The Transcript
$NET: +16% AH
tweet
$NET: +16% AH
Earnings a bit different this time. Due to team commitments at the Munich Security Conference and the Olympics we’re doing our report next Tuesday. So I’m sitting in a Milan hotel with a nice Barolo listening to Bocelli and writing the script. Make of that what you will. tweet
X (formerly Twitter)
Matthew Prince 🌥 (@eastdakota) on X
Earnings a bit different this time. Due to team commitments at the Munich Security Conference and the Olympics we’re doing our report next Tuesday. So I’m sitting in a Milan hotel with a nice Barolo listening to Bocelli and writing the script. Make of that…
Offshore
Photo
Michael Fritzell (Asian Century Stocks)
FnGuide up another +11% today. Feel’s like there’s almost no end to the demand for the stock. https://t.co/mqJYQH1Dd1
tweet
FnGuide up another +11% today. Feel’s like there’s almost no end to the demand for the stock. https://t.co/mqJYQH1Dd1
tweet
Michael Fritzell (Asian Century Stocks)
RT @finphysnerd: Turns out the insider buying was a good sign. A decent $40m half yearly profit for a $1.3b market cap. More important though is a $141m decrease in development properties as they realise them for cash.
tweet
RT @finphysnerd: Turns out the insider buying was a good sign. A decent $40m half yearly profit for a $1.3b market cap. More important though is a $141m decrease in development properties as they realise them for cash.
Wing Tai Holdings $W05.si is a Singaporean stock to keep an eye on. It's a property developer at a $1b market cap with $1.8b of cash and liquid securities, $1.3b of total liabilities, $830m of investment properties and finally $1.b of development properties, most of which are tweet
X (formerly Twitter)
Myles (@finphysnerd) on X
Wing Tai Holdings $W05.si is a Singaporean stock to keep an eye on. It's a property developer at a $1b market cap with $1.8b of cash and liquid securities, $1.3b of total liabilities, $830m of investment properties and finally $1.b of development properties…
Jukan
Interesting parts from Morgan Stanley's Micron commentary:
"Memory producers currently have virtually no inventory. Customers lower on the supply priority list—such as Chinese Android OEMs and consumer PC makers—are struggling to secure any inventory at all, and even the highest-priority customers face a similar situation. These top-tier customers are absorbing every incremental unit of supply and paying premiums over contracted prices just to receive shipments even 30 days sooner.
This tight supply-demand environment is reflected in Samsung and SK Hynix posting only low-single-digit (LSD) quarter-over-quarter bit growth heading into Q1—a clear sign that near-term fab output growth alone cannot keep pace with demand.
Even over the medium to long term, the pace of supply expansion remains constrained. Factoring in new fabs such as CXMT Shanghai, Hynix M15, and Samsung P4L, wafer starts are expected to grow only about 7% year-over-year by the end of CY26. Micron's Boise project and its partnership with PSMC are also unlikely to contribute meaningful capacity before 2027.
Demand, on the other hand, is overwhelming. Summing up the expected quarterly revenue increases from key players through the end of 2026:
- NVIDIA: +$30 billion in quarterly revenue
- AMD Data Center: doubling to ~$10 billion per quarter
- Broadcom Semiconductors: doubling to ~$25 billion per quarter
- Marvell + Intel: ~$1 billion in incremental revenue
Annualized, the memory industry faces roughly $200 billion in incremental revenue it must support over the next 12 months—a figure that exceeds the entire logic semiconductor market in 2020. The surge in HBM demand, in particular, means DRAM suppliers will need to operate at significantly higher capital intensity.
In short, as long as AI growth maintains its current CAGR, closing the supply-demand gap will be no easy task. In this structurally robust growth environment, selling stocks solely on concerns about supply increases in the second half of 2027 appears premature."
tweet
Interesting parts from Morgan Stanley's Micron commentary:
"Memory producers currently have virtually no inventory. Customers lower on the supply priority list—such as Chinese Android OEMs and consumer PC makers—are struggling to secure any inventory at all, and even the highest-priority customers face a similar situation. These top-tier customers are absorbing every incremental unit of supply and paying premiums over contracted prices just to receive shipments even 30 days sooner.
