haYN Capital

Look for value where value can be found. posts NFA. please challenge any of my posts/ideas Check out my Substack:

[haYN Capital @killapabkai]

I do however think this is marginally bearish for neoclouds

Hyperscalers have better financing and engineering teams to achieve higher roi on data center CapEx

I still think it will be profitable for these neoclouds to rent compute based off the roi of their customers for deploying compute is high therefore they are willing to pay rates that are profitable for the neoclouds to buy, deploy, and lease hardware

Just marginally less so as they have more compute supply to compete with

[haYN Capital @killapabkai]

$META Sellng Compute

My Interpretation and Trading Plan

I think it is wrong to directly interpret this solely as ”compute is not scarce”

There are 3 main ways I personally interpet it this

1. Maximize Utilization

Utilization % of hardware directly affects how much revenue that hardware earns -> profitable the cost to acquire and deploy that hardware as its mostly a fixed depreciated cost base rather than variable.

Any downtime on a compute rack is unprofitable so if $META is not using a rack for an hour it makes sense to route it to an external customer.

2. Return on Compute

For $META now that they have maximized return on compute through maximizing utilization they have another way to maximize return on compute assets.

Internal Use or External Sales

Another way to Interpret their entry into selling compute is that selling compute on the market is more profitable than them using all of it internally.

Based off $META historically lagging with LLMs I believe that they have just realized that the ROI of selling compute is > using all of it internally.

3. Bear Case Compute is actually not that scarce.

I think the main view the market is taking is that META realized they have too much compute and they don’t need it all so they are trying to make up for their investment by selling compute on the market.

And if META realized they have too much compute then other Hyperscalers may realize this as well and all of them overbuilt on compute.

Based off of how much compute $META has and is expected to bring online a single company would never be able to consume all of it internally.

Training and Inference are 2 use cases for a data center.

Training: Who tf uses Metas LLM?

Just give up on the LLM and use training compute to train the ad models etc

So this frees up GWs they don’t need for training an LLM

Inference: $META does not need multi GW of compute capacity for internal inference/agent workloads

If one company needed that much compute for inference/agentic work then it would be too expensive relative to human labor hours.

This announcement is noise not signal.

What would be signal?

1. META cutting CapEx

Cutting CapEx would mean that the ROI on AI hardware is unprofitable for both internal use and selling it on the market.

2. Market Rates GPU/hr

Now that $META is using less gpu internally this means more supply available on the public market.

If market rates for gpu rentals drop significantly from here that means that there really is not enough demand to soak up that compute supply and compute is not scarce.

[haYN Capital @killapabkai]

I am going to look for dips to add because my view is that the ROI on inference/agentic workloads is very very strong

Given a fixed task the token cost to complete that task is multiples cheaper compared to human labor hours to complete the same task = High ROI on compute
@ExponentialView has a great research report and slides that are relevant to this
@SemiAnalysis_ also does have a nice post on the roi of their internal token spend

Given this I believe that if the roi on gpu/hr is this high it will continue to stay high for a while

[haYN Capital @killapabkai]

Why Neoclouds selling off on $META compute business news?

They saw that the economics of selling compute was so strong that they decided to sell some of it instead of trying to use it themselves (likely cause Llama model sucks)

If $META thinks the economics of selling compute is good why are the other sellers of compute selling off

[haYN Capital @killapabkai]

Ballpark numbers i got roughly 400m won asp per tester

560 * 400m =224B revenues

Assume same gm of 40% gm = 88b gp

Assume 50% increase in sga = 15b opex

73b ebit * 10x fy27 ebit = 700b valuation
/4.5m shares ~150k won PT

[Cash_Foxx @Simple_happyy]

The new plant will add an annual production capacity of 360 systems to the existing 200 systems at its Cheonan facility, increasing total capacity to over 560 systems per year. This expansion is expected to eliminate production bottlenecks and enable the company to respond more quickly to growing global customer demand.

[haYN Capital @killapabkai]

An under looked aspect of Mirae Corp $025560.KS is that production can be brought online immediately per the quote below.

“Notably, the new Asan factory is located about 10 minutes by car from the Cheonan headquarters, providing high efficiency in both production and logistics. In addition, key infrastructure required for semiconductor equipment manufacturing—such as a Class 10,000 cleanroom, air conditioning systems, and temperature and humidity control facilities—are already in place, allowing for immediate production without the need for additional large-scale facility investments.”

https://t.co/Tno14FhIIa

[haYN Capital @killapabkai]

Combine this capacity expansions news with the capacity ramps for DRAM and NAND fabs for 2027 & 2028.

Then a 20-25x multiple on FY26 earnings doesn’t look to far fetched if 2027 earnings are a double from FY26

*Referring to my last post on the FY26 model

[Cash_Foxx @Simple_happyy]

Mirae Corporation announced today that it will acquire additional land and buildings for KRW 14.5B to meet rapidly increasing customer demand. With its current production facility operating at nearly 100% utilization, the acquisition is expected to expand production capacity by 2.8x.

[haYN Capital @killapabkai]

@jukan05 Mirae Corp $025560.KS Test handler sales scale with CXMT wafer capacity

[haYN Capital @killapabkai]

Earnings growth has accelerated to FY growth of 24%.

85% of companies have beat estimates with the average beat being 16% over estimates.

And you are bearish anon?

