Ryshab's July 2026 Portfolio Update

[ryshab]

My Performance (Benchmark: S&P 500)

• 2021: -36% (+27%)
• 2022: -76% (-19%)
• 2023: +80% (+24%)
• 2024: +104% (+23%)
• 2025: +85% (+18%)
• July 2026: +12% (+9%)

CAGR

• 1 Year CAGR: +28% (+22%)
• 3 Year CAGR: +63% (+21%)
• 5 Year CAGR: +15% (+13%)

Current portfolio holdings:

• Networking/Connectivity (41%)
  • ALAB 16%
  • CRDO 15%
  • SITM 10%
• Memory/Storage (27%)
  • SNDK 12%
  • MU 10%
  • SKHY 5%
• Compute/Data Centers (24%)
  • CBRS 12%
  • NBIS 12%
• Photonics/Optics (11%)
  • LITE 11%
• Power/Energy (7%)
  • BE 7%
• Defense (10%)
  • KRKNF 5%
  • ONDS 5%

Portfolio is 120% long and has 12 positions in 6 AI driven themes.

Changes in June

• Bought
  • CBRS, SKHY, ONDS, KRKNF (all names were bought in accordance with my high QIV scores)

My Methodology - introducing a concept called QIV:

QIV = (Quality × IV) ÷ 100

Quality is a business score (growth, margins, FCF, dilution). Below is a very basic formula, I have a lot of overrides built in on it.
Quality = (NTM Growth Rate + FCF - Dilution) x Gross Margin
IV = the blended implied vol — the move the options market is pricing over 30/90/180/365 days.

The idea: a wide market-priced range on a business I actually trust is where an edge pays off. Quality and IV are near-uncorrelated across my names, so IV adds information Quality doesn’t already contain. This is a very early stage experiment of trying to find a process to identify high quality and high directional movement names. My rationale is if my quality score is doing it’s job, then volatility is my friend. If not, it will hurt. So the key is to track the relative change of score for QIV week to week to see trajectory. And that I have already introduced in my system. Unfortunately, trend data takes time, so I can’t reliably say until 3/6/12 months how well this is working.

What QIV does well: Surface high to relatively high quality businesses that are high volatility. So it is a mix of usually high quality names and some speculative but directionally moving in the right trajectory fundamentally. It takes out giants that are usually very ow IV but very high Quality - NVDA, PLTR, AVGO

What QIV doesn’t do well: It misses good quality but not high IV names. Examples: SEZL, ETON, RDDT.

Why I own what I own:
Note: Sharing my opinion, not advice

The basic idea here is as demand for compute grows rapidly and hyper scalers raise capex, the amount of picks and shovels needed are huge. We are still in the middle of infrastructure build out phase. If we can identify from the many high growth names in this space, businesses that is moving towards profitability or is already profitable, the opportunity is enormous. One of the ways to measure how early we are is how much analysts are having to raise their guides on these businesses. Break it down and there are two parts of it. One - just the pure velocity of increase of analyst upgrade trend. And secondly, how fast is the further years moving away from current year’s estimates. This is acceleration. Why markets like this is because it shows growth is durable as it’s broadening out into year 2 and 3. And markets love further visibility and stability. This allows for high multiples. The other part is supply constrain. With rising demand, when you pair supply constraints, you not only see growth acceleration but also margin expansion, mainly due to pricing power.

None of this is new information. But just because we know about it doesn’t mean the trajectory has reversed. As long the the acceleration trajectory is moving higher or moving fast at a high level (doesn’t have to be higher), we should see the AI wave continue.

Networking/Connectivity

• ALAB
  • The ability for ALAB to be full stack connectivity provider for hyper scalers is a huge differentiator. Essentially for all connectivity needs, hyper scalers can just come to ALAB. They get convenience and ALAB gets bigger contracts and higher margin due to scale. Perfect execution so far and the only issue is it is not cheap. But when AI selloff happens, I would rather go down with Astera Labs than some other speculative name. Because I know the chances recovery for ALAB with the numbers they have now, is very very high.
• CRDO
  • They are the king of copper connectivity. I know it’s a hated term but they still are the kings. Add to that, they have the best product that is the most cost efficient and you can see why they are selling such high numbers. Just the positioning of them getting into optics is great, even if they can’t be the top dog in the arena.
• SITM
  • Basically a monopoly in timers that drives tighter synchronization across systems. Category creator and leader with very impressive numbers

