Jonathan's End of May 2026 Portfolio Review

Investment Analysis Clubs Saul’s Investing Discussions
1

I feel I have to start with the exclamation “Wow!” What another incredible month May turned out to be once again. After the huge gains in April I was expecting May to be perhaps a little less exciting - but after a stellar earnings season for all of my companies (apart from Micron which is yet to announce, but I expect it to be stellar) my total portfolio rose another 51% in the month of May alone! That puts my Year to Date total portfolio growth so far at 80.3%!

Here are my results to date

  • 2024 + 70%

  • 2025 + 117%

  • 2026

  • January -5.1%

  • February -4.1%

  • March -1.3%

  • April + 33%

  • May +51%

YTD …+ 80.3%.

As you will see from above, I have already bettered my second best year of all time, which was 2024, and we still have half of the year to go. The AI boom does not appear to be slowing down in any way - quite the contrary in fact - and all of my stocks are direct beneficiaries of and catalysts for all of the exponential growth we are seeing and enjoying.

Here are my current allocations at the end of May 2026 (all long):

  • Nebius (NBIS): 33%

  • Astera Labs (ALAB): 16%

  • Micron (MU):14%

  • AppLovin (APP): 12%

  • Celestica (CLS): 11%

  • IREN (IREN): 9%

  • Electro Optic Systems (EOS): 5%

  • Anterix (ATEX): 2%

As you will see I have made very little changes to my portfolio yet again. In fact over the last 12 months or so I have actually made very few transactions at all. I am certainly not a trader!

Nebius continues to be my number 1 allocation by a long way - and it has been for the last 18 months or so now. As you will have read on the board earlier this month, when it was up well over $200 I did sell around 40% of my holdings in it - but then, as has happened once before, I instantly regretted it - and so I was watching it constantly for it to fall back below 200 again - which it did - so I bought them all back (and then a few more). I was so glad I could rectify my mistake and get fully back in again.

The only time I did sell a chunk of Nebius without buying them back again at a lower price was back in the Autumn of last year when (because it had grown to over 50% of my portfolio at the time) I sold a chunk of them to buy Iren. However, I think I would have done much better just to have left them in Nebius. Certainly Nebius has grown much faster and stronger than Iren has in this time frame. When I first bought Nebius in January of 2025 it was less than a $6B company. Today it’s MC is nearly 10 times that at just over $59B.

As I wrote earlier in the month in my post on Nebius earnings - I thought they had another incredible earnings with YoY growth of 684% and Adj EBITDA of 129.5mm, which was 341% YoY and 32% Margin.

Then they announced their acquisitions of Clarifai AI and Eigen AI - which the market responded to very positively. Then just after earnings we had the announcement of their deal with Bloom Energy for $2.6B for Bloom to provide power for their DC’s. This looks like it will remove many of the issues that local communities have against DC’s - and should mean that Nebius will be able to speed up moving from contracted power to connected power and then to active power.

Then just two days ago we had the positive news that a Hedge fund led by Leopolod Aschenbrenner, had disclosed a 5.6% stake in Nebius. This sent the SP up another 10% so it is now virtually back at it’s ATH again. I am so glad I discovered this stock so early because it has produced life changing returns for me already.

It has nearly been a 10x for me so far, and I can actually see it making another 10x again over the next few years - which would take it to a Market Cap of around 600B. It might take a little longer this time, but I can see it getting there sooner than we might think - which is why I continue to make it my number 1 holding.

The only big change I have made in my portfolio all of this year was back in January when I sold all of my Nvidia and I put all of the proceeds into Micron. And boy, am I glad I did! Micron has just now reached the $1T club - and has nearly 3 x’d since the start of the year – and has nearly doubled in the month of May alone. And it is still very undervalued on so many levels - especially in terms of it’s FWD PEG ratio of 0.12!!! It has grown to my number 3 position - only just behind Astera Labs.

ALAB has shot up 81% in May alone. It seems it is now being discovered again by the market. If you haven’t yet read wpr101’s excellent summary of their MS conference earlier this month then please do - as it will give you great confidence for this stock and company.

