Corry Wang
@corrywang.bsky.social
Compute @ Anthropic | Formerly AI strategy @ Google and tech equity research @ Bernstein Research
1/4 If I had a nickel for every time Mark Zuckerberg blew a few billion dollars trying to hire a team of star researchers to build a second place frontier AI research lab, I’d have two nickels. Which isn’t a lot, but it’s weird that it happened twice
November 11, 2025 at 4:43 PM
1/4 If I had a nickel for every time Mark Zuckerberg blew a few billion dollars trying to hire a team of star researchers to build a second place frontier AI research lab, I’d have two nickels. Which isn’t a lot, but it’s weird that it happened twice
1/ Last week I finally got around to reading Genius Makers - this was Cade Metz's 2021 book on the history of ML
It was really fascinating, but in the same way you'd be fascinated reading a history of Newtonian physics published 3 months before Einstein invented relativity
It was really fascinating, but in the same way you'd be fascinated reading a history of Newtonian physics published 3 months before Einstein invented relativity
November 10, 2025 at 5:07 AM
1/ Last week I finally got around to reading Genius Makers - this was Cade Metz's 2021 book on the history of ML
It was really fascinating, but in the same way you'd be fascinated reading a history of Newtonian physics published 3 months before Einstein invented relativity
It was really fascinating, but in the same way you'd be fascinated reading a history of Newtonian physics published 3 months before Einstein invented relativity
Something that I didn’t really appreciate until I left Big Tech is that for a startup to succeed, you really have to be doing stuff that sounds crazy all the time
If it didn’t sound crazy to an L8 at Google, then Google would’ve done it already!
If it didn’t sound crazy to an L8 at Google, then Google would’ve done it already!
November 9, 2025 at 10:05 PM
Something that I didn’t really appreciate until I left Big Tech is that for a startup to succeed, you really have to be doing stuff that sounds crazy all the time
If it didn’t sound crazy to an L8 at Google, then Google would’ve done it already!
If it didn’t sound crazy to an L8 at Google, then Google would’ve done it already!
From 1840 to 1850, private Britons cumulatively invested 40% of British GDP into the country’s first rail network. For reference, the equivalent today would be the tech industry spending like, $10 trillion dollars on a single thing
Anyways it’s confirmed, guess we’re all doing this again guys
Anyways it’s confirmed, guess we’re all doing this again guys
October 8, 2025 at 10:07 PM
From 1840 to 1850, private Britons cumulatively invested 40% of British GDP into the country’s first rail network. For reference, the equivalent today would be the tech industry spending like, $10 trillion dollars on a single thing
Anyways it’s confirmed, guess we’re all doing this again guys
Anyways it’s confirmed, guess we’re all doing this again guys
I don't think Americans realize that outside the US, you can now just buy Ozempic online for $150/month. This will ultimately fall to <$50/month
This actually might've ended up as the important thing in global society in the 2020s, it weren't for the whole, yknow, AI thing
This actually might've ended up as the important thing in global society in the 2020s, it weren't for the whole, yknow, AI thing
August 3, 2025 at 2:49 PM
I don't think Americans realize that outside the US, you can now just buy Ozempic online for $150/month. This will ultimately fall to <$50/month
This actually might've ended up as the important thing in global society in the 2020s, it weren't for the whole, yknow, AI thing
This actually might've ended up as the important thing in global society in the 2020s, it weren't for the whole, yknow, AI thing
There's a famous anecdote about the invention of the cellphone: in 1981 McKinsey estimated it'd have a TAM of <1M people, so AT&T exited the market
Turns out this anecdote is made up. AT&T's marketing team did claim this, but the engineers just ignored them and launched anyways
Turns out this anecdote is made up. AT&T's marketing team did claim this, but the engineers just ignored them and launched anyways
July 13, 2025 at 3:22 AM
There's a famous anecdote about the invention of the cellphone: in 1981 McKinsey estimated it'd have a TAM of <1M people, so AT&T exited the market
Turns out this anecdote is made up. AT&T's marketing team did claim this, but the engineers just ignored them and launched anyways
Turns out this anecdote is made up. AT&T's marketing team did claim this, but the engineers just ignored them and launched anyways
