Rocket engineering is like building a complex software program where you can only test tiny pieces on a slightly different machine. The very first time you run the whole thing on the actual hardware it has to work flawlessly with basically zero major bugs
Many people have claimed that with AI-assisted bug finding, secure code (and hence trustless anything) will be impossible.
I have a much more optimistic take, and AI-assisted formal verification is a major part of the reason why:
https://t.co/0ceMBZ6uqj
It's deeply irresponsible to put dangerous AGI into the hands of the general public. I'm boycotting every model company that spreads bioterrorism and cybersecurity risks
That's why I quit OpenAI years ago when they launched GPT-2
@elonmusk Running the chips hotter in space is actually a big optimization, and it's an underrated detail in this interview. Radiated power scales with T^4, so raising the temperature by ~20% in Kelvin can cut the required radiator area roughly in half
@yuris Individual 'grindmaxxing' is overrated, although one must be capable of that.
Teamworkmaxxing in retrospect has always created more value and generated more Right directional signal. It's more Complex to join a great team, than to just work hard, but that's more winning than 1/
Sometimes you gotta risk being the aggressively uncompromising non-conformist to protect the taste and design. You say No to everyone else who wants it differently & "fashionably". Thatâs how to prevent it from getting ruined or ending up mediocre.
Why did xAI hand over a 220,000-GPU cluster to Anthropic?
The technical backdrop to xAI's decision to hand Colossus 1 over to Anthropic in its entirety is more interesting than it appears. xAI deployed more than 220,000 NVIDIA GPUs at its Colossus 1 data center in Memphis. Of these, roughly 150,000 are estimated to be H100s, 50,000 H200s, and 20,000 GB200s. In other words, three different generations of silicon are mixed together inside a single cluster â a "heterogeneous architecture."
For distributed training, however, this configuration is close to a disaster, according to engineers familiar with the setup. In distributed training, 100,000 GPUs must finish a single step simultaneously before the cluster can advance to the next one. Even if the GB200s finish their computation first, the remaining 99,999 chips have to wait for the slower H100s â or for any GPU that has hit a stack-related snag â to catch up. This is known as the straggler effect. The 11% GPU utilization rate (MFU: the share of theoretical FLOPs actually realized) at xAI recently reported by The Information can be read as the numerical fallout of this problem. It stands in stark contrast to the 40%-plus MFU figures achieved by Meta and Google.
The problem runs deeper still. As discussed earlier, NVIDIA's NCCL has traditionally been optimized for a ring topology. It works beautifully at the 1,000â10,000 GPU scale, but once you push into the 100,000-unit range, the latency of data traversing the ring once around becomes punishingly long. GPUs need to churn through computations rapidly to keep MFU high, but while they sit waiting endlessly for data to arrive over the network fabric, more than half of the silicon falls into idle. Google sidestepped this bottleneck with its own custom topology (Google's OCS: Apollo/Palomar), but xAI, by my read, has not yet reached that stage.
Layer Blackwell's (GB200) "power smoothing" issue on top, and the picture comes into focus. According to Zeeshan Patel, formerly in charge of multimodal pre-training at xAI, Blackwell GPUs draw power so aggressively that the chip itself includes a hardware feature for smoothing power delivery. xAI's existing software stack, however, was optimized for Hopper and does not understand the characteristics of the new hardware; when it imposes irregular loads on the chip, the silicon physically destructs â literally melts. That means the modeling stack must be rewritten from scratch, which in turn means scaling is far harder than most of us imagine.
Pulling all of this together points to a single conclusion. xAI judged that training frontier models on Colossus 1 simply was not efficient enough to be worthwhile. It therefore moved its own training workloads wholesale onto Colossus 2, built as a 100% Blackwell homogeneous cluster. Colossus 1, on the other hand â whose mixed architecture is far less crippling for inference, which parallelizes more forgivingly â was leased in its entirety to an Anthropic that desperately needed inference capacity.