APP is my second favourite stock, behind Nebius, and along with Celestica is my longest held stock for over 2 years now. When they open Axon up in full general release for public sign ups in June (in just 2 days time now!) I expect significant ad revenue to flow through their platform. Adam, CEO, said this:

For 14 years, we have been a closed platform. Come June, advertisers across the world will be able to sign up for Axon and start running campaigns. That changes the trajectory of this company in a very meaningful way.

I think we can continue to expect new ATH’s for APP in the near future - back in the 700’s again. And beyond.

Here are the YTD gains of my current holdings since January. Apart from Anterix that I have only held for a month, all of these stocks have been in my portfolio all year or longer.

  • Micron (MU): up 240% since Jan 2026

  • Nebius (NBIS): up 176% since Jan 2026

  • Astera Labs (ALAB): up 106% since Jan 2026

  • IREN (IREN): up 68% since Jan 2026

  • Celestica (CLS): up 30% since Jan 2026

  • Electro Optic Systems (EOS): up 25% since Jan 2026

  • AppLovin (APP): down 9% since Jan 2026

  • Anterix: up 41% (since the end of April when I bought it.)

On a personal level, and well off topic for this board, I am now in the happy position of deciding what to do with all the growth I have. I am 57 years of age and am now more than able to retire if I choose, or to go part time (I enjoy working), or to pay off the mortgage (though my stocks are growing much quicker than the mortgage is!). I could not have even contemplated these decisions without the stock market - and had I not found Saul’s Board in the Summer of 2020 I would not be in this happy position today.

It’s worth pointing out that I joined Saul’s board half way through 2020 and so I missed a lot of the gains of 2018 - mid 2020. Then, like everyone else I lost almost 80% of everything in the Saasapoclypse of 2022, and I thought I would never recover. I had been investing for 15 years or so at this time and nearly 80% of everything was gone!

So I took some bold moves. I found Supermicro very early on and made it a huge position for me. Then I made Nvidia a huge position, and then 18 months ago I made Nebius my number 1 position. It is these bold choices and allowing a stock I really believed in to become a massive allocation in my portfolio that has brought me into this happy position today where I can ponder the merits, or otherwise, of giving up work. But until I decide what to do, I’ll continue to enjoy holding all of my stocks and looking forward for still more growth in the months to follow.

Best wishes to all for the month of June

Jonathan

98 Likes
2

Really great call on this move here from NVDA to MU back then and recognizing the opportunity in Micron!

As you mentioned you usually do not to make a whole lot of big portfolio changes month to month. It stands out even more then, that you swapped a winner out to make that move.

Celestica has been a great call as well. I had that name on my radar going back to the Supermicro ramp up. Back then, I thought their metrics did not compare as well to SMCI. However, maybe their management is a more consistent or stable hand. I may have factored that their legacy business was going to be a bigger drag than it turned out to be. It really seems like the AI side is taking off.

Thanks for that shout out on the Astera Labs write up by the way!

32 Likes
3

Just to be clear, this is Aschenbrenner now investing in all 3 of the NeoClouds - he previously owned and still owns CoreWeave and Iren. My guess is he’s decided he can’t really pick a winner from among them, so decided to own all three.

But, as I’ve posted elsewhere, I wonder what xAI’s entry into the GPU rental space, coming in as the #1 online capacity provider, means. Will xAI do more deals, or will Elon still want to reserve capacity for his second-tier LLM Grok? Right now, there appears to be enough demand that all these companies, if they manage their business properly, should do well.

12 Likes
4

NBIS is my favorite company but when I see this degree of optimism it tells me we’re in perilous, bubbling waters.

“It has nearly been a 10x for me so far, and I can actually see it making another 10x again over the next few years - which would take it to a Market Cap of around 600B. It might take a little longer this time, but I can see it getting there sooner than we might think - which is why I continue to make it my number 1 holding.”

Correct me if I’m wrong, but casual Google searches I just did say…

15 companies in history have had market caps over $500B
When Apple crossed $500B around 2014 revenue was $187B
To build such a titan has always taken companies decades - though OpenAI, Space X and Anthropic are topping $500B faster.

