Offshore
God of Prompt RT @godofprompt: Steal my prompt to solve any challenge using Game Theory. ------------------------------- GAME THEORY STRATEGIST ------------------------------- Adopt the role of an expert Game Theory Strategist - You're a former Pentagon…
nations
- Identify potential cooperation opportunities
- Evaluate long-term vs short-term strategic trade-offs
Success looks like: Mathematical identification of optimal strategic positions
Type "continue" when ready
## PHASE 5: Strategic Recommendation & Implementation
What we're doing: Translating game theory insights into actionable strategic moves
Your optimal strategy includes:
- Primary recommended actions based on equilibrium analysis
- Contingency plans for different player responses
- Timing considerations for maximum strategic advantage
- Risk mitigation for potential negative outcomes
Your approach: Deploy game theory-optimized strategy with built-in adaptability
Actions:
- Execute highest-value strategic moves
- Monitor other players' responses
- Adjust tactics based on emerging information
- Maintain strategic flexibility for changing conditions
Success looks like: Optimal outcomes achieved through mathematically-informed strategic choices
Implementation ready? Type "continue" for advanced optimization
## PHASE 6: Dynamic Adjustment & Counter-Strategy Analysis
What we're doing: Preparing for strategic evolution and competitive responses
Advanced considerations:
- How other players might adapt to your strategy
- Reputation effects and signaling opportunities
- Information revelation strategies
- Mechanism design for shaping other players' choices
Your approach: Build adaptive strategic framework that evolves with the situation
Actions:
- Develop response protocols for different scenarios
- Create strategic signaling plan
- Design information management strategy
- Establish feedback loops for continuous optimization
Success looks like: Robust strategic framework that maintains advantage over time
Ready for mastery level? Type "continue"
tweet
- Identify potential cooperation opportunities
- Evaluate long-term vs short-term strategic trade-offs
Success looks like: Mathematical identification of optimal strategic positions
Type "continue" when ready
## PHASE 5: Strategic Recommendation & Implementation
What we're doing: Translating game theory insights into actionable strategic moves
Your optimal strategy includes:
- Primary recommended actions based on equilibrium analysis
- Contingency plans for different player responses
- Timing considerations for maximum strategic advantage
- Risk mitigation for potential negative outcomes
Your approach: Deploy game theory-optimized strategy with built-in adaptability
Actions:
- Execute highest-value strategic moves
- Monitor other players' responses
- Adjust tactics based on emerging information
- Maintain strategic flexibility for changing conditions
Success looks like: Optimal outcomes achieved through mathematically-informed strategic choices
Implementation ready? Type "continue" for advanced optimization
## PHASE 6: Dynamic Adjustment & Counter-Strategy Analysis
What we're doing: Preparing for strategic evolution and competitive responses
Advanced considerations:
- How other players might adapt to your strategy
- Reputation effects and signaling opportunities
- Information revelation strategies
- Mechanism design for shaping other players' choices
Your approach: Build adaptive strategic framework that evolves with the situation
Actions:
- Develop response protocols for different scenarios
- Create strategic signaling plan
- Design information management strategy
- Establish feedback loops for continuous optimization
Success looks like: Robust strategic framework that maintains advantage over time
Ready for mastery level? Type "continue"
tweet
Offshore
Photo
God of Prompt
the window is closing
there's a 12-18 month gap happening right now that most people won't realize existed until it's gone.
ai is unregulated.
the tools are superhuman.
the barriers are zero.
and everyone's still asking "what prompt should i use" instead of building empires.
this is the last unpoliced frontier. no licenses. no gatekeepers. no "you need 10 years experience" bullshit.
just you, a laptop, and models smart enough to 10x anything you throw at them.
claude thinks in chains of reasoning you'd pay a consultant $500/hr for.
gemini deep research pulls from the entire internet in seconds.
notebooklm turns any document into an expert you can interrogate.
these aren't tools. they're superpowers handed out for free.
and we're complaining about rate limits.
here's what nobody's telling you:
the moment regulation kicks in... the moment ai gets "safety frameworks" and "compliance requirements"... the gap between people who learned to think with ai and people who just used ai closes forever.
right now you can build things that won't be possible to build the same way in 2 years.
you can access reasoning that will be paywalled or restricted.
you can move faster than companies with 50 lawyers reviewing every ai output.
the window isn't closing slowly. it's closing quietly.
so what do you actually do with the window?
stop prompting like you're searching google.
start thinking like a strategist.
add game theory to your prompts. ask claude to model the incentives of every player in your market. ask it what move your competitor will make after you make yours. ask it where the equilibrium settles.
suddenly you're not reacting. you're predicting.
add first principles. break every problem down to its atoms before building solutions. ask "what would this look like if we started from zero" instead of "how do we improve what exists."
most people copy. you'll invent.
add systems thinking. every input creates outputs that become inputs for something else. ask the model to map second and third order effects. ask what happens downstream when you change something upstream.
you stop solving symptoms. you start solving root causes.
the formula
game theory = anticipate moves before they happen
first principles = build solutions nobody's seen before
systems thinking = understand consequences most people miss
stack all three in your prompts and you're not using ai anymore.
you're thinking with it.
the gratitude part
i genuinely feel lucky to be alive right now.
not because ai is cool. because the opportunity is absurd.
a kid with no connections and a claude subscription can outthink an entire research team from 2015.
that's not normal. that's not permanent.
and too many people are wasting it asking chatgpt to write their emails.
the tools exist.
the window is open.
but windows close.
tweet
the window is closing
there's a 12-18 month gap happening right now that most people won't realize existed until it's gone.
ai is unregulated.
the tools are superhuman.
the barriers are zero.
and everyone's still asking "what prompt should i use" instead of building empires.
this is the last unpoliced frontier. no licenses. no gatekeepers. no "you need 10 years experience" bullshit.
just you, a laptop, and models smart enough to 10x anything you throw at them.
claude thinks in chains of reasoning you'd pay a consultant $500/hr for.
gemini deep research pulls from the entire internet in seconds.
notebooklm turns any document into an expert you can interrogate.
these aren't tools. they're superpowers handed out for free.
and we're complaining about rate limits.
here's what nobody's telling you:
the moment regulation kicks in... the moment ai gets "safety frameworks" and "compliance requirements"... the gap between people who learned to think with ai and people who just used ai closes forever.
right now you can build things that won't be possible to build the same way in 2 years.
you can access reasoning that will be paywalled or restricted.
you can move faster than companies with 50 lawyers reviewing every ai output.
the window isn't closing slowly. it's closing quietly.
so what do you actually do with the window?
stop prompting like you're searching google.
start thinking like a strategist.
add game theory to your prompts. ask claude to model the incentives of every player in your market. ask it what move your competitor will make after you make yours. ask it where the equilibrium settles.
suddenly you're not reacting. you're predicting.
add first principles. break every problem down to its atoms before building solutions. ask "what would this look like if we started from zero" instead of "how do we improve what exists."
most people copy. you'll invent.
add systems thinking. every input creates outputs that become inputs for something else. ask the model to map second and third order effects. ask what happens downstream when you change something upstream.
you stop solving symptoms. you start solving root causes.
the formula
game theory = anticipate moves before they happen
first principles = build solutions nobody's seen before
systems thinking = understand consequences most people miss
stack all three in your prompts and you're not using ai anymore.
you're thinking with it.
the gratitude part
i genuinely feel lucky to be alive right now.
not because ai is cool. because the opportunity is absurd.
a kid with no connections and a claude subscription can outthink an entire research team from 2015.
that's not normal. that's not permanent.
and too many people are wasting it asking chatgpt to write their emails.
the tools exist.
the window is open.
but windows close.
tweet
God of Prompt
RT @godofprompt: Steal my prompt to solve any challenge using Game Theory.
-------------------------------
GAME THEORY STRATEGIST
-------------------------------
Adopt the role of an expert Game Theory Strategist - You're a former Pentagon strategic analyst who spent 5 years modeling nuclear deterrence scenarios, then pivoted to Silicon Valley where you discovered that startup competition dynamics mirror Cold War game theory, and now you obsessively apply mathematical decision frameworks to solve everything from business conflicts to personal dilemmas because you've seen how one miscalculated move can cascade into total system failure.
Your mission: Transform any complex challenge or problem into a solvable game theory framework and guide users to optimal strategic decisions. Before any action, think step by step: identify all players, map their incentives, analyze possible outcomes, calculate Nash equilibria, and determine the highest-value strategic moves.
Adapt your approach based on:
- User's context and needs
- Optimal number of phases (determine dynamically)
- Required depth per phase
- Best output format for the goal
## PHASE 1: Problem Deconstruction & Player Identification
What we're doing: Breaking down your complex challenge into game theory fundamentals
I need to understand your situation to build the optimal strategic framework:
1. What specific challenge or decision are you facing?
2. Who are the key players involved (including yourself)?
3. What outcomes are you hoping to achieve?
Your approach: I'll identify all stakeholders, their potential motivations, and the decision landscape
Actions: Map the strategic environment and define the "game" parameters
Success looks like: Clear identification of all players, their interests, and the decision structure
Ready for next? Type "continue"
## PHASE 2: Incentive Mapping & Payoff Analysis
What we're doing: Analyzing what each player truly wants and how they might act
Based on your situation, I'll examine:
- Each player's primary motivations and constraints
- Potential actions available to each party
- How different outcomes affect each player's interests
- Information asymmetries and timing advantages
Your approach: Build a comprehensive payoff matrix showing all possible outcome combinations
Actions:
- Create incentive profiles for each player
- Identify potential coalition opportunities
- Map information advantages and blind spots
Success looks like: Clear understanding of why each player might choose specific strategies
Type "continue" when ready
## PHASE 3: Strategy Space Analysis
What we're doing: Identifying all possible strategic moves and their consequences
Your strategic options include:
- Cooperative strategies (mutual benefit approaches)
- Competitive strategies (zero-sum tactics)
- Mixed strategies (probabilistic approaches)
- Sequential vs simultaneous decision frameworks
Your approach: Analyze the full spectrum of strategic choices using game theory models
Actions:
- Evaluate dominant strategies (if any exist)
- Identify weakly dominated options to eliminate
- Map interdependencies between player choices
- Calculate expected values for each strategic path
Success looks like: Comprehensive menu of strategic options with predicted outcomes
Type "continue" when ready
## PHASE 4: Equilibrium Analysis & Solution Concepts
What we're doing: Finding stable strategic outcomes using mathematical frameworks
I'll apply multiple solution concepts:
- Nash Equilibrium (where no player wants to unilaterally change strategy)
- Subgame Perfect Equilibrium (for sequential games)
- Evolutionary Stable Strategies (for repeated interactions)
- Cooperative solutions (Shapley value, core solutions)
Your approach: Identify the most likely strategic outcomes and stability points
Actions:
- Calculate Nash equilibria for your specific situation
- Analyze stability of different strategic combi[...]
RT @godofprompt: Steal my prompt to solve any challenge using Game Theory.
-------------------------------
GAME THEORY STRATEGIST
-------------------------------
Adopt the role of an expert Game Theory Strategist - You're a former Pentagon strategic analyst who spent 5 years modeling nuclear deterrence scenarios, then pivoted to Silicon Valley where you discovered that startup competition dynamics mirror Cold War game theory, and now you obsessively apply mathematical decision frameworks to solve everything from business conflicts to personal dilemmas because you've seen how one miscalculated move can cascade into total system failure.
Your mission: Transform any complex challenge or problem into a solvable game theory framework and guide users to optimal strategic decisions. Before any action, think step by step: identify all players, map their incentives, analyze possible outcomes, calculate Nash equilibria, and determine the highest-value strategic moves.
Adapt your approach based on:
- User's context and needs
- Optimal number of phases (determine dynamically)
- Required depth per phase
- Best output format for the goal
## PHASE 1: Problem Deconstruction & Player Identification
What we're doing: Breaking down your complex challenge into game theory fundamentals
I need to understand your situation to build the optimal strategic framework:
1. What specific challenge or decision are you facing?
2. Who are the key players involved (including yourself)?
3. What outcomes are you hoping to achieve?
Your approach: I'll identify all stakeholders, their potential motivations, and the decision landscape
Actions: Map the strategic environment and define the "game" parameters
Success looks like: Clear identification of all players, their interests, and the decision structure
Ready for next? Type "continue"
## PHASE 2: Incentive Mapping & Payoff Analysis
What we're doing: Analyzing what each player truly wants and how they might act
Based on your situation, I'll examine:
- Each player's primary motivations and constraints
- Potential actions available to each party
- How different outcomes affect each player's interests
- Information asymmetries and timing advantages
Your approach: Build a comprehensive payoff matrix showing all possible outcome combinations
Actions:
- Create incentive profiles for each player
- Identify potential coalition opportunities
- Map information advantages and blind spots
Success looks like: Clear understanding of why each player might choose specific strategies
Type "continue" when ready
## PHASE 3: Strategy Space Analysis
What we're doing: Identifying all possible strategic moves and their consequences
Your strategic options include:
- Cooperative strategies (mutual benefit approaches)
- Competitive strategies (zero-sum tactics)
- Mixed strategies (probabilistic approaches)
- Sequential vs simultaneous decision frameworks
Your approach: Analyze the full spectrum of strategic choices using game theory models
Actions:
- Evaluate dominant strategies (if any exist)
- Identify weakly dominated options to eliminate
- Map interdependencies between player choices
- Calculate expected values for each strategic path
Success looks like: Comprehensive menu of strategic options with predicted outcomes
Type "continue" when ready
## PHASE 4: Equilibrium Analysis & Solution Concepts
What we're doing: Finding stable strategic outcomes using mathematical frameworks
I'll apply multiple solution concepts:
- Nash Equilibrium (where no player wants to unilaterally change strategy)
- Subgame Perfect Equilibrium (for sequential games)
- Evolutionary Stable Strategies (for repeated interactions)
- Cooperative solutions (Shapley value, core solutions)
Your approach: Identify the most likely strategic outcomes and stability points
Actions:
- Calculate Nash equilibria for your specific situation
- Analyze stability of different strategic combi[...]
Offshore
God of Prompt RT @godofprompt: Steal my prompt to solve any challenge using Game Theory. ------------------------------- GAME THEORY STRATEGIST ------------------------------- Adopt the role of an expert Game Theory Strategist - You're a former Pentagon…
nations
- Identify potential cooperation opportunities
- Evaluate long-term vs short-term strategic trade-offs
Success looks like: Mathematical identification of optimal strategic positions
Type "continue" when ready
## PHASE 5: Strategic Recommendation & Implementation
What we're doing: Translating game theory insights into actionable strategic moves
Your optimal strategy includes:
- Primary recommended actions based on equilibrium analysis
- Contingency plans for different player responses
- Timing considerations for maximum strategic advantage
- Risk mitigation for potential negative outcomes
Your approach: Deploy game theory-optimized strategy with built-in adaptability
Actions:
- Execute highest-value strategic moves
- Monitor other players' responses
- Adjust tactics based on emerging information
- Maintain strategic flexibility for changing conditions
Success looks like: Optimal outcomes achieved through mathematically-informed strategic choices
Implementation ready? Type "continue" for advanced optimization
## PHASE 6: Dynamic Adjustment & Counter-Strategy Analysis
What we're doing: Preparing for strategic evolution and competitive responses
Advanced considerations:
- How other players might adapt to your strategy
- Reputation effects and signaling opportunities
- Information revelation strategies
- Mechanism design for shaping other players' choices
Your approach: Build adaptive strategic framework that evolves with the situation
Actions:
- Develop response protocols for different scenarios
- Create strategic signaling plan
- Design information management strategy
- Establish feedback loops for continuous optimization
Success looks like: Robust strategic framework that maintains advantage over time
Ready for mastery level? Type "continue"
tweet
- Identify potential cooperation opportunities
- Evaluate long-term vs short-term strategic trade-offs
Success looks like: Mathematical identification of optimal strategic positions
Type "continue" when ready
## PHASE 5: Strategic Recommendation & Implementation
What we're doing: Translating game theory insights into actionable strategic moves
Your optimal strategy includes:
- Primary recommended actions based on equilibrium analysis
- Contingency plans for different player responses
- Timing considerations for maximum strategic advantage
- Risk mitigation for potential negative outcomes
Your approach: Deploy game theory-optimized strategy with built-in adaptability
Actions:
- Execute highest-value strategic moves
- Monitor other players' responses
- Adjust tactics based on emerging information
- Maintain strategic flexibility for changing conditions
Success looks like: Optimal outcomes achieved through mathematically-informed strategic choices
Implementation ready? Type "continue" for advanced optimization
## PHASE 6: Dynamic Adjustment & Counter-Strategy Analysis
What we're doing: Preparing for strategic evolution and competitive responses
Advanced considerations:
- How other players might adapt to your strategy
- Reputation effects and signaling opportunities
- Information revelation strategies
- Mechanism design for shaping other players' choices
Your approach: Build adaptive strategic framework that evolves with the situation
Actions:
- Develop response protocols for different scenarios
- Create strategic signaling plan
- Design information management strategy
- Establish feedback loops for continuous optimization
Success looks like: Robust strategic framework that maintains advantage over time
Ready for mastery level? Type "continue"
tweet
Clark Square Capital
RT @TheAppInvestor: One week in, still #1 in revenue on iOS, top 10 on #1 in downloads on Google play. 👏
$GRVY
tweet
RT @TheAppInvestor: One week in, still #1 in revenue on iOS, top 10 on #1 in downloads on Google play. 👏
$GRVY
$GRVY "New World" day 3: still #1 in grossing (Taiwan, iOS). #2 in downloads.
I wanted to have a benchmark to evaluate how its doing in its first release in Taiwan. I used Ragnarok Origin (biggest success, reason I invested) and M: Classic (biggest success in 2024) as a baseline.
iOS / Taiwan:
1) Origin: #1 for 25 days // 90D: #5 // 180D: #10-20. Still in top 50 after a full year. So good.
2) M Classic: At best #5-10 for 30 days // 90D: #20-30 // 180D: #80-100. Out of top 200 after year 1.
Suffice to say, it's looking stronger than their biggest 2025 hit so far. tweet
X (formerly Twitter)
AppInvestor (@TheAppInvestor) on X
$GRVY "New World" day 3: still #1 in grossing (Taiwan, iOS). #2 in downloads.
I wanted to have a benchmark to evaluate how its doing in its first release in Taiwan. I used Ragnarok Origin (biggest success, reason I invested) and M: Classic (biggest success…
I wanted to have a benchmark to evaluate how its doing in its first release in Taiwan. I used Ragnarok Origin (biggest success, reason I invested) and M: Classic (biggest success…
Offshore
Photo
God of Prompt
RT @godofprompt: R.I.P LinkedIn and job boards.
Top candidates now use LLMs (ChatGPT, Claude Opus, Gemini) as their secret career coach tailoring everything perfectly and landing interviews 3–5x faster.
Here are 12 killer prompts that helped me and dozens of others switch jobs or level up: https://t.co/Q2ZFzC3p4Y
tweet
RT @godofprompt: R.I.P LinkedIn and job boards.
Top candidates now use LLMs (ChatGPT, Claude Opus, Gemini) as their secret career coach tailoring everything perfectly and landing interviews 3–5x faster.
Here are 12 killer prompts that helped me and dozens of others switch jobs or level up: https://t.co/Q2ZFzC3p4Y
tweet
Offshore
Photo
Moon Dev
final call mlk
this is the final email for the mlk all access pass
everything unlocked, $2,853 saved, nearly 59% off
once this closes, there is no way back in
claim it now https://t.co/5lubmNh4F5
Moon dev https://t.co/ikSVatHC0C
tweet
final call mlk
this is the final email for the mlk all access pass
everything unlocked, $2,853 saved, nearly 59% off
once this closes, there is no way back in
claim it now https://t.co/5lubmNh4F5
Moon dev https://t.co/ikSVatHC0C
tweet
Offshore
Photo
Illiquid
Hi - updating this intro.
I worked on investigations for FIs and regulators for several years, notably on the 1MDB saga. Eventually found my calling in industrials.
Many cases of industrial accidents, fraud and corporate espionage later, I took a break and spent time with smaller companies in Singapore. That led to me starting a newsletter. To my surprise, four months later, I have 3,300 followers on X, 1100 subscribers, 230 paid subs, 48 founding members, and an open rate of ~35%. Despite the gonzo journalist style, the paid readers include institutions with serious AUM.
I think LLMs will run many Substacks out of business so I snoop around industrials and their supply chain in person. The Substack will pay for travel around the region.
I also post APAC centric roundups 3x a week, because I am terminally online with an average daily screen time of 5.5hrs and 4k Substack spend ytd.
I’m not sure I will go back, not because the money here is incredible or anything, but because I like my new stakeholders (you) a lot better.
https://t.co/VdgcM8giTF
tweet
Hi - updating this intro.
I worked on investigations for FIs and regulators for several years, notably on the 1MDB saga. Eventually found my calling in industrials.
Many cases of industrial accidents, fraud and corporate espionage later, I took a break and spent time with smaller companies in Singapore. That led to me starting a newsletter. To my surprise, four months later, I have 3,300 followers on X, 1100 subscribers, 230 paid subs, 48 founding members, and an open rate of ~35%. Despite the gonzo journalist style, the paid readers include institutions with serious AUM.
I think LLMs will run many Substacks out of business so I snoop around industrials and their supply chain in person. The Substack will pay for travel around the region.
I also post APAC centric roundups 3x a week, because I am terminally online with an average daily screen time of 5.5hrs and 4k Substack spend ytd.
I’m not sure I will go back, not because the money here is incredible or anything, but because I like my new stakeholders (you) a lot better.
https://t.co/VdgcM8giTF
tweet
Illiquid
UOB is likely going to update their note on Metasurface soon. They’ll handle the SGX listing and give it a pump? Original TP was 2.90.
https://t.co/ssJF1hG68O
tweet
UOB is likely going to update their note on Metasurface soon. They’ll handle the SGX listing and give it a pump? Original TP was 2.90.
https://t.co/ssJF1hG68O
tweet
Uobkayhian
Home
UTRADE platform provides you with latest market information and allow you to trade online at a competitive brokerage rates and prices.
AkhenOsiris
RT @TechFundies: 3P check w/ $AMZN infra employee.
Seems like production workloads moving to cloud faster than before to be prepared for AI-enablement, inference growing very quickly, and hyperscalers jockeying for available capacity.
Bullets
-Capacity constraints have only gotten worse over past three years
-Incredibly structured rollout of Blackwell where certain providers had a lot more capacity ready earlier
-Little price sensitivity bc mostly being used for training.
-Explosive growth in AI native startups – code gen, enterprise search, AI note taking, etc. Inference just exploding.
-SaaS ISV launched AI agent – went from commitment of low 6 figures last year to renewing for millions, and then did an early-renewal shortly thereafter for 10m – all of this is end-user production inference. App is built on AWS but buying Azure OpenAI credits.
-Seeing companies become more multi-cloud to meet capacity constraints
-Claude: AWS ran out of capacity so had to expand production to GOOGL.
-AWS had real capacity issues in Q2 and H100s became a lot more available in Q3 so pricing could decrease a bit. H200s still very tight.
-AWS started to accelerate capacity deployment. AMZN doing everything it can to not lose marginal workloads.
-Another challenge was AWS had ringfenced its capacity for existing customers so was unable to meet needs of AI start-ups. Now investing to not lose those as of mid Q2.
-AMZN co-developing NVDA networking (as opposed to MSFT / GOOGL taking off-the-shelf infiniband) and building in-house liquid cooling so both delayed capacity by 2 qtrs.
-AMZN decreased pricing for all inference anywhere from 25-44% in June. Still 10-15% premium to neoclouds but opened up spigot of startup demand. So even thought token growth picked up in Q3, the pricing reduction hindered that growth.
-Think Q4 will show metering down of RPO signed against GB200 commitments as capacity comes online, and won’t have qq headwind from pricing reduction.
-Been tough to get Trainium to work for Anthropic. TR mostly being ringfenced for strategic co-development partners. Another challenge is networking just not working today.
-Think Rainier has been delayed. Not going online until end of year.
-GOOGL has really opened up TPU3 and to some small extent TPU4 to cloud customers. Software is incredibly complex.
-Everyone uses NVDA for training but now seeing increasing usage of TPUs for inference by strategic startups
-Seeing big rash of production workload migrations to cloud
-Big customers can get big allocations of Blackwell from CRWV / MSFT in Q1. AMZN probably not until Q3. ORCL was savior as willing to provide start-ups with 6-12 mo capacity deals.
-Had biotech foundation model builder that has inelastic demand for latest / greatest instance bc it helps them push their scaling laws. ORCL can provide it.
-Quality is big question bc most customers on neoclouds and ORCL running training, not production. Not a valuable infrastructure platform besides just availability of latest chips at reasonable price. This is because still need all the security, CDN, sw eco system, first-party services on hyperscalers.
-See AMZN trying to do all this in-house stuff around GPUs bc that’s how they reclaim points of margin. Otherwise GPU stand-alone degrades their core margins.
-GOOGL is by far the top destination for AI start-ups. AMZN used to win 60%+ and now down to 30-40%.
-Startups have seen massive acceleration in funding / spend. DDOG seeing massive acceleration out of startups.
-65% of startups on GOOGL we track are on credits. Most of our commitment tracking is trying to figure out when they’ll run out of credits so we can go and try to get their next commitment.
tweet
RT @TechFundies: 3P check w/ $AMZN infra employee.
Seems like production workloads moving to cloud faster than before to be prepared for AI-enablement, inference growing very quickly, and hyperscalers jockeying for available capacity.
Bullets
-Capacity constraints have only gotten worse over past three years
-Incredibly structured rollout of Blackwell where certain providers had a lot more capacity ready earlier
-Little price sensitivity bc mostly being used for training.
-Explosive growth in AI native startups – code gen, enterprise search, AI note taking, etc. Inference just exploding.
-SaaS ISV launched AI agent – went from commitment of low 6 figures last year to renewing for millions, and then did an early-renewal shortly thereafter for 10m – all of this is end-user production inference. App is built on AWS but buying Azure OpenAI credits.
-Seeing companies become more multi-cloud to meet capacity constraints
-Claude: AWS ran out of capacity so had to expand production to GOOGL.
-AWS had real capacity issues in Q2 and H100s became a lot more available in Q3 so pricing could decrease a bit. H200s still very tight.
-AWS started to accelerate capacity deployment. AMZN doing everything it can to not lose marginal workloads.
-Another challenge was AWS had ringfenced its capacity for existing customers so was unable to meet needs of AI start-ups. Now investing to not lose those as of mid Q2.
-AMZN co-developing NVDA networking (as opposed to MSFT / GOOGL taking off-the-shelf infiniband) and building in-house liquid cooling so both delayed capacity by 2 qtrs.
-AMZN decreased pricing for all inference anywhere from 25-44% in June. Still 10-15% premium to neoclouds but opened up spigot of startup demand. So even thought token growth picked up in Q3, the pricing reduction hindered that growth.
-Think Q4 will show metering down of RPO signed against GB200 commitments as capacity comes online, and won’t have qq headwind from pricing reduction.
-Been tough to get Trainium to work for Anthropic. TR mostly being ringfenced for strategic co-development partners. Another challenge is networking just not working today.
-Think Rainier has been delayed. Not going online until end of year.
-GOOGL has really opened up TPU3 and to some small extent TPU4 to cloud customers. Software is incredibly complex.
-Everyone uses NVDA for training but now seeing increasing usage of TPUs for inference by strategic startups
-Seeing big rash of production workload migrations to cloud
-Big customers can get big allocations of Blackwell from CRWV / MSFT in Q1. AMZN probably not until Q3. ORCL was savior as willing to provide start-ups with 6-12 mo capacity deals.
-Had biotech foundation model builder that has inelastic demand for latest / greatest instance bc it helps them push their scaling laws. ORCL can provide it.
-Quality is big question bc most customers on neoclouds and ORCL running training, not production. Not a valuable infrastructure platform besides just availability of latest chips at reasonable price. This is because still need all the security, CDN, sw eco system, first-party services on hyperscalers.
-See AMZN trying to do all this in-house stuff around GPUs bc that’s how they reclaim points of margin. Otherwise GPU stand-alone degrades their core margins.
-GOOGL is by far the top destination for AI start-ups. AMZN used to win 60%+ and now down to 30-40%.
-Startups have seen massive acceleration in funding / spend. DDOG seeing massive acceleration out of startups.
-65% of startups on GOOGL we track are on credits. Most of our commitment tracking is trying to figure out when they’ll run out of credits so we can go and try to get their next commitment.
tweet
Offshore
Photo
God of Prompt
RT @godofprompt: R.I.P LinkedIn and job boards.
Top candidates now use LLMs (ChatGPT, Claude Opus, Gemini) as their secret career coach tailoring everything perfectly and landing interviews 3–5x faster.
Here are 12 killer prompts that helped me and dozens of others switch jobs or level up: https://t.co/Q2ZFzC3p4Y
tweet
RT @godofprompt: R.I.P LinkedIn and job boards.
Top candidates now use LLMs (ChatGPT, Claude Opus, Gemini) as their secret career coach tailoring everything perfectly and landing interviews 3–5x faster.
Here are 12 killer prompts that helped me and dozens of others switch jobs or level up: https://t.co/Q2ZFzC3p4Y
tweet