Dan Zhuge
@zhug3.xyz
quantitative ramblings at: https://zhug3.xyz
The hyperscaler capex arms race in one chart.
2024: $256B combined
2026: $602B combined
Amazon, Microsoft, Google, and Meta each spending $100B+ individually.
Expected AI revenue from all this? ~$25B. That's a 24:1 investment-to-revenue ratio.
2024: $256B combined
2026: $602B combined
Amazon, Microsoft, Google, and Meta each spending $100B+ individually.
Expected AI revenue from all this? ~$25B. That's a 24:1 investment-to-revenue ratio.
February 4, 2026 at 10:40 AM
The hyperscaler capex arms race in one chart.
2024: $256B combined
2026: $602B combined
Amazon, Microsoft, Google, and Meta each spending $100B+ individually.
Expected AI revenue from all this? ~$25B. That's a 24:1 investment-to-revenue ratio.
2024: $256B combined
2026: $602B combined
Amazon, Microsoft, Google, and Meta each spending $100B+ individually.
Expected AI revenue from all this? ~$25B. That's a 24:1 investment-to-revenue ratio.
Major trade deals signed in 2025-2026:
EU-India: 2 billion people, 25% of global GDP
EU-Mercosur: 700 million people
India-UK, India-Australia, India-UAE...
What do they all have in common?
The US isn't in any of them.
EU-India: 2 billion people, 25% of global GDP
EU-Mercosur: 700 million people
India-UK, India-Australia, India-UAE...
What do they all have in common?
The US isn't in any of them.
February 4, 2026 at 9:36 AM
Major trade deals signed in 2025-2026:
EU-India: 2 billion people, 25% of global GDP
EU-Mercosur: 700 million people
India-UK, India-Australia, India-UAE...
What do they all have in common?
The US isn't in any of them.
EU-India: 2 billion people, 25% of global GDP
EU-Mercosur: 700 million people
India-UK, India-Australia, India-UAE...
What do they all have in common?
The US isn't in any of them.
Palantir just reported 70% revenue growth. US commercial up 137%.
Rule of 40 score: 127%.
Trades at 141x earnings.
Rule of 40 score: 127%.
Trades at 141x earnings.
February 4, 2026 at 8:15 AM
Palantir just reported 70% revenue growth. US commercial up 137%.
Rule of 40 score: 127%.
Trades at 141x earnings.
Rule of 40 score: 127%.
Trades at 141x earnings.
People aren't broke because of delivery fees. They're cooking because they're broke.
BLS data shows Americans are eating out less and cooking more since 2019.
The shift is largest among:
- Lowest income quintile (+2.5pp)
- Under-25s (+1.7pp)
BLS data shows Americans are eating out less and cooking more since 2019.
The shift is largest among:
- Lowest income quintile (+2.5pp)
- Under-25s (+1.7pp)
February 3, 2026 at 2:48 PM
People aren't broke because of delivery fees. They're cooking because they're broke.
BLS data shows Americans are eating out less and cooking more since 2019.
The shift is largest among:
- Lowest income quintile (+2.5pp)
- Under-25s (+1.7pp)
BLS data shows Americans are eating out less and cooking more since 2019.
The shift is largest among:
- Lowest income quintile (+2.5pp)
- Under-25s (+1.7pp)
The Musk empire is now two distinct pillars:
Pillar 1: SpaceX-xAI-X
- Space, AI, social media
- $1.25T combined
Pillar 2: Tesla
- Vehicles, energy, robotics
- ~$1T market cap
Tesla invested $2B in xAI last month. Now owns a small indirect stake in SpaceX.
Pillar 1: SpaceX-xAI-X
- Space, AI, social media
- $1.25T combined
Pillar 2: Tesla
- Vehicles, energy, robotics
- ~$1T market cap
Tesla invested $2B in xAI last month. Now owns a small indirect stake in SpaceX.
February 3, 2026 at 10:34 AM
The Musk empire is now two distinct pillars:
Pillar 1: SpaceX-xAI-X
- Space, AI, social media
- $1.25T combined
Pillar 2: Tesla
- Vehicles, energy, robotics
- ~$1T market cap
Tesla invested $2B in xAI last month. Now owns a small indirect stake in SpaceX.
Pillar 1: SpaceX-xAI-X
- Space, AI, social media
- $1.25T combined
Pillar 2: Tesla
- Vehicles, energy, robotics
- ~$1T market cap
Tesla invested $2B in xAI last month. Now owns a small indirect stake in SpaceX.