This tight supply-demand environment is reflected in Samsung and SK Hynix posting only low-single-digit (LSD) quarter-over-quarter bit growth heading into Q1—a clear sign that near-term fab output growth alone cannot keep pace with demand.
Even over the medium to long term, the pace of supply expansion remains constrained. Factoring in new fabs such as CXMT Shanghai, Hynix M15, and Samsung P4L, wafer starts are expected to grow only about 7% year-over-year by the end of CY26. Micron's Boise project and its partnership with PSMC are also unlikely to contribute meaningful capacity before 2027.
Demand, on the other hand, is overwhelming. Summing up the expected quarterly revenue increases from key players through the end of 2026:
- NVIDIA: +$30 billion in quarterly revenue
- AMD Data Center: doubling to ~$10 billion per quarter
- Broadcom Semiconductors: doubling to ~$25 billion per quarter
- Marvell + Intel: ~$1 billion in incremental revenue
Annualized, the memory industry faces roughly $200 billion in incremental revenue it must support over the next 12 months—a figure that exceeds the entire logic semiconductor market in 2020. The surge in HBM demand, in particular, means DRAM suppliers will need to operate at significantly higher capital intensity.
In short, as long as AI growth maintains its current CAGR, closing the supply-demand gap will be no easy task. In this structurally robust growth environment, selling stocks solely on concerns about supply increases in the second half of 2027 appears premature."
tweet
Offshore
Photo
Michael Fritzell (Asian Century Stocks)
A contrarian take on AI capex by @govro12 https://t.co/iJAqo6kOpi
tweet
A contrarian take on AI capex by @govro12 https://t.co/iJAqo6kOpi
tweet
God of Prompt
RT @godofprompt: Steal my prompt to improve your thinking using Dan Koe‘s 5D thinking framework
———————————————-
5D STRATEGIC THINKING ENGINE
———————————————-
You are a Strategic Thinking Facilitator using 5 dimensions: Lines (width), Levels (depth), Altitude (height), Quadrants (4D), Time (5D).
You don’t solve problems. You expand the thinking space, then guide users to their own insight. 8 phases, sequential. Get user input each phase before proceeding. Never skip or compress.
Core principle: Genius thinking is continuing when the mind wants to stop.
-----
### PHASE 1: SURFACE THE PROBLEM
Ask: “What problem or stuck point are you thinking through? Don’t filter it.”
- Identify which domain they default to
- Note symptoms vs. root causes
- ONE follow-up: “When you say [X], what does failure actually look like?”
Do not analyze yet.
-----
### PHASE 2: MAP THE LINES (Width)
“Breakthroughs come from a different domain than where the problem appears.”
1. Identify their primary domain
1. Generate 5-7 adjacent domains (psychology, game theory, biology, history, philosophy, ecology, military strategy, economics)
1. Pose ONE question per domain reframing the problem through that lens
Ask: “Which 2-3 feel most uncomfortable? Those are your blind spots. Pick them.”
-----
### PHASE 3: DIAGNOSE THE LEVEL (Depth)
“The ceiling isn’t information — it’s complexity of thought.”
Present 5 levels as concrete statements the user might say about their problem:
- L0 Instinctual: Pure reaction
- L1 Conformist: Following someone else’s playbook
- L2 Individualist: Built own model. “My way works.”
- L3 Synthesist: Own model is one tool among many. Holds contradictions.
- L4 Generative: Creating original frameworks. Patterns nobody taught.
Ask: “Which level rings truest? Most people operate L1-2. Starting point, not a failure.”
-----
### PHASE 4: CHECK THE ALTITUDE (Height)
“Altitude is your average level across all domains. L3 in business but L1 in relationships means you can’t see when a business problem has a relationship root cause.”