[haYN Capital @killapabkai]

Yeah fair enough.

But I’ll add this as a sort of bull case

Looking at KWSPM (wafer starts) capacity being added in 2027 compared to 2026 it is very strong. 2028 is also strong.

The KWSPM numbers in 2027 supports a higher amount of equipment purchases than KWSPM capacity adds in 2026

This supports a possible continuation of another large growth year for FY27 which may make the 25x Forward FY 26 PE cheap on a FY27 PE

Additionally, there is a possibility their HBM testing product picks up traction as well.

If this bull case plays out FY 26 25x PE is a fair valuation maybe even cheap depending on the strength of growth.

But yes I agree in the base case a 15x earnings maybe 20x max may be fair value

[haYN Capital @killapabkai]

$025560.KS Mirae Corp FY 26E model

I have built a model for FY 26 using announced contracts and the revenues they are to be recognized (using delivery date as revenue recognition for IFRS)

Then used a fixed assumption of additional ATE and MAI (mai is legacy mounterse business) revenues that are below the >5% of FY 25 revenue threshold that would require them to file a supply contract announcement.

Using a conservative ballpark average gross margin of 40%
Keeping latest Q1 26 SG&A and R&D expenses fixed across the rest of 2026

Using these assumptions I arrive at 116B revenues and a 30% ebit margin = 35B ebit

If Mirae rerates closer to peer multiples say 25x ebit this would trade at a 875B won valuation
/4.5m shares ~ just below 200k won share price

Current share price hovering around ~45k won

A 4 bagger+ based off of already announced contracts and relatively conservative assumptions for gross margin and additional ATE and smt revenues

[haYN Capital @killapabkai]

@babyfolio 25x ebit is a multiple loosely based off relative valuations of related AI equipment cos.

I will add to the model an implied valuation at a range of different multiples eventually this was just back of napkin math to show discount on a relative valuation basis

[haYN Capital @killapabkai]

I am buying the dip because inflation + rate hike fears are overblown

Just pure rate hikes are not inherently negative for equity markets.

Credit spreads remain muted right now + nominal growth and AI earnings are very strong.

Rate hikes are just one component.

Look at the attached screenshot that shows markets in 2017-2019.

Notice how the market continued its uptrend in the face of a rate hiking cycle while credit spreads remained tight throughout the hiking cycle

[haYN Capital @killapabkai]

Sheltering in place right now.

Got stopped out of a couple positions.

Still bullish

Just waiting for further $SMH compression near 50D and for vol to compress a hit before adding exposure

[haYN Capital @killapabkai]

@jimcx0 @PositiveDrift_ @aleabitoreddit Yeah I took @PositiveDrift_ initial graph as inspiration to paste into claude code and had it build my own model to add more chrats and relevant data

[haYN Capital @killapabkai]

A look back at FY 19 is an example of the operating leverage they have if utilization approaches peak levels

Revenues jumped from 85B to 210B and they had mid 30s ebit margins

R&D and SG&A stayed flat while only product input costs rose and margins accrued as they cleared fixed cost base of their manufacturing facilities.

Module capacity has not rose since 2019

What has been added since 2019 is in-house laser chip lines so they can now manufacture transceivers using their own laser chips. Adding margin to their transceivers.

+

Transceivers will have higher ASP than the 5G modules they sold in 2019

So data center cycle = more revenues and higher margins than 2019 5G cycle and likely a longer cycle

$138080.KQ OE Solutions

This bull thesis relies on their data center product lines qualifying with end customers

I think the probability of them succeeding in entering volume production of data center optics has never been higher than right now

Due to
1. huge optics demand reducing # of options customers have of suppliers with capacity
2. Rising importance of sovereign data centers making them an important source of photonics for a domestically sourced Korean build out of data centers

[haYN Capital @killapabkai]

https://t.co/l5MwJ7ZmWL

[haYN Capital @killapabkai]

First Holy fuck paris is hot. Im dying

Second Where are all the french people?

90% of the people ive seen are african/middle eastern/tourists

[haYN Capital @killapabkai]

On a shorter time frame Q2 earnings should be a significant step up from Q1 due to delivery timing of previously announced contracts piling up in Q2.

Model says revenues should be around 35b
75% QoQ growth from Q1 20B
450% growth YoY from Q2 25

I suspect this may act as a catalyst once people can more visibly see Miraes growth on financial screeners/official revenue numbers instead of digging through dart filings

$025560.KS

[haYN Capital @killapabkai]

$025560.KS Mirae Corp FY 26E model

I have built a model for FY 26 using announced contracts and the revenues they are to be recognized (using delivery date as revenue recognition for IFRS)

Then used a fixed assumption of additional ATE and MAI (mai is legacy mounterse business) revenues that are below the >5% of FY 25 revenue threshold that would require them to file a supply contract announcement.

Using a conservative ballpark average gross margin of 40%
Keeping latest Q1 26 SG&A and R&D expenses fixed across the rest of 2026

Using these assumptions I arrive at 116B revenues and a 30% ebit margin = 35B ebit

If Mirae rerates closer to peer multiples say 25x ebit this would trade at a 875B won valuation
/4.5m shares ~ just below 200k won share price

Current share price hovering around ~45k won

A 4 bagger+ based off of already announced contracts and relatively conservative assumptions for gross margin and additional ATE and smt revenues