I sometimes wonder these days what % of equity research is just written by ChatGPT. But then I see UBS publish a paragraph like this and realize I'm still getting 100% authentic human content
July 6, 2025 at 11:50 PM
I sometimes wonder these days what % of equity research is just written by ChatGPT. But then I see UBS publish a paragraph like this and realize I'm still getting 100% authentic human content
It's quite striking that despite everything that's happened in AI over the last 3 years, the world is still spending *less* capex building semiconductor foundries today than in 2022
All of AI is still small enough to be washed away by consumers buying -10% fewer Android phones
All of AI is still small enough to be washed away by consumers buying -10% fewer Android phones
June 21, 2025 at 3:20 PM
It's quite striking that despite everything that's happened in AI over the last 3 years, the world is still spending *less* capex building semiconductor foundries today than in 2022
All of AI is still small enough to be washed away by consumers buying -10% fewer Android phones
All of AI is still small enough to be washed away by consumers buying -10% fewer Android phones
1/ I don’t think people have realized how much this new Waymo scaling laws paper is basically an admission that “Waymo was wrong, Tesla was right”
Hopefully this becomes a call to action internally within Waymo
Hopefully this becomes a call to action internally within Waymo
According to new research by waymo, self driving cars neural nets perform better according to power scaling laws. More data and compute = better performance. waymo.com/blog/2025/06...
June 19, 2025 at 12:42 PM
1/ I don’t think people have realized how much this new Waymo scaling laws paper is basically an admission that “Waymo was wrong, Tesla was right”
Hopefully this becomes a call to action internally within Waymo
Hopefully this becomes a call to action internally within Waymo
When Goldman Sachs initiated on NVIDIA in 2016, there was general amazement that any semiconductor company could actually grow revenues sustainably
June 3, 2025 at 2:03 PM
When Goldman Sachs initiated on NVIDIA in 2016, there was general amazement that any semiconductor company could actually grow revenues sustainably
Life update: I joined Anthropic at the start of the month!
(1/5)
(1/5)
May 14, 2025 at 3:40 AM
Life update: I joined Anthropic at the start of the month!
(1/5)
(1/5)
More than one friend has asked me in the last few months… so here’s my rule of thumb:
1 ChatGPT query costs, like, 5 H100 seconds
An H100 consumes roughly the same electricity as the average American house
So 1 ChatGPT query = turning on the lights in your house for 5 seconds
1 ChatGPT query costs, like, 5 H100 seconds
An H100 consumes roughly the same electricity as the average American house
So 1 ChatGPT query = turning on the lights in your house for 5 seconds
I wrote a cheat sheet version of my Using ChatGPT is not bad for the environment post. It's paired down and more focused on simple responses to common objections, without a long intro or my background environmental philosophy. andymasley.substack.com/p/a-cheat-sh...
A cheat sheet for conversations about ChatGPT and the environment
Arm yourself with knowledge
andymasley.substack.com
April 29, 2025 at 4:02 AM
More than one friend has asked me in the last few months… so here’s my rule of thumb:
1 ChatGPT query costs, like, 5 H100 seconds
An H100 consumes roughly the same electricity as the average American house
So 1 ChatGPT query = turning on the lights in your house for 5 seconds
1 ChatGPT query costs, like, 5 H100 seconds
An H100 consumes roughly the same electricity as the average American house
So 1 ChatGPT query = turning on the lights in your house for 5 seconds
It’s kind of wild how much China has run away with the electric vehicle battery market in the last 5 years
The largest Chinese battery maker CATL now controls 37% of global industry sales, and 90%+ of industry profits (ex-BYD)
The largest Chinese battery maker CATL now controls 37% of global industry sales, and 90%+ of industry profits (ex-BYD)
April 24, 2025 at 8:12 PM
It’s kind of wild how much China has run away with the electric vehicle battery market in the last 5 years
The largest Chinese battery maker CATL now controls 37% of global industry sales, and 90%+ of industry profits (ex-BYD)
The largest Chinese battery maker CATL now controls 37% of global industry sales, and 90%+ of industry profits (ex-BYD)
1/ Was this the biggest miss in the history of pharma? Apparently in 1990, Pfizer preemptively abandoned development of the first GLP1 drugs