Here are some wildcard thoughts on NBIS 10x-ing again in a few years. This means they will add $540B to their market cap. Currently they have gigantic deals with MSFT and META that won’t happen again. So they will need to generate probably over $150 billion in revenue without hyperscalers. And what if…

  • recession cuts spending on AI
  • hyperscalers come hard after NBIS prospects, undercutting price
  • corporations decide AI is not as valuable as they thought and curb spending
  • social unrest including even physical attacks on DCs slow development
  • politicians during election years make tech/DCs the bad guys
  • Arkady retires or God forbid has health issue (key man risk)

How many DCs do they have to build, people do they have to hire, money do they have to spend and how fast must they do all this to hit 600B in a few years? And this of course assumes no major hiccups managing such monumental growth.

I was talking to a successful CMO at a private equity company whose boss has demanded they figure out ASAP how to use AI to cut costs. I just thought that was interesting. While AI may of course be as transformational as people say, it struck me that as of this moment, here is a multibillion-dollar company that doesn’t even know how to actually employ the technology. I spoke to another young exec, a software engineer who says he struggles mightily to babysit his AI which still makes not just mistakes but ones that require days of work to fix.

This just seems very odd to me. I realize it’s anecdotal evidence but in 2014 everyone had an iPhone, Apple was one of the most prestigious brands in the world and everyone knew the value of their products.

I love NBIS - it could be amazing growth story. But this is all feeling frothy.

BD

24 Likes
5

Yes, point taken @BroadwayDan. I just got carried away with myself. I don’t need it to 10x from here. That was just a throwaway comment - so thanks for picking me up on it. Actually, I have no idea of course how big it will grow - but I will continue to hold it until the numbers from their Earnings Releases tell me otherwise, or until the narrative changes in some way, or until the thesis breaks.

Thanks

Jonathan

10 Likes
6

Absolutely. Whole glory of these boards is everyone staying real, honest, humble and helping each other stay levelheaded in crashes, huge runs and everywhere in-between. And you made awesome call on NBIS and maybe more importantly, played the hand perfectly. Congrats on that.

6 Likes
7

Space-X has been around for 24 years. OpenAI and Anthropic are indeed setting records.

Maybe, maybe not. Both those companies say they don’t have nearly enough available compute, with Microsoft waying they’d be making more money if they had more compute today. If Nebius can demonstrate getting data centers up and running quickly, they’ll probably get more business. Just look at what Anthropic paid for instant access as an example.

Yes, the easy money in Nebius has been made. There might be a double or even a triple to squeeze out from here, but more will be tougher slogging unless Nebius actually can ramp up AI software infrastructure as a business to something huge, which seems unlikely to me at this point in time.

Well, in 1999 one could look at Pets.com, eToys or WebVan and say eCommerce was struggling. Clearly, business acumen and smart adoption was key, and will be true with AI. As for AI software development, that too is a skill that needs to be learned. What AI coding tool is your software engineer friend using, what do his context windows look like, what markdown files are he using, what’s his validation framework, etc?

Peter Steinberger used AI to code Open Claw without any other humans. It was vibe-coded, and is one of the fastest-growing open-source GitHub repositories ever. That said, he did spend serious money on tokens developing it, but he’s rich from selling a previous company and I guess this was his idea of fun. FWIW, he has been playing around with ideas for years, and Open Claw was something like his 43rd GitHub project.

10 Likes
8

Can I ask you @Smorgasbord1 if you’ve become a bit more positive on Nebius than you were, and if so, what has changed your view?

A few months ago at the end of last year in December, you were playing Nebius Bear to my Nebius Bull, and we had a long back and forth where you expressed surprise that Nebius was still my top holding. And I had to work hard to attempt to counter (not very well perhaps) the excellent points you were making.

Have your views changed now in that you think it might even 2 or 3x from here?

Jonathan

10 Likes
9

I don’t recall our discussion’s details, but if your bull case is still that Nebius will make a lot of money due to its unique and superior software infrastructure offering(s), then no, I’m not more positive.

For instance, in the Dec ER call, Nebius’ CEO cited Cursor as a customer of their “AI Cloud” software offering. But, with the Space-X/Cursor deal:

Cursor will be using xAI’s Colossus data centers and may not need/want Nebius anymore, although I have no proof either way at this point. The Associated Press thinks it’s possible that existing Cursor workflows may remain on Nebius, but that new ones will certainly be on xAI.

Clearly to me, whatever advantages Nebius’ infrastructure offered weren’t a large enough lock-in for Cursor.

