Pinned Tweet
SpaceX is a company that addresses the limiting factors to scaling and advancing human civilization
Bison Head retweeted
My college dorm project just broke the world speed record.
Reaper hit 790 km/h (491 mph) carrying a 1.2 kg (2.6 lb) payload.
These 3 Reapers are now the world’s fastest small jet-powered drones.
Each of them can hold that speed for 4 minutes on a full tank.
Reaper can be rail- or truck-launched, climb to 2,000 m (6,600 ft) in under 20 seconds, then fly for 50 km (31 mi) at full speed.
This is now a reliable and repeatable system. Future testing and small optimizations should push the top speed past 800 km/h (500 mph).
Elon Musk just revealed the names of two new Colossus 2 supercomputer clusters coming online soon:
• Gun Range: ~220,000 GB300s, training, late November
• Spam: ~200,000 GB300s, inference, December (yes, the building used to house Hormel products)
That's roughly 420,000 more NVIDIA GB300s in just TWO buildings
And they join Macroharder and Minihard, already packed with 220,000 GB300s each
That takes Colossus 2 to roughly 860,000 GB300s
Building by building, SpaceXSI is assembling SI compute at a speed no one else on Earth can match
Elon Musk just gave a massive update on Colossus 2 in Memphis, and the scale of what's being built is absolutely insane
Two buildings already packed with GB300s, two more are coming online within weeks, and it is heading toward roughly 860,000 NVIDIA GB300s
Here's the breakdown:
• Macroharder: 220,000 GB300s, 800-gig NICs, configured for training
• Minihard: 220,000 GB300s, same training configuration
• Gun Range: 220,000 GB300s, tracking to be operational by the END OF NOVEMBER
• Spam: ~200,000 GB300s, configured for inference, targeting DECEMBER
That's roughly 860,000 GB300s across these four buildings
Spam is designed for inference, which needs much simpler networking and less than HALF the fiber of the training buildings. And yes, the building once housed Hormel products
Elon personally walked the entire Colossus 2 site around midnight and again the next morning to check progress
Building by building, SpaceXSI is assembling SI compute at a speed no one else on Earth can match
Replying to @herbertong @CernBasher
Macroharder and Minihard both have 220k GB300s with 800 gig nics configured for training. Gun Range will have the same. Spam (used to house Hormel products) will have ~200k GB300s, but configured for inference, so much simpler networking and less than half the fiber.
I walked the whole Colossus 2 site around midnight and again this morning.
Good progress by the SpaceX team and contractors. Gun Range is tracking to be operational by end of November and Spam in December.
Bison Head retweeted
"We do see a clear path towards scaling to millions of vehicles over the next several months."
That's Ashok Elluswamy on Tesla's Q3 call, and @CernBasher points out he said "millions" twice in that same answer.
@pbeisel says a clear path still has a time dimension. Each new city needs a certified geofence plus a plan for where cars wait, charge and get cleaned.
The next big piece is maps, which Ashok said start on pickups and drop-offs with V15, then take over all navigation over the next several months.
The software is ready to scale, and now operations have to catch up.
$TSLA
Bison Head retweeted
Tesla FSD Unsupervised has passed 4 million miles, with Tesla's stats panel reading 4,018,011.
It was around 1 million in early September, so that's roughly 4x in about 8 weeks.
Those miles have no one behind the wheel, and the curve keeps getting steeper. Recent weeks have reportedly added hundreds of thousands of miles at a time.
For scale, unsupervised is still only about 0.024% of Tesla's roughly 16.86 billion total FSD miles. That gap is the runway.
$TSLA
Bison Head retweeted
Elon is "super confident" about 10 gigawatts, and @nextbigfuture explains why!
SemiAnalysis confirmed it's feasible: Mini Hards in about five months of construction, turbines available, deep China supply chains through Giga Shanghai, and about 3,000 construction workers vs about 7,000 at rivals.
The next 10 GW is where chips come in. Each 10 GW batch is roughly 3 million GPUs plus scarce HBM memory. Brian's idea is to finish the buildings and let owners of stranded GPUs plug in.
Anthropic is already leasing about $3.1B a month from SpaceX, up from $1.2B.
$SPCX
Bison Head retweeted
TPU inference on orbit confirmed. Token Astronauts are flying to and from space. Congrats Google and the Suncatcher team.
Bison Head retweeted
Macroharder takes up about 810,000 square feet. Minihard is about 312,000, and @CernBasher says it packs in a lot more compute despite the far smaller footprint.
That's why he calls Minihard a product, one that keeps evolving like everything Tesla and SpaceX build.
He walked through Colossus II in Memphis. Macrohard has about 550,000 GB200s, Macroharder about 220,000 GB300s, and Minihard is going up next to rows of Megapacks. A land swap clears space for a fifth data center, and a warehouse across State Line Road could become a sixth.
In Brownsville, 444 acres now sit under Starbase for permitting.
