Thomas Chua
@steadycompound.bsky.social
Filling my stock portfolio with steady compounders and sharing my analysis at http://steadycompounding.com
ROIC across industries.
Interestingly, ROIC has been trending upwards in recent years, particularly for winners.
Interestingly, ROIC has been trending upwards in recent years, particularly for winners.
January 15, 2026 at 10:29 PM
ROIC across industries.
Interestingly, ROIC has been trending upwards in recent years, particularly for winners.
Interestingly, ROIC has been trending upwards in recent years, particularly for winners.
MELI has delivered 27 straight quarters of 30%+ revenue growth.
No company at this scale has matched that consistency.
But Shopee is doubling down in Brazil. Amazon is expanding. Competition is heating up.
Can they keep the streak alive?
Part 3: steadycompounding.com/investing/m...
No company at this scale has matched that consistency.
But Shopee is doubling down in Brazil. Amazon is expanding. Competition is heating up.
Can they keep the streak alive?
Part 3: steadycompounding.com/investing/m...
January 8, 2026 at 10:30 PM
MELI has delivered 27 straight quarters of 30%+ revenue growth.
No company at this scale has matched that consistency.
But Shopee is doubling down in Brazil. Amazon is expanding. Competition is heating up.
Can they keep the streak alive?
Part 3: steadycompounding.com/investing/m...
No company at this scale has matched that consistency.
But Shopee is doubling down in Brazil. Amazon is expanding. Competition is heating up.
Can they keep the streak alive?
Part 3: steadycompounding.com/investing/m...
But when forced to pick ONE company for the next 5 years, he chose Stripe over Adyen.
Why?
"We're entering a period of technological chaos over the next three to five years, and I believe they're better positioned than anyone to handle that."
Why?
"We're entering a period of technological chaos over the next three to five years, and I believe they're better positioned than anyone to handle that."
December 24, 2025 at 11:10 AM
But when forced to pick ONE company for the next 5 years, he chose Stripe over Adyen.
Why?
"We're entering a period of technological chaos over the next three to five years, and I believe they're better positioned than anyone to handle that."
Why?
"We're entering a period of technological chaos over the next three to five years, and I believe they're better positioned than anyone to handle that."
What about Checkout, Nuvei, Braintree?
"Those three companies will continue to survive and possibly thrive, but there's still a lot of share to be taken from legacy players."
Translation: They'll win some deals. They won't change the structural picture.
"Those three companies will continue to survive and possibly thrive, but there's still a lot of share to be taken from legacy players."
Translation: They'll win some deals. They won't change the structural picture.
December 24, 2025 at 11:10 AM
What about Checkout, Nuvei, Braintree?
"Those three companies will continue to survive and possibly thrive, but there's still a lot of share to be taken from legacy players."
Translation: They'll win some deals. They won't change the structural picture.
"Those three companies will continue to survive and possibly thrive, but there's still a lot of share to be taken from legacy players."
Translation: They'll win some deals. They won't change the structural picture.
They're not identical though.
Adyen: Traditional retail, physical POS, enterprise (<10,000 customers)
Stripe: Non-traditional enterprises, platforms, millions of merchants
"Stripe is winning in non-traditional enterprises, whereas Adyen has succeeded in traditional retail."
Adyen: Traditional retail, physical POS, enterprise (<10,000 customers)
Stripe: Non-traditional enterprises, platforms, millions of merchants
"Stripe is winning in non-traditional enterprises, whereas Adyen has succeeded in traditional retail."
December 24, 2025 at 11:10 AM
They're not identical though.
Adyen: Traditional retail, physical POS, enterprise (<10,000 customers)
Stripe: Non-traditional enterprises, platforms, millions of merchants
"Stripe is winning in non-traditional enterprises, whereas Adyen has succeeded in traditional retail."
Adyen: Traditional retail, physical POS, enterprise (<10,000 customers)
Stripe: Non-traditional enterprises, platforms, millions of merchants
"Stripe is winning in non-traditional enterprises, whereas Adyen has succeeded in traditional retail."
Scale becomes self-reinforcing:
• More volume → better data → better fraud detection
• Lower unit costs → pricing flexibility
• Single platform → faster feature deployment
"As market needs become more intensely technical, Adyen and Stripe are best positioned to differentiate."
• More volume → better data → better fraud detection
• Lower unit costs → pricing flexibility
• Single platform → faster feature deployment
"As market needs become more intensely technical, Adyen and Stripe are best positioned to differentiate."
