NeoCloud rankings from SemiAnalysis

[ActonUp]

There seems to be a lot of love here for various Neocloud vendors, including Nebius, so I thought I’d pass along this excellent analysis of the many competitors in the market, done by the team at SemiAnalysis (Dylan Patel and friends).

“Neoclouds”, if you don’t know the lingo, refers to a hyperscaler that is focused on providing GPU-based services for rental to customers.

This is the 2nd iteration of their detailed analysis, including extensive testing of the most important features of each service, along with interviews with many NeoCloud customers.

At a time of significant technological change, this kind of analysis can be extraordinarily helpful to investors in the space because it helps to separate those who are “all hat and no cattle”, as the saying goes, from those who are truly delivering for their customers.

Using this kind of approach, you can sometimes discover clues to future quarterly earnings reports. In an earlier time of significant change in the technology industry, I was fortunate enough to be associated with a group performing similar evaluation and measurement work, which led to me investing in Cisco (post-IPO but at the IPO price), along with several other soon-to-be-discovered tech industry leading lights.

Here are their overall rankings:

I’m not an investor in any of the NeoClouds but I’m invested in Microsoft (Azure is ranked in the 2nd tier) and Amazon (AWS is ranked in the 3rd tier).

ActonUp

[dlbuffy]

Apropos of nothing much, but flash back to the dot com days and to see Akamai way down at bottom brings sadness. I believed in them SOOOOOO much as everything was booming. Never really made money and didn’t know anything about evaluation at the time. It was part of the MF story back in the dark ages of online investment forums.

Anyway, on topic, due to conversations here I had added coreweave to my watch list. Nice to see its ranking.

[coldmountain]

@ActonUp Thanks much for posting that! I didn’t realize just how little I understood about this industry before I read it. Extremely humbling.

For those who haven’t gone to the link (and are invested in IREN, Coreweave, Nebius, etc), I really recommend it. It is a monster report, and it helped me understand much more about what is meant by Nebius’ engineering/technical expertise. The report discusses all the companies shown in the screenshot (all the various tiers); regarding IREN, and why it did not make the grade in this review, it says:

IREN/Iris Energy

IREN is one of the most aggressive crypto mining companies trying to convert their facilities into a neocloud. Unlike their competitors such as TeraWulf, Core Scientific, and Cipher Mining, which have all realized significant value from their existing investments by pursuing the powered shell datacenter infrastructure business (aka colocation), IREN is intent on doing things the hard way and building a neocloud all by themselves, with no relevant experience on their team, now with nearly 100K GPUs committed for current and future customers.

We have tested IREN in March 2025 and found the service to be severely lacking, with multiple basic configuration errors on the hardware such as ACS not being disabled and GPU Direct RDMA not being enabled. In March 2025, our two node NCCL test on the AllReduce collective showed that IREN machines had around 129.27GB/s at 128MiB msg size when the Nvidia reference numbers and our testing for top tier neoclouds is well above >= 300GB/s busBW. It was later confirmed to us by IREN engineers that the root cause was due to their team not disabling the ACS setting on the system’s PCIe switch, which meant the GPU couldn’t talk directly to the NIC but instead had to go through the root complex of the CPU. While this should be a simple fix and checks or remediation is easy to automate with software, we have not been able to verify that IREN has made any changes. For this round of testing, IREN has claimed they have no capacity available to test for over 3 months straight.

Recently, IREN has had some success, signing a $9.7 billion offtake deal with Microsoft targeting a portion of their 750MW site in Childress, Texas. It is known within the industry that IREN offers below market rate prices compared to providers in the ClusterMAX Silver, Gold, or Platinum tiers. We think the reason is twofold:

• Cheaper-than-average cost structure, through ownership of the datacenter and site selection centered on areas with cheap power costs (typical in the Bitcoin mining business)

• Inferior service quality, relative to the market average

For a deeper analysis of the economics or IREN’s publicly announced AI cloud contracts, our AI Cloud TCO Model is the best tool. It is trusted by many major GPU buyers, as well as their financial sponsors.

[twillo]

My take is something doesn’t “smell” right here. Common sense argues that an enterprise of MSFT’s stature wouldn’t enter into a transaction with IREN if it had genuine cause to believe that these allegations were true. And they certainly wouldn’t make a $2B prepayment.

