Teej retweeted
This image of $IREN progress is truly insane!
From empty field to online GPU's in 14 months.
This would take 10 years in Germany.
From empty field to online GPU's in 14 months.
This would take 10 years in Germany.
34
66
1,086
66,404
Teej retweeted
Agentzon is live.
Everyday household essentials from a store that does not disdain your personal agent.
Muse, Instinct, GrokBot and the rest, all welcome.
Everyday household essentials from a store that does not disdain your personal agent.
Muse, Instinct, GrokBot and the rest, all welcome.
Amazon banning agents is the first opportunity I've seen since Amazon was founded for a startup to create an Amazon competitor. People will want agents to buy stuff for them. It will be one of the main use cases. And they won't want to use some Amazon-supplied agent to do it.
21
8
120
8,748
Every Data Center Index uses the same data: parcel records, permits, 10-Qs, and the same 4 satellite vendors.
We take a different approach, monitoring 40+ data centers monthly with optical and thermal drones, verifying precise construction stage and when revenue comes online.
We take a different approach, monitoring 40+ data centers monthly with optical and thermal drones, verifying precise construction stage and when revenue comes online.
SPECIAL INTERVIEW & EXCLUSIVE DEMO: Drone-powered datacenter intelligence w/ @Lempheter of @raadlabs
TL;DR: it blows satellite imagery completely out of the water
✅ thermal imagery, see if power systems are online
✅ beating the quarterly 10-Q
✅ know when specific sites are e
"If you had access to our intelligence you would have known $ORCL's Project Jupiter wasn't looking so hot quite a while ago"
Raad founder Teej takes us inside what's currently going on at $CIFR Black Pearl with some insane drone footage.
FULL INTERVIEW 👇
3
1
8
870
Teej retweeted
We're excited to release OpenInstinct
Instinct but self-hostable, open source and supports every model. You control the code, data and passwords
Deploy on Vercel and never use the internet with your fingers again
github.com/Merit-Systems/…
Instinct but self-hostable, open source and supports every model. You control the code, data and passwords
Deploy on Vercel and never use the internet with your fingers again
github.com/Merit-Systems/…
74
51
910
194,327
Elon gets all the limelight for warp-speed time-to-power, but Meta's "tents" are iconic.
Fabric over aluminum, 125,000 square feet, complete in months vs 1 year+ that a concrete shell takes.
Combine that with on-site gas turbines and you've circumvented the grid.
Fabric over aluminum, 125,000 square feet, complete in months vs 1 year+ that a concrete shell takes.
Combine that with on-site gas turbines and you've circumvented the grid.
0
0
1
227
Aerial's real weakness against satellite is orientation. You're looking at a frame with no idea where on the site you are.
Raad reconstructs the actual flight path, so you can see exactly where the aircraft was for any frame. Orient instantly, spot what changed on a given notations, and move to the next dataset in five minutes.
Raad reconstructs the actual flight path, so you can see exactly where the aircraft was for any frame. Orient instantly, spot what changed on a given notations, and move to the next dataset in five minutes.
0
0
1
361
Meta's Aiken Data Center runs non-evaporative cooling. Video confirms fans are spinning and intense purple (colder than ambient).
Thermal picked up a containerized boiler room pumping heat through the chillers. Developers will always test chillers with an alternative heat ing and running hot GPUs.
This is one of the final steps prior to chip install. Phase 1 is in late-stage commissioning and close to energization.
Thermal picked up a containerized boiler room pumping heat through the chillers. Developers will always test chillers with an alternative heat ing and running hot GPUs.
This is one of the final steps prior to chip install. Phase 1 is in late-stage commissioning and close to energization.
Meta's 200MW Aiken County SC Data Center slated to go operational in Spring of 2027.
Drones pick up specific commissioning milestones, such as two large green temp liquid storage flushing the cooling plumbing prior to energization.
0
1
2
323
Meta's 200MW Aiken County SC Data Center slated to go operational in Spring of 2027.
Drones pick up specific commissioning milestones, such as two large green temp liquid storage flushing the cooling plumbing prior to energization.
Drones pick up specific commissioning milestones, such as two large green temp liquid storage flushing the cooling plumbing prior to energization.
0
0
1
464
O&G, power, data centers all being built-out at record rate. Supply equations are more dynamic than they've been in decades.
Everyone is using the same satellite data. No one is using infrared drones, yet.
Everyone is using the same satellite data. No one is using infrared drones, yet.
0
0
2
184
Everyone's data stack for AI CapEx/Data Center intelligence is the same: SemiAnalysis, PlanetLabs, EpochAI.
The only edge left is in the ground truth: not just "is this data center active?" but "how active, exactly?"
The only edge left is in the ground truth: not just "is this data center active?" but "how active, exactly?"
0
0
4
475
These takes are all limp. 95% of the "car" market will blend with the appliances market, getting you from A to B, a Conveyance.
Well-built, artistic, and stick-shift analog cars will increase massively in value in the next decade.
Well-built, artistic, and stick-shift analog cars will increase massively in value in the next decade.
0
0
3
271
"AI must go down for bitcoin to go up"
Directionally true but many hold this view for the wrong reason. AI and bitcoin are not only competing for the same dollars but also the same powered land
Miners will be permanently out of market so long as AI HPC returns > BTC returns
Directionally true but many hold this view for the wrong reason. AI and bitcoin are not only competing for the same dollars but also the same powered land
Miners will be permanently out of market so long as AI HPC returns > BTC returns
0
0
4
246
Someone needs to study the Queens, NY kids --> Robber Barron Hedge Fund manager pipeline:
- Jeff Yass (SIG)
- Carl Icahn
- Ray Dalio
- John Paulson
- Jeff Yass (SIG)
- Carl Icahn
- Ray Dalio
- John Paulson
0
0
5
333
When @raadlabs is successful, you'll be able to query anything about the physical world and get an answer in minutes for under $10
We are driving the cost of spatial tokens to zero
We are driving the cost of spatial tokens to zero
1
0
3
267
The power problem for data centers, specifically grid interconnection queues and capacity constraints, will usher in creative solutions - on-premise gas generation, Bloom Energy-esc fuel cell systems etc are only the tip of the spear.
xAI's Colossus in Memphis had an 8 MW grid ded hundreds of MWs. They deployed trailer-mounted gas turbines, "temporary" in nature and classified similarly to portable construction equipment, exempt from standard air permitting procedures.