December 24, 2025 at 11:09 AM
Scale becomes self-reinforcing:
• More volume → better data → better fraud detection
• Lower unit costs → pricing flexibility
• Single platform → faster feature deployment
"As market needs become more intensely technical, Adyen and Stripe are best positioned to differentiate."
• More volume → better data → better fraud detection
• Lower unit costs → pricing flexibility
• Single platform → faster feature deployment
"As market needs become more intensely technical, Adyen and Stripe are best positioned to differentiate."
The problem for incumbents isn't features—it's infrastructure.
"Their competitors are bogged down with numerous acquisitions and disjointed infrastructure."
Adyen and Stripe win with unified single-stack platforms. Legacy can't replicate that without starting over.
"Their competitors are bogged down with numerous acquisitions and disjointed infrastructure."
Adyen and Stripe win with unified single-stack platforms. Legacy can't replicate that without starting over.
December 24, 2025 at 11:09 AM
The problem for incumbents isn't features—it's infrastructure.
"Their competitors are bogged down with numerous acquisitions and disjointed infrastructure."
Adyen and Stripe win with unified single-stack platforms. Legacy can't replicate that without starting over.
"Their competitors are bogged down with numerous acquisitions and disjointed infrastructure."
Adyen and Stripe win with unified single-stack platforms. Legacy can't replicate that without starting over.
His view: Adyen and Stripe aren't really competing with each other.
They're both taking share from legacy players—Fiserv, Worldpay, Chase Merchant Services.
"There are a few others nibbling around the edges, but Stripe and Adyen have really taken the market by storm."
They're both taking share from legacy players—Fiserv, Worldpay, Chase Merchant Services.
"There are a few others nibbling around the edges, but Stripe and Adyen have really taken the market by storm."
December 24, 2025 at 11:09 AM
His view: Adyen and Stripe aren't really competing with each other.
They're both taking share from legacy players—Fiserv, Worldpay, Chase Merchant Services.
"There are a few others nibbling around the edges, but Stripe and Adyen have really taken the market by storm."
They're both taking share from legacy players—Fiserv, Worldpay, Chase Merchant Services.
"There are a few others nibbling around the edges, but Stripe and Adyen have really taken the market by storm."
Most people play life like Monopoly without ever buying a single asset.
steadycompounding.com/life/monopoly/
steadycompounding.com/life/monopoly/
December 23, 2025 at 10:32 PM
Most people play life like Monopoly without ever buying a single asset.
steadycompounding.com/life/monopoly/
steadycompounding.com/life/monopoly/
On Transformer architecture:
Transformers have a problem—context window is limited to maybe 1 million tokens today.
That's only ~500k words. Not enough to process all your emails and documents.
The Transformer architecture is not AGI. Something else will replace it.
Transformers have a problem—context window is limited to maybe 1 million tokens today.
That's only ~500k words. Not enough to process all your emails and documents.
The Transformer architecture is not AGI. Something else will replace it.
December 21, 2025 at 7:03 AM
On Transformer architecture:
Transformers have a problem—context window is limited to maybe 1 million tokens today.
That's only ~500k words. Not enough to process all your emails and documents.
The Transformer architecture is not AGI. Something else will replace it.
Transformers have a problem—context window is limited to maybe 1 million tokens today.
That's only ~500k words. Not enough to process all your emails and documents.
The Transformer architecture is not AGI. Something else will replace it.
On Anthropic's chip strategy:
It's smart for Anthropic not to rely solely on Google TPU or AWS.
By using multiple options, they can negotiate pricing.
Everyone wants to work with Anthropic—which might lead NVIDIA to offer a lower price.
It's smart for Anthropic not to rely solely on Google TPU or AWS.
By using multiple options, they can negotiate pricing.
Everyone wants to work with Anthropic—which might lead NVIDIA to offer a lower price.
December 21, 2025 at 7:03 AM
On Anthropic's chip strategy:
It's smart for Anthropic not to rely solely on Google TPU or AWS.
By using multiple options, they can negotiate pricing.
Everyone wants to work with Anthropic—which might lead NVIDIA to offer a lower price.
It's smart for Anthropic not to rely solely on Google TPU or AWS.
By using multiple options, they can negotiate pricing.
Everyone wants to work with Anthropic—which might lead NVIDIA to offer a lower price.
Who wins in edge?
Apple is the absolute monopoly for consumer devices.
In automotive, it's mainly NVIDIA and Tesla. Tesla makes their own chips.
Apple is the absolute monopoly for consumer devices.
In automotive, it's mainly NVIDIA and Tesla. Tesla makes their own chips.
December 21, 2025 at 7:03 AM
Who wins in edge?
Apple is the absolute monopoly for consumer devices.