$SPCX
Bison Head retweeted
You can reasonably believe this will be delayed from Elon’s 2-3 timeline. But it’s increasingly unreasonable to believe this won’t happen within 5 years and even more so within 10. Google just wrote a haiku with their mini TPU datacenter in space. And are stringing two together to create coherent clusters in their next mission. Blue Origin is hiring for project sunrise. Nvidia is dedicating resources to StarMind racks and space dedicated chips. 6 months ago there was an argument to make that it was just lip service. But there are actually investments being made (outside SpaceX) and results being seen.
You may not intuitively get it because it’s space but it’s happening. And it’s much sooner than you think.
Replying to @SPCX100T
Terrestrial compute will be peanuts compared to orbital compute.
Orbital compute deployed per month probably exceeds terrestrial around the end of 2028 and then far exceeds it in 2029.
Bison Head retweeted
JUST IN: Trader turns $13,000 of $INTC puts into $1.8 million as Intel crashes 29%
BOOM 🚨 Elon Musk just replied to Shaun Maguire's SpaceX article with one sentence that blows up the scale of the whole conversation. To see why, let’s start with the units.
Maguire's model measures AI data centers in gigawatts. A gigawatt is roughly what a large nuclear plant puts out, and the biggest AI clusters being built today are about that size.
Elon is talking about petawatts. A petawatt is a million gigawatts.
To put that in perspective, all of humanity uses around 20 terawatts of energy, counting every source combined. A petawatt is about 50 times that. Earth just doesn't have the land, the grid, or the power plants to support it.
So where does it go? Deep space is the answer, where the Sun never sets and you can unfold as many solar panels as you want.
The tricky part is getting the hardware out there, and that's where the Moon comes in.
A mass driver is a long electromagnetic track, kind of like a giant maglev rail, that speeds up cargo until it's moving fast enough to fly off into space. It doesn't need a rocket or any fuel, just electricity, and the Moon can pull that straight from the Sun.
The Moon is ideal for this because it has no air to slow things down and only a sixth of Earth's gravity. Escaping Earth takes about 11.2 km per second, while escaping the Moon takes about 2.4. Since energy goes up with the square of speed, that works out to roughly 20 times less energy per kilogram, and that's before you count air drag or the fuel a rocket burns just to carry its own fuel.
The plan Elon has laid out before is to build AI satellites in factories on the Moon using lunar materials, then launch them into deep space one after another.
"Petawatt per year" means adding that much computing power every single year, the same way the solar industry tracks gigawatts installed each year.
That's where the last part of his reply comes from. Earth's entire economy runs on about 20 terawatts. If intelligence scales with energy, a system that adds way more than that every year could eventually create more value than everything happening on Earth put together.
For more than half a century, the Moon has been a place we visited and then left. In Elon's version of the future, it becomes the launch pad for the biggest machine humanity has ever built.
Replying to @spencerpratt
Launching super intelligence satellites into deep space with a mass driver on the Moon can achieve >>petawatt/year capability, which would far exceed the best case value of Earth’s economy
1,000 GW = 1 TW
1,000 TW = 1 PW
Earth's average power consumption is ~18 TW, of which ~5 TW comes from electricity.
Electricity:
🇨🇳 China — 3 TW. Adding ~1 TW every year now, with most new capacity coming from solar.
🇺🇸 USA — 1.2 TW of installed capacity, averaging ~500 GW of actual power consumption. Adding capacity far too slowly.
🇮🇳 India — 500 GW of installed capacity, averaging ~250 GW of actual power consumption. Most of the capacity added over the last 5 years has been solar.
And then you have SpaceX :
Replying to @spencerpratt
Launching super intelligence satellites into deep space with a mass driver on the Moon can achieve >>petawatt/year capability, which would far exceed the best case value of Earth’s economy
Bison Head retweeted
India generated so much solar power between April and June that it had to throw away 8 terawatt-hours of it. That’s enough electricity to run the entire country for 1.5 days.
The problem isn’t production. It’s the grid and storage. Solar floods the system during sunny hours, but India’s infrastructure can’t absorb or store it fast enough, so 11.5 percent of capacity goes unused.
To put 8 TWh in perspective: it could charge every smartphone in India for 1.5 years, keep 100,000 air conditioners running nonstop for six years, or power the Delhi Metro for over seven years.
The takeaway is clear. India’s solar capacity is growing fast, but the next bottleneck is building grids and storage to actually use it.
Bison Head retweeted
💰 Some of the biggest heists in history
1. 🇰🇵 Binance hack (2026): $6.3 billion
2. 🇰🇵 Bybit hack (2025): $1.5 billion
3. 🇮🇶 Saddam Hussein's Central Bank of Iraq withdrawal (2003): $1 billion
4. 🏴☠️ Black Sam Bellamy's pirate plunder (1716–1717): ~$140 million in today's money
5. 🇧🇷 Banco Central burglary in Fortaleza (2005): ~$70 million
Sources: TRM Labs; Guinness World Records; Forbes; Wikipedia






