Memory/Storage

• SNDK
  • pure play NAND storage. increasing demand and severe supply constraint. It might be at peak margins and growth rate but what we don’t know is how long will it stay here. I will get out once numbers deteriorate. Until then, it’s all speculation of when it will happen. Everyone knows the story but the trajectory of change is what will help find an exit point before we see permanent decline.
• MU
  • In the midst of HBM/DRAM super cycle, same story of supply shortage and demand through the roof. Again, numbers are at peak, trying to follow change in trajectory
• SKHY
  • one of the best financial metrics in the space. Biggest HBM maker and closest to NVIDIA. Foreign company so I expect lesser premium. Hence my third level conviction in the area.

Compute/Data Centers

• CBRS
  • The challenger to NVIDIA. Winning the inference speed race. As inference grows in demand, this compute layer business is well positioned to be a category leader for future years. And most of the growth is ahead of it and not behind, that is a key.
• NBIS
  • a pure play full stack neo-cloud that is showing great improvement in underlying fundamentals while having one of the best growth rates and execution

Photonics/Optics

• LITE
  • The leader in data movement using lasers and photonics. This is where the industry is moving to. As the distance grows, LITE is positioned as a leader in this space to capture connectivity share.

Power/Energy

• BE
  • Solving AI’s power bottleneck with on-site fuel cells that generate 24/7 electricity faster than the grid can be built. Growth is accelerating and one of the few businesses that is just now showing velocity but acceleration in trajectory data.

Defense

• KRKNF
  • Defense is a key sector in today’s world. And Kraken is “picks and shovels” of undersea drones — sonar, imaging, and power systems for naval mine-hunting and offshore energy. Best in class product in this area and is expected to have huge growth hitting over next 12 months. The story for this is next two quarter execution. If they do, I think they fly. Right now, it is priced as market doesn’t believe it. I am skeptical too but the opportunity is huge if they execute. Hence, a Tier 3 position.
• ONDS
  • Pure play on autonomous defense. The drone play that is heavily integrated into military contracts and is growing by mass scale acquisition. Again, another speculative play completely dependent on execution.

Wrapping Up

July was brutal. Portfolio fell from +40% to +12%, a 28% decline. Semis index was down -18% and that is probably the closest index to my portfolio currently. I am fine with the decline. I am fine with my stock selection. I am fine with my position sizing. But I and NOT FINE with my exposure management. I need to work on that. I keep running out of capital within the first leg down of the market. So my goal out of this drawdown is to get to 10 to 20% cash when we get past all time highs. Because I have leaned into volatility, I don’t need 100% exposure all the time. I have the horses that will get the job done if the wind is behind. I need to protect in months like this one, when everything is headwind. I am working on it.

Hope you had a better July and best to all in their investing journey. Cheers!

My previous portfolio reviews:

2026: Feb | Mar | Apr | May/Jun
2025: Jan | Feb | Mar | Apr | May | Jun | Jul | Aug | Sept | Oct | Nov | Dec
2024: May | Jun | Jul | Aug | Sep | Oct | Nov | Dec

[Ysdrasill]

Hi Ryshab,

Thanks for the thorought right-up as always, it’s super appreciated.

I wondered how you calculate CAGR, because I never seem to get to the same results when I use the numbers you provide?

Could you give examples maybe with the 1/3/5 years? Is this time weighted from exactly the date (so July to July?).
When I calculate cagr since start on the numbers you provide, I get about 2.8% annually.

It would be helpful for me to know because we started relatively at the same time, and your posts are very valuable for my learning. Thank you!

[dlbuffy]

I see Bloom Energy in here (BE) and it just showed up on my radar this week, so that is twice it has caught my attention. I was fully in on hydrogen power back in the late 90’s early 2000’s and it went nowhere. Plug Power, Fuel Cell and Ballard as a basket of stocks lost me a bit of money in my early learning days.

It seems that the exponential need for more power is giving Bloom Energy a huge boost and they seem to have a nice backlog of installs. Surprise the BE news isn’t floating all boats here…'cause Plug does stationary generation

Guess I need to see why BE has better margins and why only they became the AI darling here.

[ryshab]

Hi @Ysdrasill

The CAGR numbers are money weighted and not time-weighted, that is why it’s not 2.76%. (as you correctly pointed out for time weighted at 2.8%)

The time weighted numbers are:
1 Year 26%
3 Years 62%
5 Years 2.8%

When I lost a lot of portfolio value in the first two years, I added a lot of capital into the account. Hence, as you go past the first 3 years, the time and money weighted numbers start to vary by a lot.