[wpr101]

The report seems to gear the rankings towards the developer experience and not towards a ranking by what is a good stock or investment.

For example CoreWeave has hired Galaxy to do their data center setup and build. Galaxy in turn is hiring multiple subcontractors, and those subcontractors likely contract out portions of the work as well. With this type of business model there’s simply no way for them to compete on margins in the long term.

The critique of IREN there says they want to build a neocloud all by themselves. It says there is “no relevant experience on their team”, as the energy part of the equation is seems to be counted for zero in this evaluation model.

The critique given seems to be entirely configuration based in that certain setting were not disabled at the start. They even say “this should have been a simple fix” and “remediation is easy”. What that means is these types of issues are not going to be big blockers for hyperscalers who want customizations anyways. This is probably a weeks worth of configurations to specs to get things up and running. However, the commentary is talking as if they are a startup company directly talking to IREN to get setup for a small allotment. IREN has said they are talking to customers looking for 100 MW allocations, and they just proved this out by signing Microsoft recently. It does seem crazy for them even knowing of the Microsoft deal to put them in the very last rated category.

I am also skeptical of polished reports which seem intent of giving tiered rankings like this as some sort of guideline. Altimeter publishes a lot of these types of reports as well through Jamin Ball, and I become somewhat suspicious how recent graphs posted of SaaS companies showed Snowflake and Rubrik as showing up the best. I know SaaS companies like Paymentus show up much better than either of those two companies yet was left off the graph. However, if you look up who Jamin Ball works for it’s Altimeter, and their largest holding is Snowflake. Rubrik is also a sizable holding as well, but Altimeter has zero shares of Paymentus.

In this case Altimeter has a $67M position in CoreWeave. Altimeter has no position in IREN or HIVE though which appear at the bottom. The author is associated with the Altimeter team, so I don’t think its too hard to figure out why CoreWeave is ranked number one above all others, and IREN is ranked the lowest.

[Chamonyx]

lol - anyone who thinks that this is a thorough and objective analysis of anything should check out the discussions relating to it on X.

I’m not saying there isn’t anything of value in there…..just suggesting caution in drawing any conclusions based on the ‘results’.

[Chamonyx]

Here is one example:

[coldmountain]

What I found interesting, and useful, about the paper was it laid out a series of objective and quantifiable metrics (and explained what each test was intended to be a proxy for) and then published its results. They also had a more subjective evaluation of the companies, based on their own interviews and user feedback. One could certainly quibble with the metrics chosen, or claim that the subjective opinions are skewed towards their agenda, but ‘I’ cannot. I don’t know enough to propose other metrics, and I don’t have any direct experience using any of the companies, nor do I know anyone who does. This being the very testosterone-fueled tech industry, it doesn’t surprise me that there are some boisterous discussions on X about the study’s methodologies, chosen metrics, evaluation of results, biases in subjective opinions, ideologies, font selection, graphic design…. Have I missed anything?

I also found it fascinating just how complex and difficult nearly every aspect of an AI data center is. It shed a bit of light on why companies such as ALAB or CLS have such booming businesses, and why they are likely to continue.

Regarding some of the comments above, the tests done on IREN’s DC were back in March, and according to the authors they failed the tests (not as a business, or as an AI data center, or anything else) for some pretty basic configuration problems. It sounded like the resolution was also straightforward, so it seems likely that IREN had that sorted out to Microsoft’s satisfaction by the time MS awarded it the contract (which has specific milestones, if I’m not mistaken). What was more interesting to me was the assertion that IREN is ‘doing it the hard way’. That raises questions in my mind as to what IREN’s particular resources are that makes this approach a good one? I’ve asked in other threads how a company with so few employees can make all this work, and this ‘doing it the hard way’ assertion does not resolve anything in my mind.

Regarding Nebius, it has been pointed out by @wpr101 and perhaps others that their supposed technological bench depth is often quoted but seldom described. This is the first time I have read something that touches on that, and it sounded impressive.

Finally, I gained a better understanding of how many different ways there are to build and run an AI DC company, ALL of which have substantial risks. It seems likely that there will be successful players of all denominations. The authors did not (as I recall) suggest which were good investments.