SpaceX-xAI at $1.25T valuation.
For context:
- Microsoft hit $1T at $92B revenue
- Google hit $1T at $183B revenue
- SpaceX has ~$16B revenue
That's a 78x revenue multiple vs ~10x for mature tech.
For context:
- Microsoft hit $1T at $92B revenue
- Google hit $1T at $183B revenue
- SpaceX has ~$16B revenue
That's a 78x revenue multiple vs ~10x for mature tech.
February 3, 2026 at 8:29 AM
SpaceX-xAI at $1.25T valuation.
For context:
- Microsoft hit $1T at $92B revenue
- Google hit $1T at $183B revenue
- SpaceX has ~$16B revenue
That's a 78x revenue multiple vs ~10x for mature tech.
For context:
- Microsoft hit $1T at $92B revenue
- Google hit $1T at $183B revenue
- SpaceX has ~$16B revenue
That's a 78x revenue multiple vs ~10x for mature tech.
how do you price agentic workloads that run inference continuously? The AI cost structure is inverting.
2023: Training was 67% of AI compute spend
2026: Inference is now 55%+
2023: Training was 67% of AI compute spend
2026: Inference is now 55%+
February 3, 2026 at 8:02 AM
how do you price agentic workloads that run inference continuously? The AI cost structure is inverting.
2023: Training was 67% of AI compute spend
2026: Inference is now 55%+
2023: Training was 67% of AI compute spend
2026: Inference is now 55%+
Current US effective tariff rates:
China: 104%
Everyone else: 10-25%
For context, pre-2018 average was 2.2%.
The Supreme Court is currently deciding whether these tariffs are even legal.
China: 104%
Everyone else: 10-25%
For context, pre-2018 average was 2.2%.
The Supreme Court is currently deciding whether these tariffs are even legal.
February 2, 2026 at 8:12 PM
Current US effective tariff rates:
China: 104%
Everyone else: 10-25%
For context, pre-2018 average was 2.2%.
The Supreme Court is currently deciding whether these tariffs are even legal.
China: 104%
Everyone else: 10-25%
For context, pre-2018 average was 2.2%.
The Supreme Court is currently deciding whether these tariffs are even legal.
FT piece on China's "genius classes" - the educational pipeline producing DeepSeek engineers, ByteDance founders, and Olympiad champions. Author Zijing Wu is herself a graduate is making the rounds today.
February 2, 2026 at 3:16 PM
FT piece on China's "genius classes" - the educational pipeline producing DeepSeek engineers, ByteDance founders, and Olympiad champions. Author Zijing Wu is herself a graduate is making the rounds today.
NFT market cap hit $3.3B this week. It's still 85% below the 2021 peak. The 30% weekly "jump" is just noise inside a range it's been stuck in for ~two years
February 2, 2026 at 2:15 PM
NFT market cap hit $3.3B this week. It's still 85% below the 2021 peak. The 30% weekly "jump" is just noise inside a range it's been stuck in for ~two years
"i function fine on 6 hours" is the cognitive equivalent of "i drive better drunk"
Study found that after 4 days of 6h sleep you stop feeling more tired but your brain keeps getting dumber for another 10 days
you're basically too tired to notice you're tired
Study found that after 4 days of 6h sleep you stop feeling more tired but your brain keeps getting dumber for another 10 days
you're basically too tired to notice you're tired
February 2, 2026 at 2:00 PM
"i function fine on 6 hours" is the cognitive equivalent of "i drive better drunk"
Study found that after 4 days of 6h sleep you stop feeling more tired but your brain keeps getting dumber for another 10 days
you're basically too tired to notice you're tired
Study found that after 4 days of 6h sleep you stop feeling more tired but your brain keeps getting dumber for another 10 days
you're basically too tired to notice you're tired
Table 1 from the Attention Is All You Need paper (arxiv.org/pdf/1706.03762). Brilliant. Beautiful!
Maximum Path Length - how far a signal travels between any two positions:
- RNNs: O(n)
- Self-Attention: O(1)
Maximum Path Length - how far a signal travels between any two positions:
- RNNs: O(n)
- Self-Attention: O(1)
February 2, 2026 at 9:31 AM
Table 1 from the Attention Is All You Need paper (arxiv.org/pdf/1706.03762). Brilliant. Beautiful!
Maximum Path Length - how far a signal travels between any two positions:
- RNNs: O(n)
- Self-Attention: O(1)
Maximum Path Length - how far a signal travels between any two positions:
- RNNs: O(n)
- Self-Attention: O(1)
Software loans in CLOs: -4% in January vs flat for everything else.