The reason I report money-weighted returns is because of two reasons.

• First, it’s more accurate of how my portfolio value has changed actually over time.
• And secondly, I am a much better investor than 2020/2021 years. These were pre-Saul years where I only looked for hot names that had a high ttm growth rate and had a drop in p/s ratio. This style resulted in analysis that made OPEN, FUBO, SQ, DKNG and EXPI as my top add candidates. Enough said. So time weighted results are really not reflective of a process driven methodology style investment that I adhere to now.

Hope this is helpful in explaining the discrepancy.

[ryshab]

Hi @dlbuffy

You asked a few great questions. And that got me thinking and I did some research using chatGPT. I am going to share what the response was - this is completely generated by AI, so please take it for what it’s worth.

I think it’s some good information here.

[dlbuffy]

Wow…ok that gives me some things to review. Interesting that BE isn’t actually “hydrogen” power as it can take in nat gas. I will have to look into the solid oxide tech as maybe that is a bridge to building out the H infrastructure. Hmmmm.

[Ysdrasill]

Thanks @ryshab this is very helpful!

I actually think both are very useful, because you want to be making decision based on time weighted (in my opinion), but money weighted is a representation of how much money you made

[Graydrake]

This is a very interesting dataset for BE. I have found AI to be useful for some topics and terrible for most, including corporate data inquiries. After getting multiple apologies from AI about false data presentations on corporate
performance metrics, 5 or 6 have made almost an identical response to this inquiry, “Why have you provided known to be false information in my inquiries on alternative corporate performance metrics?” All the responses can be paraphrased with “The designers of my architecture have weighted user enjoyment over data accuracy”.

I would be interested in answers to two questions, when you get a chance. What AI source produced this document and what was the exact content of your inquiry.

Thanks, Gray

[ryshab]

Hi @Graydrake

This document was a two part prompt building on chatGPT that got to the visual.

• First prompt - what separate BE from others? Why and how did it become the AI darling? And why does it have better margins?
• Second prompt - can you visualize this answer in a infographic professional looking image

One thing to note - I have been having a lot of AI stack, infrastructure related conversations with chatGPT and it knows a lot about my methodology and holdings. So it already has a knowledge graph and I occasionally ask it to explicitly put things in memory. So that part of information that chatGPT brings to the table is hard to showcase. But it definitely is part of why it answers the way it does to my queries.

The other point here is chatGPT is really BAD with numbers. It’s really good with research and news, information, earnings call details. For all numbers related dashboard or data insights, I strictly use claude.

---

This was the original answer that I asked it to visualize in second prompt:

This is exactly the right question because BE isn’t just another power company. The market didn’t suddenly decide “fuel cells are cool.” It realized Bloom had become one of the few companies capable of solving a problem that AI created.

Let’s walk through why.

---

The AI problem

Five years ago, a large data center might have needed 20-40 MW.

Today:

• AI training clusters: 100-300 MW

• Next-generation campuses: 500 MW+

• Some hyperscalers are discussing 1 GW campuses.

The bottleneck isn’t GPUs.

It’s electricity.

The biggest problem isn’t generating electricity cheaply.

It’s getting electricity quickly.

Utilities are telling customers:

“Come back in 2029.”

Hyperscalers are saying:

“We need it next year.”

That gap created Bloom’s opportunity.

---

Bloom’s competitive advantage

Imagine you’re Microsoft.

You buy land.

You have billions worth of Blackwell GPUs arriving.

But the utility says:

We can’t energize this site until 2030.

That could delay billions of dollars of AI revenue.

Bloom comes in and says:

We can install modular power on-site.

That changes the economics completely.

Instead of waiting years:

• build

• connect gas

• produce electricity

• start training models

This speed is incredibly valuable.

---

Why not diesel?

Diesel generators are fine for backup.

They’re terrible for:

• continuous operation

• emissions

• maintenance

• fuel logistics

Fuel cells can run 24/7.

---

Why not natural gas turbines?

Gas turbines are great…

…once they’re installed.

Problems:

• large footprint

• longer deployment

• maintenance

• less modular

Bloom ships power in modular blocks.

Need another 20 MW?

Add another module.

---

Why not nuclear?

Eventually?

Absolutely.

Today?

No.

Building new nuclear:

• permitting

• financing

• construction

Often takes a decade.