In automotive, it's mainly NVIDIA and Tesla. Tesla makes their own chips.
Apple is the absolute monopoly for consumer devices.
In automotive, it's mainly NVIDIA and Tesla. Tesla makes their own chips.
On edge inference:
Eventually it's going to be all on the edge.
Edge isn't just phones—it's self-driving cars and humanoid robots.
10-year view: 80% on the edge.
5-year view: 50/50.
Eventually it's going to be all on the edge.
Edge isn't just phones—it's self-driving cars and humanoid robots.
10-year view: 80% on the edge.
5-year view: 50/50.
December 21, 2025 at 7:03 AM
On edge inference:
Eventually it's going to be all on the edge.
Edge isn't just phones—it's self-driving cars and humanoid robots.
10-year view: 80% on the edge.
5-year view: 50/50.
Eventually it's going to be all on the edge.
Edge isn't just phones—it's self-driving cars and humanoid robots.
10-year view: 80% on the edge.
5-year view: 50/50.
On how NVIDIA might respond:
NVIDIA is nimble—they can adapt quickly, unlike Google.
They might develop custom ASICs that are less general purpose but cheaper.
They do everything in-house, so they can potentially offer lower costs than Google.
NVIDIA is nimble—they can adapt quickly, unlike Google.
They might develop custom ASICs that are less general purpose but cheaper.
They do everything in-house, so they can potentially offer lower costs than Google.
December 21, 2025 at 7:03 AM
On how NVIDIA might respond:
NVIDIA is nimble—they can adapt quickly, unlike Google.
They might develop custom ASICs that are less general purpose but cheaper.
They do everything in-house, so they can potentially offer lower costs than Google.
NVIDIA is nimble—they can adapt quickly, unlike Google.
They might develop custom ASICs that are less general purpose but cheaper.
They do everything in-house, so they can potentially offer lower costs than Google.
But losing share doesn't mean losing revenue.
As NVIDIA loses share to Google, the entire market is growing—so revenues will continue to grow.
If NVIDIA loses 10% share but the market grows 20%, revenue still grows.
As NVIDIA loses share to Google, the entire market is growing—so revenues will continue to grow.
If NVIDIA loses 10% share but the market grows 20%, revenue still grows.
December 21, 2025 at 7:03 AM
But losing share doesn't mean losing revenue.
As NVIDIA loses share to Google, the entire market is growing—so revenues will continue to grow.
If NVIDIA loses 10% share but the market grows 20%, revenue still grows.
As NVIDIA loses share to Google, the entire market is growing—so revenues will continue to grow.
If NVIDIA loses 10% share but the market grows 20%, revenue still grows.
His five-year inference forecast:
NVIDIA's market share will be low because all hyperscalers will have their own chips.
~30% AWS, ~30% Google, ~30% Azure.
NVIDIA's market share will be low because all hyperscalers will have their own chips.
~30% AWS, ~30% Google, ~30% Azure.
December 21, 2025 at 7:03 AM
His five-year inference forecast:
NVIDIA's market share will be low because all hyperscalers will have their own chips.
~30% AWS, ~30% Google, ~30% Azure.
NVIDIA's market share will be low because all hyperscalers will have their own chips.
~30% AWS, ~30% Google, ~30% Azure.
The cost gap is massive.
NVIDIA charges ~10x manufacturing cost.
Google charges 2-4x.
NVIDIA can do this today because they don't have much competition.
NVIDIA charges ~10x manufacturing cost.
Google charges 2-4x.
NVIDIA can do this today because they don't have much competition.
December 21, 2025 at 7:03 AM
The cost gap is massive.
NVIDIA charges ~10x manufacturing cost.
Google charges 2-4x.
NVIDIA can do this today because they don't have much competition.
NVIDIA charges ~10x manufacturing cost.
Google charges 2-4x.
NVIDIA can do this today because they don't have much competition.
On inference:
The customer doesn't have a choice. You have no idea which hardware OpenAI actually runs your query on.
The hyperscaler decides. And for them, it's mainly about total cost of ownership.
The customer doesn't have a choice. You have no idea which hardware OpenAI actually runs your query on.
The hyperscaler decides. And for them, it's mainly about total cost of ownership.
December 21, 2025 at 7:03 AM
On inference:
The customer doesn't have a choice. You have no idea which hardware OpenAI actually runs your query on.
The hyperscaler decides. And for them, it's mainly about total cost of ownership.
The customer doesn't have a choice. You have no idea which hardware OpenAI actually runs your query on.
The hyperscaler decides. And for them, it's mainly about total cost of ownership.