Software is 12% of the leveraged loan index. Largest single sector.
Trigger is AI disruption fears. The selloff is hitting embedded enterprise systems and point solutions equally.
Software is 12% of the leveraged loan index. Largest single sector.
Trigger is AI disruption fears. The selloff is hitting embedded enterprise systems and point solutions equally.
February 1, 2026 at 6:13 PM
Software loans in CLOs: -4% in January vs flat for everything else.
Software is 12% of the leveraged loan index. Largest single sector.
Trigger is AI disruption fears. The selloff is hitting embedded enterprise systems and point solutions equally.
Software is 12% of the leveraged loan index. Largest single sector.
Trigger is AI disruption fears. The selloff is hitting embedded enterprise systems and point solutions equally.
Japan showed you can zero out your debt service costs through financial repression
Japan also showed you pay for it in currency value
Japan also showed you pay for it in currency value
February 1, 2026 at 3:33 PM
Japan showed you can zero out your debt service costs through financial repression
Japan also showed you pay for it in currency value
Japan also showed you pay for it in currency value
This kind of outrages honesty!!
Noam Shazeer in GLU Variants Improve Transformer (arxiv.org/abs/2002.052...)
Noam Shazeer in GLU Variants Improve Transformer (arxiv.org/abs/2002.052...)
February 1, 2026 at 2:18 PM
This kind of outrages honesty!!
Noam Shazeer in GLU Variants Improve Transformer (arxiv.org/abs/2002.052...)
Noam Shazeer in GLU Variants Improve Transformer (arxiv.org/abs/2002.052...)
Reposted by Dan Zhuge
It’s the beginning of history! 😃
Nomura: total returns for software debt in collateralized loan obligations have been lower than all other sectors in 2026, amid fears about AI's impact on SaaS (Bloomberg)
Main Link | Techmeme Permalink
Main Link | Techmeme Permalink
February 1, 2026 at 5:18 AM
It’s the beginning of history! 😃
These 2020 ARK Invest slides live rent free in my head
January 31, 2026 at 7:01 PM
These 2020 ARK Invest slides live rent free in my head
Reposted by Dan Zhuge
What could a serious company do with 10 billion?
Amazon, a potential new investor in OpenAI, is in talks to invest more than $10 billion and possibly over $20 billion. https://thein.fo/3LVnyHu
Nvidia, Microsoft, Amazon in Talks to Invest Up to $60 Billion in OpenAI
OpenAI is lining up huge checks from some of its biggest tech partners, as it seeks to raise up to $100 billion to fund growing demand for its AI. Nvidia, an existing investor whose chips power OpenAI...
thein.fo
January 31, 2026 at 6:08 PM
What could a serious company do with 10 billion?
Unreal how much time I lately spent creating dishes with Nano Banana Pro
January 31, 2026 at 4:33 PM
Unreal how much time I lately spent creating dishes with Nano Banana Pro
The crash has been six months away for 20 years. Cassandra trades are indistinguishable from being wrong. Markets can remain irrational longer than you can remain solvent.
wilsoniumite.com/2026/01/27/s...
wilsoniumite.com/2026/01/27/s...
Surely the crash of the US economy has to be soon – Wilsons Blog
wilsoniumite.com
January 31, 2026 at 3:56 PM
The crash has been six months away for 20 years. Cassandra trades are indistinguishable from being wrong. Markets can remain irrational longer than you can remain solvent.
wilsoniumite.com/2026/01/27/s...
wilsoniumite.com/2026/01/27/s...
Silver: -31.35% (z = -15.4σ)
Gold: -11.37% (z = -10.3σ)
A 15-sigma event shouldn't happen once in the lifetime of the universe under normal distributions. Of course, returns aren't normal. But still..
Gold: -11.37% (z = -10.3σ)
A 15-sigma event shouldn't happen once in the lifetime of the universe under normal distributions. Of course, returns aren't normal. But still..
January 31, 2026 at 10:43 AM
Silver: -31.35% (z = -15.4σ)
Gold: -11.37% (z = -10.3σ)
A 15-sigma event shouldn't happen once in the lifetime of the universe under normal distributions. Of course, returns aren't normal. But still..
Gold: -11.37% (z = -10.3σ)
A 15-sigma event shouldn't happen once in the lifetime of the universe under normal distributions. Of course, returns aren't normal. But still..