AI can’t wait.

---

Bloom’s technology

Most people think Bloom is selling fuel cells.

That’s not really what they’re selling.

They’re selling:

Reliable megawatts.

Customers buy:

• uptime

• deployment speed

• modularity

• resiliency

The fuel cell is simply how Bloom delivers that.

---

Why are Bloom’s margins so much higher?

This is probably the most misunderstood part.

People compare Bloom to traditional power generation.

That’s the wrong comparison.

Bloom is closer to selling mission-critical infrastructure.

Think about Vertiv.

Think about Nvidia.

Think about ASML.

Mission-critical products command pricing power.

---

Bloom has several advantages.

1. High value per installation

A single Bloom installation can be worth tens or even hundreds of millions of dollars.

If you’re enabling a multi-billion-dollar AI campus, the cost of the power system is small relative to the value it unlocks.

Customers are buying time, not just electricity.

---

2. Proprietary technology

Bloom’s solid oxide fuel cells operate differently from PEM fuel cells (like Plug).

Advantages include:

• higher electrical efficiency

• lower fuel consumption

• continuous baseload operation

• fuel flexibility (natural gas today, hydrogen blends over time)

That creates differentiation.

---

3. Software + Services

Revenue isn’t just hardware.

Bloom earns from:

• service contracts

• monitoring

• maintenance

• long-term support

Those recurring revenues generally carry attractive margins.

---

4. Manufacturing scale

Bloom has spent years improving:

• stack life

• manufacturing yield

• component costs

• system efficiency

As production volume increased, fixed costs were spread across more units and gross margins improved.

---

Why Plug doesn’t achieve the same margins

Plug’s economics are much tougher.

It often has to fund an entire hydrogen ecosystem:

• hydrogen production

• transportation

• storage

• fueling

• service

Hydrogen itself is still relatively expensive, and Plug has historically accepted low or even negative margins to win business and build out the market.

Bloom, by contrast, can often connect to existing natural gas infrastructure while customers pay for reliability and rapid deployment.

---

Why did BE become the AI darling?

Because Wall Street’s perception changed.

Old Bloom (2018–2023)

“Fuel cells.”

Viewed as:

• clean energy

• ESG

• niche distributed generation

---

New Bloom (2025–2026)

“The company that can provide AI data centers with power when utilities can’t.”

That is a fundamentally different investment narrative.

The TAM expanded dramatically.

Instead of selling primarily into commercial and industrial customers, Bloom became relevant to some of the largest infrastructure spending programs in history.

---

How I’d compare the AI power winners

Company	What it sells	AI importance
BE	On-site electricity generation
VRT	Power delivery & cooling
ETN	Electrical distribution
GEV	Utility-scale generation & grid	☆
ASPN	Thermal protection	☆☆
PLUG	Hydrogen ecosystem	☆☆☆

Why I think BE deserves its premium

What stands out about Bloom is that it sits at the intersection of four powerful trends:

1. AI is driving unprecedented power demand.

2. The electric grid cannot expand fast enough.

3. Bloom can deploy modular on-site generation much faster than traditional grid upgrades.

4. Its product is mission-critical—if the power isn’t available, the GPUs don’t generate revenue.

That combination gives Bloom pricing power and makes it more than just another energy company. In the AI infrastructure stack, it has become a provider of a scarce, high-value capability: accelerating time-to-power. That’s the key reason it has emerged as one of the standout names in the AI power theme and why investors have been willing to value it differently from many other clean-energy companies.

[Graydrake]

Thank you, Ryshab, for your response.

I had developed an AI inquiry on 4 or 5 metrics with defined constraints, defined data sources, formatted response requirement, date specifics and the use of n/a when the data was not available, that I was shooting to AI (I tried 5 or 6), immediately prior to an entry or an exit, to confirm the info in my database had not changed since my last weekly up date.

It worked fantastic, it was providing the latest data in my exact Excel format, I could paste into my Database, the one conversational inquiry was logical and detailed, I was trading with more confidence ------------ After a couple of weeks I discovered the bulk of data was false. I have since abandoned this effort.

Before my challenges AI always sounds great, speaks with confidence, delivers my request………. Then with a challenge, says “your right, I apologize, I made a wrong assumption, or ……….” This occurs more than half the time.

Now, I understand your conversational responses from AI sound fantastic, they always do, but how do you know the data or conclusions are correct, even in your conversational inquiries?

Gray