Bennett Hoffman retweeted
This image is pretty great, because first, the creators' idea of a great Communist plan is "farming". Under the Czar, Russia was a major exporter of food. Under Lenin, it became a major importer, as harvest after harvest failed. Communists are renowned to be terrible at farming.
But it did get one thing right - it shows modern Americans being forced to farm with 18th century tools and bison rampaging through the fields, which is probably what would happen if we became communists.
Bennett Hoffman retweeted
If any elected leader opposes data centers or the reindustrialization of America, just say you support them in Chinese. Make every reply below their posts in Chinese.
They will eventually go insane.
SpaceX will attempt to catch both the Super Heavy booster and Starship on the upcoming flight.
If both are successful, it will mark the first time that all parts of an orbital rocket return to Earth intact.
This is the critical breakthrough needed to make life multiplanetary.
More than a dozen flight tests and years of development, with countless hours worked by SpaceX engineers and technicians, all come down to this: the first orbital Starship launch carrying advanced Starlink satellites bound for space.
Watch “Go For Orbit” and experience the moments behind flying the most ambitious Starship mission yet
people clearly feel democracy on a spiritual level. the new batch of elected don’t get sworn in until next year, but you already see everybody acting differently from a week ago. including those who have nothing to gain from political posturing
Bennett Hoffman retweeted
20 days ago we built a system that writes GPU code. Today its kernels outperform the state of the art. We just open-sourced 9 megakernels, which outperform NVIDIA’s stack on DeepSeek and run self-driving, robotics, protein folding and world models 2–3× faster than the current best systems. Every model runs in a single launch instead of thousands, and most of our kernels were generated in under an hour.🧵
We invented a recursive megakernel generator called Hyperkernel, the most complete solution to megakernels and GPU efficiency.
Hyperkernel-generated megakernels achieve state of the art performance on 11 different domains: inference (DeepSeek, Kimi and GLM), pretraining (Qwen), hardware fault recovery, robotics (π₀-FAST, ACT) diffusion and real time world models (Matrix-Game), full self driving (OpenPilot), voice models (Moshi), genomics (Evo 2) and protein folding (ESMFold).
All of this in a single kernel launch.
For context, Hyperkernel generated fully optimized megakernels for robotics, genomics and voice in under 50 minutes, and doubled the efficiency for >5 models generally using only about 60-100 lines of code per megakernel. This speaks to the unparalleled standard for code quality in the automated generations.
Our first release today open sources the megakernels themselves. Over the next couple weeks, we will release the entire Hyperkernel stack, along with the complete serving stack for Kimi and GLM (20-30% inference efficiency gains across all concurrencies).
Out of the box GPU machinery performs terribly for unorthodox research. We built Hyperkernel only 20 days ago for internal use; but given its multi-modality, we have decided to open source it.
Creative research should not be bottlenecked by the lack of dedicated infra teams. Hyperkernels offer a one stop shop infra stack for all research. We would love for the open source community to add to it :)
GitHub repo: github.com/Hyperstitioncc/Hy…
NOTE: We primarily use NVIDIA B300s, but our kernel generation extends to other hardware (B200s, AMD, TPUs).
1/9
We absolutely crush NVIDIA and achieve the best low latency inference for DeepSeek v4.1 Flash.
Compared with NVIDIA’s published vLLM results, Hyperkernel delivers:
• 36.7% higher per-user generation speed at concurrency 1.
• 28.2% higher per-user speed while serving two users, versus NVIDIA serving one.
• 15% higher per user speed while serving four users, versus NVIDIA serving one.
• 90.6% higher per-user generation speed at concurrency 4, compared with the highest published C4 AgentX interactivity for TP4 B300s.
• 28.5% higher per-user generation speed at concurrency 8, compared with NVIDIA's published result.
• 36% faster median per-request decode speed.
Fuck yeah
Today, after 31 years, we are saddened to announce that Salon is ceasing publication. It has been an extraordinary privilege. salon.com/2026/11/05/goodbye…
I find the “patch the planet” initiative quite inspiring. openai and others are spending huge sums of money on updating and securing all modern open source software infrastructure. it will be a continual effort to make the world more robust
This is the "upcoming cyber model" from yesterday. Besides 948 Linux vulns, it found 111 renderer RCEs in Firefox, among which, 81 were novel. 30 were exploits for bugs known to developers that were thought to be unexploitable.
We are using this technology to scan more OSS projects and secure the Internet infrastructure.
This is just a start.
