Toby Pohlen

@jukan05 Whoever wrote this has a lot of half-knowledge

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 — 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 — or for any GPU that has hit a stack-related snag — 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 1,000–10,000 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 — 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 — whose mixed architecture is far less crippling for inference, which parallelizes more forgivingly — 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 — built entirely on Blackwell — 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 — the MFU-11% debacle — 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 — long the "sore finger" — 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 — the chief weakness of a mixed cluster — 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 $5–6 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 $5–6 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)

@fraser_cook @xai Thanks for all the amazing work 🐐

Today marks the end of my time at @xai I was incredibly fortunate to have joined at a time where I was able to help take us from nothing to something across a few surfaces. It has been a privilege to work alongside some of the world’s best for almost two years and share in the ups, downs and everything in-between. I wish all of them well, past and present, and want to thank them for what has been a very special period of my life!

@finbarrtimbers @xai Will do! I have a ton of notes actually.

@TobyPhln @xai Please write more about this! I’d love to read more about you reflecting on your Xai decisions.

I'm back in Austin three years after spending my first day at @xai here. On the flight I thought about all the decisions I made during that time - many of which were actually poor. I compiled a selection below.

@stalmico @xai Not splitting my time better between London and PA

@TobyPhln @xai curious which one stings most

@tugot17 I've been living here for a long time and love the city. It's also great for building companies.

@TobyPhln why London?

The most impactful decision I made was to build a team in London instead of moving to Palo Alto. It opened up a great talent pool for the company but also meant I couldn't effectively lead a big area. I probably should have at least split my time evenly between the locations.

I avoid company politics; it's painful. But as a lead you can't ignore bad decisions and submit PRs instead. I wish I had been more vocal about areas such as prod reliability, security, and feature roadmaps.

@giffmana @unixpickle 😭

Hey now, this took a ton of effort. Design doc, cross org alignment, getting stakeholder buy in, lawyer approval, security review, leads sponsorship, and finally getting an intern to write the code, iterating until it fits the readability style guide in all 7 languages that were needed for this, fighting the 4 rollbacks it received, and finally writing the launch doc thanking everyone who ever sat in a related meeting or left any comment on the doc. This is important groundwork to eventually making it do useful things!

did google seriously add an annoying Gemini button to the bottom of all google doc pages?

@SimonCarGuy Remember this?

U.K. proudly produces some of the most desirable cars in the world including Rolls Royce, Bentley, Aston Martin, Range Rover etc Our industry (which employs nearly 900k people) needs government to get behind it. But we send Their Majesties to the White House in a BMW 7 Series?

@MarcoJunk 💯

Lieferkettengesetz für die Wirtschaft bis zur letzten Schraube aus Indien vs. der Aufwand zu klären, welche NGO wieviel Steuergelder erhalten hat, ist uns zu groß, in ein und demselben Land, ist unerträglich.

@skcd42 Sounds like Louis

> I've watched sandeep grow from a security hellhole, to someone I only have occasional day nightmares about. I'll add him to the privileged list and we can revoke if he acts up. 🧘‍♂️

@yacineMTB It was great meeting you. Sorry for giving you such a hard time in the interview, lol

Can't believe I only joined this website a year ago. A month after I joined, roon followed me when i was at 800 followers and then everything changed. I left my comfortable big tech job, started a successful startup called dingboard, got hired at twitter, had a kid, got fired from twitter, became an xai shareholder, then became a spacex shareholder, became an electrical engineer, and gained 250k followers. Life really comes at you fast

@mrcgngr611 @BigTechAlert @jmusk Fixed, lol

@BigTechAlert @jmusk @TobyPhln We know each other for more than 18 years… He‘s still not following me… @TobyPhln 🥲

🚫 @jmusk is no longer following @TobyPhln

@BigTechAlert @jmusk bro, wth. We watched ducks march through a hotel lobby together. This is the kind of experience that creates life-long bonds

British humor

@DavidHorne @LNER That's great

We now have high definition cameras on nine @LNER trains monitoring the infrastructure. Defects are identified and alerts sent to engineers who can fix the problem before services are disrupted. We have identified and dealt with 19 overhead line defects on the East Coast route using this approach - avoiding an estimated 11 days of delays to customers. A good example of how the railway is using modern technology to improve everyone’s journeys - more of this to come with Great British Railways!

Exiting a platform is the most severe punishment in the eyes of these people. They simply haven't comprehended that quite literally no one cares if they're here. Their ideology, viewpoints, and voices don't matter. And that's a bitter pill to swallow.