Ozempic, Zepbound, Wegovy, Mounjaro, etc. were doing $60B+ in runrate revenues at the end of last year. None are made by Pfizer
Ozempic, Zepbound, Wegovy, Mounjaro, etc. were doing $60B+ in runrate revenues at the end of last year. None are made by Pfizer
April 23, 2025 at 3:49 PM
1/ Was this the biggest miss in the history of pharma? Apparently in 1990, Pfizer preemptively abandoned development of the first GLP1 drugs
Ozempic, Zepbound, Wegovy, Mounjaro, etc. were doing $60B+ in runrate revenues at the end of last year. None are made by Pfizer
Ozempic, Zepbound, Wegovy, Mounjaro, etc. were doing $60B+ in runrate revenues at the end of last year. None are made by Pfizer
I did not know this - the failed first attempt to build the Panama Canal in the 1880s cost 10% of France’s GDP
For reference, that would be equivalent to a single startup raising about $300B today
For reference, that would be equivalent to a single startup raising about $300B today
April 22, 2025 at 7:54 PM
I did not know this - the failed first attempt to build the Panama Canal in the 1880s cost 10% of France’s GDP
For reference, that would be equivalent to a single startup raising about $300B today
For reference, that would be equivalent to a single startup raising about $300B today
1/8 Daniel Kokotajlo (ex-OpenAI) just published ai-2027.com, his forecast for AI takeoff by 2027
I am deeply skeptical… but I also deeply believe the best predictor of future forecasting performance is past forecasting performance
Conveniently, Daniel wrote a prior set of predictions in 2021!
I am deeply skeptical… but I also deeply believe the best predictor of future forecasting performance is past forecasting performance
Conveniently, Daniel wrote a prior set of predictions in 2021!
April 4, 2025 at 1:16 PM
1/8 Daniel Kokotajlo (ex-OpenAI) just published ai-2027.com, his forecast for AI takeoff by 2027
I am deeply skeptical… but I also deeply believe the best predictor of future forecasting performance is past forecasting performance
Conveniently, Daniel wrote a prior set of predictions in 2021!
I am deeply skeptical… but I also deeply believe the best predictor of future forecasting performance is past forecasting performance
Conveniently, Daniel wrote a prior set of predictions in 2021!
In retrospect, the cost of inference has been a crazy red herring for the whole AI industry over the last two years
If you think a particular LLM costs too much to ship in your product today, maybe ship it anyways… the unit economics might just turn out to work 6 months later
If you think a particular LLM costs too much to ship in your product today, maybe ship it anyways… the unit economics might just turn out to work 6 months later
March 14, 2025 at 2:53 AM
In retrospect, the cost of inference has been a crazy red herring for the whole AI industry over the last two years
If you think a particular LLM costs too much to ship in your product today, maybe ship it anyways… the unit economics might just turn out to work 6 months later
If you think a particular LLM costs too much to ship in your product today, maybe ship it anyways… the unit economics might just turn out to work 6 months later
Apparently, one of the most effective ways to improve LLM performance is just to manually insert the word “wait” at the end of every answer, to force it to think harder
You have to admit, sometimes working on LLMs feels less like engineering and more like a wizard casting spells
You have to admit, sometimes working on LLMs feels less like engineering and more like a wizard casting spells
February 17, 2025 at 4:47 PM
Apparently, one of the most effective ways to improve LLM performance is just to manually insert the word “wait” at the end of every answer, to force it to think harder
You have to admit, sometimes working on LLMs feels less like engineering and more like a wizard casting spells
You have to admit, sometimes working on LLMs feels less like engineering and more like a wizard casting spells
1/ If you really believe LLMs will dramatically compress the cost of software development in 3-5 years, doesn't this obviate the reason for independent software vendors to exist?
This doesn't seem obviously crazy to me - it'd just be a return to the days of mainframes
This doesn't seem obviously crazy to me - it'd just be a return to the days of mainframes
February 16, 2025 at 11:29 PM
1/ If you really believe LLMs will dramatically compress the cost of software development in 3-5 years, doesn't this obviate the reason for independent software vendors to exist?