Bennett Hoffman retweeted
Notice the anti-datacenter ban started after the Iran offensive
This means that it was a psyop by Iran / CCP
This needs to be investigated by the CIA/FBI
Her post caused 2 trillion in damages to markets and she should be held accountable – at Titusville, FL
Bennett Hoffman retweeted
I’m not a Dan Crenshaw fan by any means but he fucking destroyed this clown
Bennett Hoffman retweeted
the median voter boomers and wine moms do not yet realize serious data center moratoriums make stonks go down. the system self equilibrates after making mistakes.
Actual progress comes from hard-to-build technology.
Which comes from competitive, creative, free-market capitalism.
Which comes from one small, precarious area and then diffuses to the rest of the world.
Snuff that out, and it’s global stagnation for an extremely long time.
Bennett Hoffman retweeted
God people are so retarded
AI is the only path forward to make medical care cheap and affordable
They nuked the data center buildouts not realizing this multi year setback will cause them their health in old age AI could have given them
Bennett Hoffman retweeted
The same people who make art now would make art under ubi. The people who doomscroll and taxi burrito now and say they would make art if they had the time would keep doomscrolling and taxiburritoing.
My most Save The Shrimp opinion in that it would be worth it for the people who did make the art.
You can tell the UBI advocates are more honest than most utopians, because they used to argue that UBI would let the masses live lives of aristocratic self-cultivation and full human expression, then in 2020 we tried it and saw most people use UBI for doordash and doomscrolling, and then they *stopped making that argument* instead of just ignoring the refutation.
Bennett Hoffman retweeted
The irony of killing AI data centers to feed a bloated healthcare system instead is that you are actually killing people relative to the world where you let AI progress unimpeded. – at Port St John, FL
Bennett Hoffman retweeted
HOLY CR*P 🚨 Official Election Administration and Voting Comprehensive Report buried all the way on page 193 of the document shows Texas processed 2,900,357 new valid voter registrations between 2022-2024 without proof of citizenship
Documented proof of citizenship like birth certificate, passport and naturalization papers is not required at the time of voter registration in Texas
56.5% of Texas voter registration were classified as “new valid registrations.”
This is extremely suspicious that 55% of Texas voter registrations were “new”
Bennett Hoffman retweeted
The US needs data centers to win
Data centers multiply your nation's money. It's an IRL money glitch
Imagine the consequences of never building railroads during the industrial revolution, but 100x
For example, SpaceX rents out one MWh of compute for ~$5700. The electricity for that costs ~$75, and with the GPUs and overhead it likely costs ~$1250 all in to produce the MWh of compute.
Fully rented, every $1 you put in gets you $4 out.
This multiple is rising as we build more datacenters and models get smarter. The arbitrage is widening, not closing. Anthropic's revenue per megawatt nearly 4x'd in a single year.
People in France aren't using Mistral.
People in Japan aren't using Sakana.
They use ChatGPT, Gemini, and Claude.
Using any LLM that isn't on the cost / intelligence frontier is foolish, since you'd spend more money for worse results. (See ArtificialAnalysis frontier below)
The only rational choice for any individual abroad is to use American superintelligence, since it gives you the most intelligence per dollar.
Around 28% of Anthropic's revenue was international in the first half of 2026. A lot of Anthropic's revenue now flows to SpaceX, which charges ~$5,700 per MWh for compute that Anthropic turns into ~$6,800 (per SemiAnalysis).
You can see why owning the datacenter is lucrative:
SpaceX makes more profit than Anthropic does for every token Anthropic serves on SpaceX's GPUs.
(Also, it's likely that around half of OpenAI's revenue comes internationally.)
So, when you convert $1 to $4 by putting it into an AI datacenter, around $1.10 to $2 of that is the United States absorbing money from abroad. This money is used to fund more AI development and more data centers, which grows the arbitrage.
Most of these dollars stay domestic, so long as we can build more datacenters here.
As time goes on, superintelligence compute will become the dominant spending category.
Every country that does not have a frontier model, or more specifically, their own datacenters to serve American frontier models, will eventually have a huge share of their spending flowing directly into the US to companies that re-invest in the US.
How does this benefit anyone here?
Many reasons:
- Domestic tax revenue will skyrocket.
- For the average American who can invest in the stock market, asset prices in American companies like SpaceX will skyrocket.
- The dollar gets very strong: other currencies are sold for dollars to buy American tokens, which makes imports much cheaper.
- The US will be able to afford inference from the best models, way more than other countries, which leads to massive increases in productivity in every sector. (With robotics, this means clean, beautiful cities - which I'm excited for).
This is the path to Universal High Income.
In the industrial revolution, this happened with railroads: countries that built railroads and industrialized, such as the US, were able to produce such a high volume of goods for the first time that they absorbed global wealth. The railroad buildout made the US rich.
We can do it again.
The only thing that matters politically in the next few years is securing the future of our datacenter buildout.


