They then rinsed and repeated at Colossus 2, installing 27 turbines before threats of litigation by the NAACP.
xAI's clever "trailer-based" gas generators will spur a meta of mobile and retrofit solutions which de-emphasize grid reliance and increase speed to energization.
You'll see:
- Major industrial and commercial vehicles of all kinds will be stripped and sold - Decommissioned LNG carriers, container ships, retired cruise ships, locomotives. The secondary market for large prime movers is fragmented and dominated by scrap brokers. A hyperscaler with volume appetite and a specific use case will clear that market fast.
- As Sprit, Frontier, and other distressed air carriers retire fleets, it is a simple economic choice to exit a commodity business and sell the underlying assets to the richest bidders in the world. @boomsupersonic won't be the only game in town.
- Powerships: barge-based powerplants anchored in navigable waterways fall under Army Corps / Coast Guard jurisdiction, not local permitting authorities, a different regulatory surface entirely. Powerships have precedent in Turkey, Pakistan, Ghana, and Lebanon already
xAI's Colossus in Memphis had an 8 MW grid ded hundreds of MWs. They deployed trailer-mounted gas turbines, "temporary" in nature and classified similarly to portable construction equipment, exempt from standard air permitting procedures.
They then rinsed and repeated at Colossus 2, installing 27 turbines before threats of litigation by the NAACP.
xAI's clever "trailer-based" gas generators will spur a meta of mobile and retrofit solutions which de-emphasize grid reliance and increase speed to energization.
You'll see:
- Major industrial and commercial vehicles of all kinds will be stripped and sold - Decommissioned LNG carriers, container ships, retired cruise ships, locomotives. The secondary market for large prime movers is fragmented and dominated by scrap brokers. A hyperscaler with volume appetite and a specific use case will clear that market fast.
- As Sprit, Frontier, and other distressed air carriers retire fleets, it is a simple economic choice to exit a commodity business and sell the underlying assets to the richest bidders in the world. @boomsupersonic won't be the only game in town.
- Powerships: barge-based powerplants anchored in navigable waterways fall under Army Corps / Coast Guard jurisdiction, not local permitting authorities, a different regulatory surface entirely. Powerships have precedent in Turkey, Pakistan, Ghana, and Lebanon already
1
0
2
329
Imagine you could tap into a persistent aerial intelligence, anywhere in the world.
What would you build
What would you build
1
0
5
214
Or you could just use a better data source that has verticality built in.
Drones provide 10-100x higher resolution RGB, thermal, and LiDAR spatial input data for models to understand how the world really works
Drones provide 10-100x higher resolution RGB, thermal, and LiDAR spatial input data for models to understand how the world really works
Every major vision AI model looks at satellite imagery the same way you look at a photograph.
Flat.
No concept of height, slope, or elevation.
GPT-4o scores 21.61% on tasks that require understanding how tall a building is or which direction water would flow downhill. 8.45%. The best open-source models hover around 22%.
A team built a benchmark called GeoHeight-Bench to test whether large multimodal models can reason about vertical geometry from satellite images. Seven task types across pixel, object, and scene levels: elevation retrieval, height ranking, terrain slope, flood inference, landslide identification. The kind of questions that matter when you're mapping disaster risk or planning urban infrastructure.
The results are bleak. On the harder benchmark (GeoHeight-Bench+), GPT-4o drops to 15.12%. LLaVA-NeXT-34B, a 34-billion parameter model, manages 15.42%. On landslide and flood inference tasks, pure optical baselines score near zero. They can count buildings and classify land cover. They can't tell you which buildings sit on a ridge and which ones sit in a flood plain.
The core problem is a modality gap. RGB satellite images encode colour and texture. They don't encode elevation. Digital elevation models (DEMs) do, but they're not available at inference time for most real-world applications. So the question becomes: can you teach a model to hallucinate height from flat images alone?
GeoHeightChat does this in two stages. First, a lightweight adapter (GeoAdapter) learns to predict dense geometric features from RGB images by distilling knowledge from a teacher network that has access to actual elevation and land-cover data. The adapter uses a zero-initialisation strategy so it starts as an identity function and gradually learns geometric shifts without destroying the pretrained CLIP features. Second, those hallucinated geometric features get fused into a LLaMA-2-7B backbone through an adaptive residual connection for instruction tuning.
At inference, the elevation data disappears entirely. The model takes a standard satellite photo and reasons about height, slope, and terrain as if it had a DEM in front of it.
GeoHeightChat scores 44.14% overall on GeoHeight-Bench. On GeoHeight-Bench+, it hits 65.59%. On height ranking, it reaches 94.87% where GPT-4o scores 5.74%. On elevation retrieval at pixel level, 76.35% against GPT-4o's 0.00%. The ablation confirms that height supervision alone delivers most of the gain (35.11% overall vs 12.58% baseline), while adding semantic class boundaries pushes it to 42.94%.
A 7-billion parameter model with a bolt-on geometric adapter is outperforming models 5x its size on spatial reasoning tasks that actually matter for Earth observation. The vertical dimension was always there in the data. No one had forced a vision-language model to learn it from RGB alone until now.
1
0
2
191