This doesn't seem obviously crazy to me - it'd just be a return to the days of mainframes
This doesn't seem obviously crazy to me - it'd just be a return to the days of mainframes
Something that surprised me on my trip to China last year was how everybody ordered food via WeChat even when dining in at a restaurant
This was true both at fast foods (where the cash register was usually closed) and dine-in (where waiters now only bring you food instead of taking orders)
This was true both at fast foods (where the cash register was usually closed) and dine-in (where waiters now only bring you food instead of taking orders)
February 16, 2025 at 12:48 AM
Something that surprised me on my trip to China last year was how everybody ordered food via WeChat even when dining in at a restaurant
This was true both at fast foods (where the cash register was usually closed) and dine-in (where waiters now only bring you food instead of taking orders)
This was true both at fast foods (where the cash register was usually closed) and dine-in (where waiters now only bring you food instead of taking orders)
On the one hand: I’m constantly amazed at how much capital the top AI labs are trying to raise today
On the other hand: in the electric vehicle market, Rivian alone is probably going to end up blowing $40B just to reach, like, 2% share of global autos in 2030
On the other hand: in the electric vehicle market, Rivian alone is probably going to end up blowing $40B just to reach, like, 2% share of global autos in 2030
February 15, 2025 at 5:01 PM
On the one hand: I’m constantly amazed at how much capital the top AI labs are trying to raise today
On the other hand: in the electric vehicle market, Rivian alone is probably going to end up blowing $40B just to reach, like, 2% share of global autos in 2030
On the other hand: in the electric vehicle market, Rivian alone is probably going to end up blowing $40B just to reach, like, 2% share of global autos in 2030
The internal version of this document was the best primer I’d ever read at Google on the fundamental math of LLM training and inference cost
Honestly, I’m kind of surprised the team got clearance to release this to the public. You’re welcome
Honestly, I’m kind of surprised the team got clearance to release this to the public. You’re welcome
Making LLMs run efficiently can feel scary, but scaling isn’t magic, it’s math! We wanted to demystify the “systems view” of LLMs and wrote a little textbook called “How To Scale Your Model” which we’re releasing today. 1/n
February 5, 2025 at 2:51 AM
The internal version of this document was the best primer I’d ever read at Google on the fundamental math of LLM training and inference cost
Honestly, I’m kind of surprised the team got clearance to release this to the public. You’re welcome
Honestly, I’m kind of surprised the team got clearance to release this to the public. You’re welcome
1/ This internal 2007 Nokia presentation on the first iPhone is a really good example of how incumbents actually get disrupted
Oftentimes, the incumbent already knows what needs to be done. It's just that organizational incentives inhibit the incumbent from doing it
Oftentimes, the incumbent already knows what needs to be done. It's just that organizational incentives inhibit the incumbent from doing it
January 19, 2025 at 7:02 PM
1/ This internal 2007 Nokia presentation on the first iPhone is a really good example of how incumbents actually get disrupted
Oftentimes, the incumbent already knows what needs to be done. It's just that organizational incentives inhibit the incumbent from doing it
Oftentimes, the incumbent already knows what needs to be done. It's just that organizational incentives inhibit the incumbent from doing it
1/ Tfw when you think AI is a once-in-generation venture opportunity, but your best investment idea is “buy a minority stake in an accounting firm and give every employee a GPT wrapper”
January 15, 2025 at 4:42 AM
1/ Tfw when you think AI is a once-in-generation venture opportunity, but your best investment idea is “buy a minority stake in an accounting firm and give every employee a GPT wrapper”
Not something I see a lot of people talk about - but nuclear fusion might be the best historical example of a technology where a scaling law ultimately failed
Fusion reactors were improving by 10x every 5 years until the mid-80s… and then basically flatlined into the 2000s
Fusion reactors were improving by 10x every 5 years until the mid-80s… and then basically flatlined into the 2000s
January 12, 2025 at 8:46 PM
Not something I see a lot of people talk about - but nuclear fusion might be the best historical example of a technology where a scaling law ultimately failed
Fusion reactors were improving by 10x every 5 years until the mid-80s… and then basically flatlined into the 2000s
Fusion reactors were improving by 10x every 5 years until the mid-80s… and then basically flatlined into the 2000s