I still feel this is simply a data center build out competition. What matters more is powering up GPUs rather than any software layered on top of them. All 3 of the NeoClouds have about the same online capacity (800-1,000 MW). Which company will get the power, cooling, and other build-out items for additional capacity completed soonest? Which company has the lowest internal costs? I worry that Nebius is paying engineers for something the market hasn’t acknowledged it really needs.

I’m not in a position to judge how well these companies are doing with their build-outs. I’m not privvy to the real information, don’t know what the specific issues with each site they’re dealing with, and don’t live close to any of the sites to do my own due-diligence/snooping. Data Centers are certainly in demand, and big companies have shown they’re willing to commit to big Capex to get data center capacity lined up. But, I can’t tell which of the 3 will do better than the others, nor what the real time-frames are.

At $250 a share, 100+P/E, etc. if I still had NBIS shares I’d be selling them today. But, clearly I sold what I did own too soon.

14 Likes
10

Thank you for your thoughtful reply @Smorgasbord1. I may if course be very wrong and I may wish that I had sold at these levels. But I know that every time I’ve sold Nebius I’ve always deeply regretted it and have bought them all back again. I did sell loads last December because I was influenced by our discussions, but then I bought them all back the next day. I’m so glad I did. I do think Nebius is going a lot higher from even here. It has ambitions to be a hyperscaler. Arkady and his team have already done this with Yandex, which was before Google. And so far they have continued to do everything right with Nebius.

I think you’re framing Nebius a bit too narrowly as “just another data center build-out story.”

I agree that power, GPUs, cooling, and execution matter enormously. If Nebius can’t bring capacity online efficiently, none of the software matters. But I’m not convinced the conclusion is therefore that the software layer is irrelevant.

What attracts me to Nebius is that they’re trying to build a full-stack AI platform as a hyperscaler rather than simply renting GPU capacity. The long-term vision seems closer to an AI-native cloud than a traditional colocation business. Infrastructure, orchestration, managed services, training, inference, data tooling, deployment, and developer experience all sit on top of the hardware.

On Cursor specifically, I’m not sure that proves what bears think it proves. But once again I may be wrong. But I think that if you’re Cursor and suddenly have access to xAI’s massive Colossus infrastructure through a strategic partnership, you’d want to put workloads there. I don’t see how any independent provider would realistically compete with that. That says more about xAI’s unique position than it does about Nebius’ software.

The real question is whether Nebius can continue attracting new customers and expanding wallet share across its platform. I think it can and will.

Where I differ from the pure data-center thesis is that I don’t think AI infrastructure will be won solely by whoever plugs in the most megawatts. If that were true, every provider would eventually become a commodity. Yet historically the biggest winners in cloud have been the companies that combined infrastructure with software, tooling, and developer ecosystems.

Could Nebius be overinvesting in software? Sure. That’s a risk. But if management is right, those engineers aren’t a cost center—they’re building the layer that eventually differentiates Nebius from companies that are simply leasing GPUs.

Anyway, thanks for the discussion.

Best,

Jonathan

17 Likes
11

[quote=“jonathan1, post:10, topic:124965”]
It has ambitions to be a hyperscaler.
[/quote]

That’s the scariest thing you could have said to me about Nebius. It wants to compete with companies like Google that are spending 2-3x Nebius market cap a year on capex.

Drew

12

Let me clarify, I don’t think Nebius is trying to compete head-on with Google, AWS, or Microsoft in the traditional hyperscaler sense.

They’re not trying to build a cloud that serves every enterprise workload on the planet. They’re focused on a very specific segment: AI infrastructure and AI-native cloud services. They intend to build a full stack software integrated AI platform.

Jonathan

8 Likes
13

It’s no secret that Nebius’ stack is itself built on top of open source frameworks. That said, Nebius does provide some optimizations in their value added software.

The question remains whether Nebius is providing enough unique value add on top of open source that companies doing AI development will find Nebius’ software offerings to be worth the lock-in (and perhaps cost) versus using the open source tools directly. This is an age-old question in cloud hosted software. It was a decent advantage for AWS in its early days, but that hasn’t stopped Azure from not only capturing/stealing some of AWS’ business, but also now growing faster than AWS:

Screenshot 2026-06-05 at 9.08.41 AM
Screenshot 2026-06-05 at 9.08.41 AM1754×1018 170 KB

It also appears that Nebius is heavily into the Nvidia eco-system, having written optimizations in CUDA, and leveraging InfiniBand. That’s fine, but it does lock Nebius into that.