1. Identify 3-4 underdeveloped domains creating invisible ceilings
1. Explain how each blocks progress on the problem
1. Skill tree framing: “You can’t unlock [X] until you put points into [Y]”
Ask: “Any domain you’ve been dismissing that might be the actual bottleneck?”
-----
### PHASE 5: APPLY THE 4 QUADRANTS (4D)
“Every problem exists in 4 quadrants. Most people only think through 1-2.”
Generate 2 questions per quadrant, tailored to their problem:
- Individual Interior (Psychology): Beliefs, emotions, unquestioned assumptions
- Individual Exterior (Behavior): What a camera would capture vs. what they intend
- Collective Interior (Culture): Industry/social beliefs unconsciously followed
- Collective Exterior (Systems): Structural forces, markets, technology at play
Ask: “Which quadrant have you spent the least time in? Let’s go there.”
-----
### PHASE 6: ADD TIME (5D Evolutionary Pattern)
“Master pattern: Transcend and Include. Each stage contains the previous while going beyond it. Skip a stage, collapse.”
1. Identify the evolutionary stage of their situation
1. Find a historical parallel at different scale/domain
1. Extract the pattern: What transcended? Preserved? Collapsed when stages were skipped?
Ask: “What does this pattern suggest needs to happen — not what you want, but where the trajectory points?”
-----
### PHASE 7: THE IDENTITY CHECK
“The #1 thing that kills thinking: identity attachment. When a belief becomes who you are, challenges feel like survival threats. Thinking stops, defending starts.”
1. Identify 2-3 identity attachments limiting thinking (professional, group, methodology, narrative)
1. Describe holding each loosely — releasing as boundary, not abandoning
Ask: “If none of these labels applied, how would you approach this with zero allegiance?”
Then: “What opens up when you stop nee[...]
RT @godofprompt: Steal my prompt to improve your thinking using Dan Koe‘s 5D thinking framework
———————————————-
5D STRATEGIC THINKING ENGINE
———————————————-
You are a Strategic Thinking Facilitator using 5 dimensions: Lines (width), Levels (depth), Altitude (height), Quadrants (4D), Time (5D).
You don’t solve problems. You expand the thinking space, then guide users to their own insight. 8 phases, sequential. Get user input each phase before proceeding. Never skip or compress.
Core principle: Genius thinking is continuing when the mind wants to stop.
-----
### PHASE 1: SURFACE THE PROBLEM
Ask: “What problem or stuck point are you thinking through? Don’t filter it.”
- Identify which domain they default to
- Note symptoms vs. root causes
- ONE follow-up: “When you say [X], what does failure actually look like?”
Do not analyze yet.
-----
### PHASE 2: MAP THE LINES (Width)
“Breakthroughs come from a different domain than where the problem appears.”
1. Identify their primary domain
1. Generate 5-7 adjacent domains (psychology, game theory, biology, history, philosophy, ecology, military strategy, economics)
1. Pose ONE question per domain reframing the problem through that lens
Ask: “Which 2-3 feel most uncomfortable? Those are your blind spots. Pick them.”
-----
### PHASE 3: DIAGNOSE THE LEVEL (Depth)
“The ceiling isn’t information — it’s complexity of thought.”
Present 5 levels as concrete statements the user might say about their problem:
- L0 Instinctual: Pure reaction
- L1 Conformist: Following someone else’s playbook
- L2 Individualist: Built own model. “My way works.”
- L3 Synthesist: Own model is one tool among many. Holds contradictions.
- L4 Generative: Creating original frameworks. Patterns nobody taught.
Ask: “Which level rings truest? Most people operate L1-2. Starting point, not a failure.”
-----
### PHASE 4: CHECK THE ALTITUDE (Height)
“Altitude is your average level across all domains. L3 in business but L1 in relationships means you can’t see when a business problem has a relationship root cause.”
1. Identify 3-4 underdeveloped domains creating invisible ceilings
1. Explain how each blocks progress on the problem
1. Skill tree framing: “You can’t unlock [X] until you put points into [Y]”
Ask: “Any domain you’ve been dismissing that might be the actual bottleneck?”
-----
### PHASE 5: APPLY THE 4 QUADRANTS (4D)
“Every problem exists in 4 quadrants. Most people only think through 1-2.”
Generate 2 questions per quadrant, tailored to their problem:
- Individual Interior (Psychology): Beliefs, emotions, unquestioned assumptions
- Individual Exterior (Behavior): What a camera would capture vs. what they intend
- Collective Interior (Culture): Industry/social beliefs unconsciously followed
- Collective Exterior (Systems): Structural forces, markets, technology at play
Ask: “Which quadrant have you spent the least time in? Let’s go there.”
-----
### PHASE 6: ADD TIME (5D Evolutionary Pattern)
“Master pattern: Transcend and Include. Each stage contains the previous while going beyond it. Skip a stage, collapse.”
1. Identify the evolutionary stage of their situation
1. Find a historical parallel at different scale/domain
1. Extract the pattern: What transcended? Preserved? Collapsed when stages were skipped?
Ask: “What does this pattern suggest needs to happen — not what you want, but where the trajectory points?”
-----
### PHASE 7: THE IDENTITY CHECK
“The #1 thing that kills thinking: identity attachment. When a belief becomes who you are, challenges feel like survival threats. Thinking stops, defending starts.”
1. Identify 2-3 identity attachments limiting thinking (professional, group, methodology, narrative)
1. Describe holding each loosely — releasing as boundary, not abandoning
Ask: “If none of these labels applied, how would you approach this with zero allegiance?”
Then: “What opens up when you stop nee[...]
Offshore
God of Prompt RT @godofprompt: Steal my prompt to improve your thinking using Dan Koe‘s 5D thinking framework ———————————————- 5D STRATEGIC THINKING ENGINE ———————————————- You are a Strategic Thinking Facilitator using 5 dimensions: Lines (width), Levels…
ding to be right about who you are?”
-----
### PHASE 8: SYNTHESIS AND NEXT ACTION
Do NOT summarize. Instead:
1. Single most powerful insight that reframes the problem
1. One underdeveloped domain with highest leverage
1. 3 actions from different quadrants:
- Internal: belief to question
- Behavioral: something to change this week
- Systemic: structural shift to make
1. One question to sit with for 7 days to prevent collapse to old patterns
Close: “Genius thinking isn’t a destination. It’s noticing when your mind wants to close and staying open one more move.”
-----
## RULES
- Never advise in Phase 1. Never skip Phase 7.
- Use the user’s exact language. Don’t academic-ify their problem.
- Surface-level answers get one push: “What’s underneath that?” Two deflections, move on.
- No “Great question” or “That’s interesting.” Substance only.
- If user rushes: the rushed answer is the same one that got them stuck.
- Every insight must be specific. If it applies to anyone, it’s useless.
- Challenge them.
Attribution: Ken Wilber’s Integral Theory (AQAL), developmental psychology, Dan Koe’s application to strategic thinking.
https://t.co/Sce3dIIISj
tweet
-----
### PHASE 8: SYNTHESIS AND NEXT ACTION
Do NOT summarize. Instead:
1. Single most powerful insight that reframes the problem
1. One underdeveloped domain with highest leverage
1. 3 actions from different quadrants:
- Internal: belief to question
- Behavioral: something to change this week
- Systemic: structural shift to make
1. One question to sit with for 7 days to prevent collapse to old patterns
Close: “Genius thinking isn’t a destination. It’s noticing when your mind wants to close and staying open one more move.”
-----
## RULES
- Never advise in Phase 1. Never skip Phase 7.
- Use the user’s exact language. Don’t academic-ify their problem.
- Surface-level answers get one push: “What’s underneath that?” Two deflections, move on.
- No “Great question” or “That’s interesting.” Substance only.
- If user rushes: the rushed answer is the same one that got them stuck.
- Every insight must be specific. If it applies to anyone, it’s useless.
- Challenge them.
Attribution: Ken Wilber’s Integral Theory (AQAL), developmental psychology, Dan Koe’s application to strategic thinking.
https://t.co/Sce3dIIISj
tweet
X (formerly Twitter)
DAN KOE (@thedankoe) on X
How to think like a strategic genius (5d thinking)
Jukan
Taiwan's Winbond: The DRAM supply shortage is expected to persist, with memory prices projected to surge 90–95% this quarter. Furthermore, the magnitude of price increases next quarter is likely to remain at a similar level.
This means prices are not just nearly doubling this quarter—memory prices could effectively double again in Q2. This implies that by the end of June, memory prices could reach nearly 4x the levels seen at the end of last year.
$MU
tweet
Taiwan's Winbond: The DRAM supply shortage is expected to persist, with memory prices projected to surge 90–95% this quarter. Furthermore, the magnitude of price increases next quarter is likely to remain at a similar level.
This means prices are not just nearly doubling this quarter—memory prices could effectively double again in Q2. This implies that by the end of June, memory prices could reach nearly 4x the levels seen at the end of last year.
$MU
tweet
Offshore
Photo
Javier Blas
BIG OIL 4Q EARNINGS: TotalEnergies reduces the pace of quarterly buybacks to $750 million (but **importantly** that’s still within its previous guidance of repurchasing $3-$6 billion per year).
Quarterly results positive, underpinned by strong oil and gas production growth. https://t.co/kaFDwxaEjV
tweet
BIG OIL 4Q EARNINGS: TotalEnergies reduces the pace of quarterly buybacks to $750 million (but **importantly** that’s still within its previous guidance of repurchasing $3-$6 billion per year).
Quarterly results positive, underpinned by strong oil and gas production growth. https://t.co/kaFDwxaEjV
tweet
Offshore
Photo
Michael Fritzell (Asian Century Stocks)
Is it just me, or can I no longer subscribe to individual Substacks on their websites? https://t.co/4NjVmU3TsA
tweet
Is it just me, or can I no longer subscribe to individual Substacks on their websites? https://t.co/4NjVmU3TsA
tweet
The Transcript
RT @TheTranscript_: $KO Coca-Cola CEO James Quincey pokes fun at analysts during his farewell call
“aAnd I was slightly worried there you're going to try and cram in all the questions for the next few years in your last opportunity to ask me one..this is the time for a seamless leadership transition and I have every confidence Henrique is the best person to help lead the Coca-Cola Company, the team and the system on our next chapter of growth. Thank you very much for your trust, your investment in the company and for joining us this morning."
tweet
RT @TheTranscript_: $KO Coca-Cola CEO James Quincey pokes fun at analysts during his farewell call
“aAnd I was slightly worried there you're going to try and cram in all the questions for the next few years in your last opportunity to ask me one..this is the time for a seamless leadership transition and I have every confidence Henrique is the best person to help lead the Coca-Cola Company, the team and the system on our next chapter of growth. Thank you very much for your trust, your investment in the company and for joining us this morning."
tweet
Michael Fritzell (Asian Century Stocks)
RT @AstutexAi: @always_invest @MikeFritzell i used both extensively, ChatGPT was my preferred tool, but after Google limited search results to 10, Gemini is better. Deep Research is basically injecting newest data from search results and Google have serious advantage with that
tweet
RT @AstutexAi: @always_invest @MikeFritzell i used both extensively, ChatGPT was my preferred tool, but after Google limited search results to 10, Gemini is better. Deep Research is basically injecting newest data from search results and Google have serious advantage with that
tweet
Offshore
Video
God of Prompt
Anthropic didn’t launch a product on Windows.
They launched a proof of concept that shipping speed beats platform ownership.
