring with each deal in an echochamber effect.

The math behind these is quite simple. Neoclouds buy the land and secure access to power on the grid (which saves hyperscalers time to compute). Once a tenant is located that agrees to pay X billions for Y MW of compute, they do a deal where they pay an advance to the neoclouds and then a fee every year for the compute.

The hyperscaler then has to build the data center (including substations, cooling, etc) and purchase the GPUs that will be plugged in.

The GPUs are typically financed by a loan against the GPUs and the data center by the advance payment and equity from the neocloud or convert note / ATM raises.

The twist is, GPUs are amortised within 5 years, the average length of the contract, and these represent 2/3 of the cost of the data center, so at the end only the data center, substantions, cooling infrastructure remains which account for only 1/3 of the cost of the project. This means the next time a contract comes along you end up with 1/3 of the costs as net margin.

So over a 10-year period, the net operating income is around 1/6th of the revenue received on the contracts or around 17-18% of the project value. DCF methods over 15 years and applying a discount rate of 6% would give you about 20% of the project as NOI.

Of course from that you would substract additional development costs, SGA, financing costs, etc but you would still end up in the 15% APY range which for high quality deals that could be monetised if sold to data center reits to get cash at a higher multiple is definitely an interesting looking option.

Neoclouds have built a large part of their valuation around their secured power, meaning they encourage investors to look at their powered land portfolio, assume it will all be built out due to heavy demand and apply valuation models based on that potential future revenue.

This is becoming the reason as to why I have now fully exited my Neoclouds position.

First of all, the run up, ever since the Iran war has been almost 2x from the bottom and that is on top of a previous run close to 2x. In any case, this surge was due for a correction, which is now playing out.

However the real reason I am out of this sector at this moment is different - neoclouds are stuck in a land where their margins will always be subpar.

One one end, hyperscalers are going through neoclouds because it is cheaper for them than building their own data centers which requires more hassle, investment costs, etc which would dilute the performance of their metrics. It is therefore highly unlikely that neoclouds have any pricing power as the replacement cost will forever be the metric judged against.

This is in addition to the fact, many companies are trying to disrupt the powered land aspect of the offering by offering to build data sources for data centers independent of the grid. This would greatly weaken current neoclouds as well if any success is seen on that end.

Perhaps even more importantly, neoclouds need to secure GPUs from NVIDIA to run their data centers, without them a data center simply can’t operate.

Now the most valuable company in history, NVDA is actually deciding which companies to sell its GPUs to and which operator it will gives its latest chips to.

This is how NvDA got the $21 billion Meta contract to Nebius for example by giving it advance access to its Ruben chips, which it received as NVDA is an investor in Nebius. However Nvidia will always want to make sure its main profit centers are on the core holding, meaning margins from neoclouds are further challenged by the strongest supplier in the market.

Neoclouds therefore find themselves at an intersection where buyers have the bargaining power of replacement costs, its powered land assets are being challenged and its supplier is in de facto control of when it can actually build and what it can sell. All of these are happening at the same time.

None of this is priced into the price after the current run up. It eventually will be. I am watching for a pullback 20-30% in the neoclouds space and would say we are lower a year from now than we are today.

In 2026, the "Neocloud" market—specialized, high-performance cloud providers that challenge the traditional "Big Three" hyperscalers—has matured into a $42 billion industry. Unlike AWS or Azure, these providers focus almost exclusively on GPU-accelerated computing, AI factory infrastructure, and sovereign data requirements.

Here are the top Neoclouds leading the market in 2026:

1. CoreWeave

Widely considered the "gold standard" for AI infrastructure, CoreWeave has solidified its lead by securing massive allocations of NVIDIA’s latest hardware.

• Best For: Massive-scale LLM training and high-concurrency inference.

• Key Advantage: Direct access to NVIDIA Blackwell (B200) and Hopper (H200) clusters with native Kubernetes orchestration.

• 2026 Vibe: The enterprise-grade alternative for teams that find hyperscalers too slow or restrictive.

2. Nebius

A rising power in the European and U.S. markets, Nebius has positioned itself as the "AI-native" cloud.

• Best For: European data sovereignty and cost-effective GPU clusters.

• Key Advantage: Operates strategic hubs in Europe and North America, offering highly commoditized GPU access that undercuts hyperscaler pricing while maintaining Tier-3 data center standards.

3. Crusoe (formerly Crusoe Energy)

Crusoe has pivoted from crypto-mining to building "AI Factories" powered by wasted energy sources.

• Best For: Sustainable, high-performance computing.

• Key Advantage: Their Leviathan modular data centers can be deployed in weeks and are powered by stranded energy (like flared natural gas or renewables), making them the choice for ESG-conscious AI labs.

4. Lambda (formerly Lambda Labs)

Lambda remains the favorite for research-intensive organizations and academic institutions.

• Best For: 1-click GPU clusters and deep learning development.

• Key Advantage: Extreme simplicity. They offer a "on-demand" experience that feels like a local machine but scales to thousands of nodes, with pre-installed ML frameworks that just work.

5. Civo

Civo has successfully carved out a niche as the leading "developer-first" low-cost GPU cloud.

• Best For: Startups and independent developers.

• Key Advantage: Zero egress fees and transparent pricing. They are the only low-cost provider in 2026 that is fully ISO 27001 and SOC 2 certified, making them "production-ready" for regulated industries.

Comparison of Specialty Tiers

If you are looking for specific performance metrics, the market has split into these categories: