Offshore
Photo
Ahmad
this is what a chad GPU owner feels next to an api-only coder
donʼt be like Sean, Buy a GPU https://t.co/hYGNggd53z
tweet
this is what a chad GPU owner feels next to an api-only coder
donʼt be like Sean, Buy a GPU https://t.co/hYGNggd53z
tweet
Offshore
Photo
Dimitry Nakhla | Babylon Capital®
Cadence Design $CDNS Q3 2025 Report 🗓️
✅ REV: $1.34B (+10% YoY)
✅ EPS: $1.93 (+17% YoY)
Raising full year revenue outlook to +14% YoY https://t.co/OvQcR1oaMU
tweet
Cadence Design $CDNS Q3 2025 Report 🗓️
✅ REV: $1.34B (+10% YoY)
✅ EPS: $1.93 (+17% YoY)
Raising full year revenue outlook to +14% YoY https://t.co/OvQcR1oaMU
2. Cadence Design $CDNS (Mon AM)
🗓️ REV Est: $1.32B (+14%)
🗓️ EPS Est: $1.79 (+9%)
💵 NTM P/E: 47x
💵 NTM FCF Yield: 1.60% https://t.co/R3To8mlv9w tweet
Offshore
Photo
AkhenOsiris
As others have opined in comments to a previous post, checked around with some buyside, $AMZN stake in Anthropic does appear to be closer to 20% while $GOOGL is in the 10-15% range (these will get diluted with further rounds, and both companies have caps).
Having said that, with the explosive growth Anthropic is seeing, and being further recognized as the enterprise play (WSJ even noting that their business model may be better than OpenAI):
If Amazon screws up from here on out, it would be an immaculate fumble of grand proportions and the guillotine will be out for Jassy.
tweet
As others have opined in comments to a previous post, checked around with some buyside, $AMZN stake in Anthropic does appear to be closer to 20% while $GOOGL is in the 10-15% range (these will get diluted with further rounds, and both companies have caps).
Having said that, with the explosive growth Anthropic is seeing, and being further recognized as the enterprise play (WSJ even noting that their business model may be better than OpenAI):
If Amazon screws up from here on out, it would be an immaculate fumble of grand proportions and the guillotine will be out for Jassy.
I hate to keep harping on AWS, but, anyone who came at looking at the key players capabilties from the viewpoint of AI compute knew this was going to be an issue for AWS several years ago.
2026 is make or break for AWS if they want to be a player in cloud AI workloads. https://t.co/TQleVFeBkU tweet
Offshore
Photo
AkhenOsiris
RT @FundaBottom: Preview| $MSFT FY1Q26: Mid-Training Driving Non-AI Growth, Copilot Studio Becoming the Next PLTR
Copilot Studio Accelerating, Azure Accelerating, OpenAI Accelerating, Mid-Training Accelerating
In our FY1Q26 MSFT Preview, we are pleased to discover new MSFT developments that make us even more bullish:
-Copilot Studio is progressing rapidly with a growth trajectory very similar to PLTR, and we are seeing customers building customer-facing agents using Copilot Studio for the first time.
-Mid-training is not only making Azure GPUs more scarce but is now also creating scarcity in Azure CPU and storage products.
-OpenAI’s contribution to Azure is actually accelerating significantly, not decelerating.
-We observed Azure significantly increasing rack orders in Taiwan today, suggesting Satya is considering a more aggressive AI capex posture. When a CEO known for prudence begins adopting a more aggressive capital expenditure strategy, it typically signals greater optimism about future growth prospects.
Ahead of the Microsoft earnings report, we conducted in-depth interviews with nine experts, providing updates on several topics we consider critical:
-M365 Copilot sales progress and expectations
-Copilot Studio progress
-Azure OpenAI API progress
-GPU rental and data center progress
-Azure AI and MSFT model updates
Detailed Report
https://t.co/Ad9qtgp0ZX
tweet
RT @FundaBottom: Preview| $MSFT FY1Q26: Mid-Training Driving Non-AI Growth, Copilot Studio Becoming the Next PLTR
Copilot Studio Accelerating, Azure Accelerating, OpenAI Accelerating, Mid-Training Accelerating
In our FY1Q26 MSFT Preview, we are pleased to discover new MSFT developments that make us even more bullish:
-Copilot Studio is progressing rapidly with a growth trajectory very similar to PLTR, and we are seeing customers building customer-facing agents using Copilot Studio for the first time.
-Mid-training is not only making Azure GPUs more scarce but is now also creating scarcity in Azure CPU and storage products.
-OpenAI’s contribution to Azure is actually accelerating significantly, not decelerating.
-We observed Azure significantly increasing rack orders in Taiwan today, suggesting Satya is considering a more aggressive AI capex posture. When a CEO known for prudence begins adopting a more aggressive capital expenditure strategy, it typically signals greater optimism about future growth prospects.
Ahead of the Microsoft earnings report, we conducted in-depth interviews with nine experts, providing updates on several topics we consider critical:
-M365 Copilot sales progress and expectations
-Copilot Studio progress
-Azure OpenAI API progress
-GPU rental and data center progress
-Azure AI and MSFT model updates
Detailed Report
https://t.co/Ad9qtgp0ZX
tweet
Offshore
Photo
Quiver Quantitative
BREAKING: Grokipedia appears to be live.
Built by @elonmusk as a competitor to Wikipedia. https://t.co/7Ia5zLz4xU
tweet
BREAKING: Grokipedia appears to be live.
Built by @elonmusk as a competitor to Wikipedia. https://t.co/7Ia5zLz4xU
tweet
Offshore
Photo
Ahmad
seeing cs students grinding this stuff is a W
oxycblt is gmi
tweet
seeing cs students grinding this stuff is a W
oxycblt is gmi
from-scratch transformer update: i may not be modeling language yet but here's some pretty convolutional filters
im still predicting digits but the convolutions i set up should enable me to implement gated convnets for language modeling. while these were obsoleted by the transformer i still want to implement them to compare how they learn language versus attention
it also turns out while implementing convnets that my tensor was still ridiculously slow. with the addition of stride arithmetic and batch parallelism i was able to reduce my training time by 80x from 2 hours/epoch to 1.5 mins/epoch
my next step is quickly implementing a simpler language model (bengio) just to set up learnable embeddings, and then ill finally be able to start making real language models tweet