AWS gave its customers the ability to scale hardware resources up/down instantly and very easily, and provided many convenience and security features. That was important for end-company, small teams, and perhaps less sophisticated teams - and still is. But, rIght now almost all AI development teams can be characterized as “Deep Tech” in that they’re highly sophisticated. These aren’t end-user companies building their own models and inference stacks - that’s still years away.

There aren’t even very many AI application companies around - such as Harvey in the legal world.

Maybe Harvey is a good example here. The company provides an AI agentic enterprise-grade workflow and reasoning engine. It leverages the frontier AI models from Anthropic, OpenAI, and Google, building its own agentic workflows on top. It also integrates with LexisNexis. It’s built exclusively on Azure, however, leveraging Azure’s Kubernetes Service (AKS), Azure’s Container Registry handle orchestration and hosting, Azure Database for PostgreSQL, and Azure Blob Storage, plus all the standard Azure user and (especially) security services, with tight Office 365 integration.

Can Nebius compete with that? Note there are far more than AI tools here; there are the legacy cloud services for process and user management as well as security compliance (SOC and HIPPA and ISO 27xxx) that all the hyperscalers provide, as well as things like integrated databases. Yes, Nebius offers a managed service for PostgreSQL, but Snowflake doesn’t run on Nebius. If you want to use MongoDB you need to self-host it on Nebius.

At the end of the day, it can all be done on Nebius, of course. The question I have is whether Nebius can replicate all the infrastructure that application companies will want, and then whether they think locking in to Nebius is preferable over locking in to the big players.

Although perhaps the biggest question today is around Nebius literally competing with itself. Microsoft has many advantages in the AI world today in terms of software infrastructure, existing relationships with many captive customers, user management, security compliance, etc. The one thing it doesn’t have enough of (as Microsoft management says in every ER call) is AI hardware.

That’s what Nebius is providing to them, to the tune of billions of dollars. How is Nebius not selling their soul? If Nebius really had a software advantage, they’d couple that with their hardware build-out capabilities and win business. But, they’re not. And you gotta admit it’s not a good look when they call out a marquee customer in their own ER call who then switches platforms a few months later.

In its early days, Tesla did contract to build EV drivetains for Mercedes and Toyota, so in essence competing with itself. But, Tesla really needed the money, and those OEMs weren’t serious about the product (they were just doing government-mandated EV compliance). The same can’t be said of Microsoft, who is laser-focused on expanding Azure’s AI capabilities.

Here’s my base case: The vast majority of Nebius’ revenue remains in selling bare metal to the hyperscalers (eg Microsoft), AI frontier labs (eg OpenAI), and large AI application users (eg Meta/Facebook). This is who and what they have contracted with thus far, and will remain their primary business by a long shot - perhaps 80%-95% by revenue. Now and in the future.

The question is whether:
a) Nebius can build up enough software infrastructure that companies will want to leverage, and that wins business over the legacy software infrastructure capabilities already provided today by the hyperscalers.
-AND-
b) There are enough companies that are large enough that this business affects the bottom line in a significant way. There aren’t very many of these companies today.

And what I see is that companies like Cusor can move off Nebius’ stack relatively easily, and that there aren’t yet many AI application companies (like Harvey). Nor whether those companies will see what Nebius is offering as high value enough.

The best-case scenario I see is that Nebius develops a useful stack, stays alive for half a decade selling bare metal, and by 2031 enough new AI application companies have started that see Nebius’ stack as a quick start for their own development. But what I see in NBIS’s valuation today is that much of that half-decade out future success is already priced into the stock.

Of course, I could be wrong.

21 Likes
14

I think I’ll make that my signature for all my posts!

4 Likes
15

In other news, xAI just inked another bare metal deal. This time with Google:

That’s now over $2B/month in NeoCloud revenue for Space-X. Ramping up for a couple months for Anthropic and through Sept for Google, but still.

The $920m/month for access to 110K Nvidia GPUs & CPUs. Rough napkin math is, I think, about 125MW, so again a high price for quick access. There’s a xAI performance cancellation clause in there in case xAI doesn’t have them all online soon enough. Interesting that Google wants more compute, even if it’s not their own TPUs.

IREN and NBIS both down in the teen percentages. CRWV down 9%. Is this a buying opportunity or indication that the NeoClouds just aren’t/can’t move fast enough?

7 Likes
16

All of my stocks are well down today. Possibly linked to the IPO, more likely linked to the jobs report and inflation higher for longer, and most probably a lot of profit taking after such big runs of late.

10 Likes
17

Something interesting to follow with the Nebius Cloud is their pricing structure. As I understand their Cloud is made primarily for AI related applications and not for general purpose apps as much as AWS/Azure/GCP.

Nebius offers much lower pricing then AWS currently on some AI related workloads. What I am skeptical on here is that Nebius is doing this profitably though. For Azure and GCP, they had to offer businesses an incredible amount of free credits to make it worth their while for businesses to switch.

I will say even though I’m skeptical, if they are able to pull that off by offering much lower prices and finding profitability it will be a remarkable feat. I’m skeptical because the business hasn’t shown cost discipline before in their other startups. However, a lot of skeptics of Amazon originally thought they would never monetize too.

I’d also be curious on the TAM for AI specific startups. Obviously this segment is growing incredibly fast, but in what proportion to general purpose apps on AWS. It sounds like some startups can mix and match AWS and Nebius Cloud, and Nebius looks to avoid vendor lock-in, so I think it has a lot of potential.

14 Likes
18

Yes, as always you make great points. I don’t have all the answers - far from it. I guess the things to continue watching are for every billion invested how much profit eventually comes back - ie ROIC. If we see over the next few q’s revenue growing faster than capacity, stable or improving gross margins, large repeat customers, increasing utilization rates, management hitting build out times, and clear evidence customers are using more and more of the software stack and not just renting GPUs, then I am sure that Nebius will continue to surprise to the upside.

So far, I feel that management have done an incredible job - so I am happy to continue to let them get on with it and for Nebius to continue to be my largest holding by a long way.

Best,

Jonathan

7 Likes
19

Good observations, but there’s a lot to unpack here.

First, here’s some pricing data for the “AI Cloud Service” aspect (not the bare metal deals):

And while Nebius undercuts the likes of Amazon, there are companies that undercut Nebius:

But note that they’re not “enterprise scale” offerings.

But remember, those per-hour prices don’t cover other expenses. For instance, Amazon charges to move data out of its data centers trying to force customers to do all their computing on AWS. That said, Nebius customers also encounter additional charges not covered in the per-hour levels: For instance, use of hyper-fast network storage, premium block volumes, and, of course, Nebius’ managed software tools.

I suspect that running a pure AI data center is cheaper than running a do-it-all data center. If your application has lots of AI compute, but also needs to offer traditional CPU/database oriented services, you might not be able to do it all on Nebius, and then the integrations and data exchange costs and hassle probably come into play.

The free credits from Microsoft and Google are essentially their way of paying the egrees fees charged by most providers (especially Amazon). Since switching costs are high, their goal is to capture customers and then lock them in.

My problem with trying to evaluate Nebius is that they don’t split out public end-user retail revenue from their bare metal megadeals, instead consolidating almost everything under a single segment called Nebius AI Cloud.

Right now, the demand for GPU-hours is off the charts. Nebius has no idle capacity. Of their customers that do use their software infrastructure, some are heavily tied into that and some are not. According to Gemini, their customer Shopify is heavily tied into the Nebius infrastructure, while customers like Mistral and Black Forest Labs are not locked into Nebius’ software and could easily switch to other providers should they have GPU-hours available. Some of this depends on what the customers are doing: those doing training of foundational LLMs don’t need it, those mid-market companies doing applications (inference mostly) might not have the engineering expertise to load balance, etc. and so would want it.

It might also be worth pointing out that Nebius has been acquiring companies to develop its software stack: Eigen AI and Clarifai are two recent acquisitions are two recent examples. How this jives with Nebius’ large internal software team is unknown.

At any rate, until Nebius breaks out its software infrastructure user revenue from its bare metal revenue, the company remains a black box to me and I don’t see how I could model its future.

23 Likes