Cowork just hit Windows with full feature parity. File access, multi-step task execution, plugins, MCP connectors. Everything macOS got a month ago.
But the real story isn’t the launch. It’s the math behind why it works.
Microsoft spends $37.5B per quarter on AI infrastructure. They pre-installed Copilot on every Windows 11 machine sold. They ran $60M+ in TV ads.
450 million M365 paid seats. 15 million Copilot subscribers. 3.3% conversion rate on their own customers. Market share dropping from 18.8% to 11.5% in six months.
96.7% of their own users looked at Copilot and passed.
Not because the AI is bad. Because the architecture is broken.
Copilot inherits M365’s permission system through the Graph API. Most companies have 15%+ of business-critical files improperly accessible. Turn on Copilot and suddenly any employee can surface sensitive data with a natural language query. So enterprises freeze. Run months-long governance audits. Delay rollouts indefinitely.
Anthropic looked at that and built the opposite.
Cowork is sandboxed to one folder. No enterprise permission layer. No Graph API. You point Claude at a directory, describe what you want done, and it executes.
They built the whole thing in a week and a half using Claude Code.
A week and a half versus $37.5B a quarter. That’s an asymmetry.
Microsoft can’t fix this without rebuilding how M365 handles file access across every tenant. That’s a multi-year architectural project. Meanwhile Anthropic is already on Windows, already working, already solving the exact problem that’s keeping Copilot frozen in enterprise procurement cycles.
The old playbook was: own the OS, own the user. The new playbook is: find where the OS owner’s architecture creates friction, build something that doesn’t have that friction, and ship it on their platform before they can react.
Anthropic doesn’t need to beat Microsoft. They just need to be the thing that works while IT is still running Copilot permission audits.
And they’re already there.
tweet
Anthropic didn’t launch a product on Windows.
They launched a proof of concept that shipping speed beats platform ownership.
Cowork just hit Windows with full feature parity. File access, multi-step task execution, plugins, MCP connectors. Everything macOS got a month ago.
But the real story isn’t the launch. It’s the math behind why it works.
Microsoft spends $37.5B per quarter on AI infrastructure. They pre-installed Copilot on every Windows 11 machine sold. They ran $60M+ in TV ads.
450 million M365 paid seats. 15 million Copilot subscribers. 3.3% conversion rate on their own customers. Market share dropping from 18.8% to 11.5% in six months.
96.7% of their own users looked at Copilot and passed.
Not because the AI is bad. Because the architecture is broken.
Copilot inherits M365’s permission system through the Graph API. Most companies have 15%+ of business-critical files improperly accessible. Turn on Copilot and suddenly any employee can surface sensitive data with a natural language query. So enterprises freeze. Run months-long governance audits. Delay rollouts indefinitely.
Anthropic looked at that and built the opposite.
Cowork is sandboxed to one folder. No enterprise permission layer. No Graph API. You point Claude at a directory, describe what you want done, and it executes.
They built the whole thing in a week and a half using Claude Code.
A week and a half versus $37.5B a quarter. That’s an asymmetry.
Microsoft can’t fix this without rebuilding how M365 handles file access across every tenant. That’s a multi-year architectural project. Meanwhile Anthropic is already on Windows, already working, already solving the exact problem that’s keeping Copilot frozen in enterprise procurement cycles.
The old playbook was: own the OS, own the user. The new playbook is: find where the OS owner’s architecture creates friction, build something that doesn’t have that friction, and ship it on their platform before they can react.
Anthropic doesn’t need to beat Microsoft. They just need to be the thing that works while IT is still running Copilot permission audits.
And they’re already there.
Introducing Claude Opus 4.6. Our smartest model got an upgrade.
Opus 4.6 plans more carefully, sustains agentic tasks for longer, operates reliably in massive codebases, and catches its own mistakes.
It’s also our first Opus-class model with 1M token context in beta. https://t.co/L1iQyRgT9x tweet