Many observers point to what looks like a contradiction: Elon Musk poured enormous capital into building Colossus, only to hand the core asset over to a direct competitor in Anthropic. Others read it as xAI capitulating because it is a "middling frontier lab." But these are surface-level reads.
Look at the numbers and a different picture emerges. xAI today holds roughly 550,000+ GPUs in total (on an H100-equivalent performance basis), and Colossus 1 (220,000 units) accounts for only about 40% of the total available capacity. Colossus 2 â built entirely on Blackwell â is already operational and continuing to expand. Elon kept the all-Blackwell homogeneous cluster (Colossus 2) for himself and leased out the older, mixed-generation Colossus 1. In other words, he handed the pain of rewriting the stack â the MFU-11% debacle â to Anthropic, while keeping his own focus on training the next generation of models.
The real point, then, is this. Elon's objective appears to be positioning ahead of the SpaceXAI IPO at a $1.75 trillion valuation, currently floated for as early as June. The narrative SpaceXAI now needs is that xAI â long the "sore finger" â is not merely a research lab burning cash, but a business with a "neo-cloud" model in the mold of AWS, capable of leasing surplus assets at high yields.
From a cost-of-capital perspective, an "AGI cash incinerator" is far less attractive to investors than a "data-center landlord generating cash."
As noted above, the most important detail of the Colossus 1 lease is that it is for inference, not training. Unlike training, inference requires far less tightly synchronized inter-GPU communication. Even when the chips are heterogeneous, the workload parcels out cleanly across them in parallel. The straggler effect â the chief weakness of a mixed cluster â is essentially neutralized for inference workloads.
Furthermore, with Anthropic occupying all 220,000 GPUs as a single tenant, the network-switch jitter (unanticipated latency) that arises under multi-tenancy disappears. The two sides' technical weaknesses end up complementing each other almost exactly.
One insight follows. As a training cluster mixing H100/H200/GB200, Colossus 1 was an asset that could only deliver an MFU of 11%. The moment it was handed over to a single inference customer, however, that asset transformed into a cash-flow asset rented out at roughly $2.60 per GPU-hour (a weighted average of the lease rates across GPU types). For xAI, what was a "cluster from hell" for training has become a "golden goose" minting $5â6 billion in annual revenue when redeployed for inference. Elon's genius, I would argue, lies not in the model but in this asset-rotation structure.
The weight of that $6 billion becomes clearer when set against xAI's income statement. Annualizing xAI's 1Q26 net loss yields roughly $6 billion in losses per year. The $5â6 billion in annual revenue generated by leasing Colossus 1 to Anthropic, in other words, almost perfectly hedges xAI's loss figure. This single deal effectively pulls xAI to break-even.
Heading into the SpaceXAI IPO, this functions as a core line of financial defense. From a cost-of-capital standpoint, if the image shifts from "research lab burning cash" to "infrastructure tollgate stably printing $6 billion a year," the entire tone of the offering can change.
(May 8, 2026, Mirae Asset Securities)
rant time: people are so fucking obsessed with building more tools, more products, more services, more "security" layers. are you guys all fucking insane?? every single thing you add is more complexity. and complexity is exactly what makes systems _dangerous_. you don't get safer by stacking abstractions on top of abstractions. you just increase the attack surface and pray the whole dependency chain doesn't collapse (hint: it will collapse!!). now you depend on 10, 50, 100 moving parts. all needing updates, all with their own bugs, all potential supply chain failures and we call that "security" like fucking retards.
dude, it's the fucking opposite. we're not building safer systems. we're building systems so complex nobody actually understands them anymore. and almost nobody is asking the obvious question: **what can we remove?** everyone wants to add. nobody wants to reduce. that's how you end up in a nightmare system (hint: we're already in that nightmare). not because of one big failure. but because of thousands of tiny dependencies you never should have had in the first place.
Software needs to be aggressively hardened more than ever
Formal verification or bust for most
Ironically, vibe coding is taking many companies in the opposite direction at the worst possible time