But NVIDIA will still lose training share.
Google's TPU is already taking share, and that's only going to get worse.
AWS will enter with Trainium. AMD is further behind—their software stack is still really bad.
Google's TPU is already taking share, and that's only going to get worse.
AWS will enter with Trainium. AMD is further behind—their software stack is still really bad.
December 21, 2025 at 7:03 AM
But NVIDIA will still lose training share.
Google's TPU is already taking share, and that's only going to get worse.
AWS will enter with Trainium. AMD is further behind—their software stack is still really bad.
Google's TPU is already taking share, and that's only going to get worse.
AWS will enter with Trainium. AMD is further behind—their software stack is still really bad.
On training:
CUDA still matters. The software stack is superior, even compared to Google's TPU.
This is why xAI probably hasn't done a deal with Google's TPU—the software stack isn't as good as NVIDIA's.
CUDA still matters. The software stack is superior, even compared to Google's TPU.
This is why xAI probably hasn't done a deal with Google's TPU—the software stack isn't as good as NVIDIA's.
December 21, 2025 at 7:03 AM
On training:
CUDA still matters. The software stack is superior, even compared to Google's TPU.
This is why xAI probably hasn't done a deal with Google's TPU—the software stack isn't as good as NVIDIA's.
CUDA still matters. The software stack is superior, even compared to Google's TPU.
This is why xAI probably hasn't done a deal with Google's TPU—the software stack isn't as good as NVIDIA's.
Training and inference have fundamentally different competitive dynamics.
Training: Software stack matters. CUDA is still superior.
Inference: Customers only see an API. They don't choose the chip. Cost per token is all that matters.
This distinction is key to understanding NVIDIA's position.
Training: Software stack matters. CUDA is still superior.
Inference: Customers only see an API. They don't choose the chip. Cost per token is all that matters.
This distinction is key to understanding NVIDIA's position.
December 21, 2025 at 7:03 AM
Training and inference have fundamentally different competitive dynamics.
Training: Software stack matters. CUDA is still superior.
Inference: Customers only see an API. They don't choose the chip. Cost per token is all that matters.
This distinction is key to understanding NVIDIA's position.
Training: Software stack matters. CUDA is still superior.
Inference: Customers only see an API. They don't choose the chip. Cost per token is all that matters.
This distinction is key to understanding NVIDIA's position.
Read an expert call with a former Google chip engineer on NVIDIA's competitive position.
His view: NVIDIA will lose market share in both training and inference.
Here's why 🧵:
His view: NVIDIA will lose market share in both training and inference.
Here's why 🧵:
December 21, 2025 at 7:03 AM
Read an expert call with a former Google chip engineer on NVIDIA's competitive position.
His view: NVIDIA will lose market share in both training and inference.
Here's why 🧵:
His view: NVIDIA will lose market share in both training and inference.
Here's why 🧵:
Seth Klarman on the five traits that made Buffett a successful investor:
December 19, 2025 at 1:19 PM
Seth Klarman on the five traits that made Buffett a successful investor:
On 9 April 2020, I published my first post on Steady Compounding.
No audience. No plan. Just a tagline: "All Big Things Come From Small Beginnings."
Almost 6 years and a little less hair later, that line turned out to be the whole point.
steadycompounding.com/life/almost...
No audience. No plan. Just a tagline: "All Big Things Come From Small Beginnings."
Almost 6 years and a little less hair later, that line turned out to be the whole point.
steadycompounding.com/life/almost...
December 15, 2025 at 11:13 PM
On 9 April 2020, I published my first post on Steady Compounding.
No audience. No plan. Just a tagline: "All Big Things Come From Small Beginnings."
Almost 6 years and a little less hair later, that line turned out to be the whole point.
steadycompounding.com/life/almost...
No audience. No plan. Just a tagline: "All Big Things Come From Small Beginnings."
Almost 6 years and a little less hair later, that line turned out to be the whole point.
steadycompounding.com/life/almost...
1/ Howard Marks just dropped 5 decades of investing wisdom in one conversation.
5 market calls in 50 years. $218 billion under management. Zero blowups.
Here's what separates him from everyone else: 🧵
5 market calls in 50 years. $218 billion under management. Zero blowups.
Here's what separates him from everyone else: 🧵
December 14, 2025 at 7:24 AM
1/ Howard Marks just dropped 5 decades of investing wisdom in one conversation.
5 market calls in 50 years. $218 billion under management. Zero blowups.
Here's what separates him from everyone else: 🧵
5 market calls in 50 years. $218 billion under management. Zero blowups.
Here's what separates him from everyone else: 🧵