The current natural gas price in West Texas (the WaHa hub, north of Coyanosa, TX) is negative. And has been for a while. The price peaked (dipped?) to -$9/MCF a couple months ago. That means gas producers had to pay $9 per MCF for it to be taken away. Oil in the Permian comes with gas, a lot of it, so to produce oil, you need to get rid of gas. Wells I'm familiar with have 4000-5000 cubic feet of gas per barrel of oil. Recall in oilfield M = thousand, so that's 4-5 MCF per bbl of oil.
There is no free gas pipeline capacity to get gas out of West Texas. Any time new pipelines are built, they are filled within months.
This makes a ton of sense for oil producers (which are also gas producers) who can sell their gas for less of a loss (potentially a profit!) and also for MSFT who can lock in long term contracts for minimal cost. I'd guess these contracts are for $1-2/MCF which is win/win for the oil companies in the area and MSFT.
This might make sense for oil producers to get rid of their natural gas, which is nearly a waste material, but I'm struggling to understand how it makes sense for Microsoft. Despite the cheap natural gas, and an abundance of investors and builders with natural gas expertise, in the competitive electricity generation market everybody is deploying solar and batteries in west Texas because it's still more profitable than gas generation.
Further, there's a gas turbine shortage so Microsoft choosing to put their (presumably limited) allocation of gas turbines in West Texas, where they have good alternatives, seems a bit mysterious. Why not save that massive amount of turbines for the northeast DCs, where renewables work far poorer yet gas is more reliable?
The reasons that seem most convincing to me:
1. Political environment is hostile to renewables and Microsoft doesn't want to paint a target on their back by choosing solar plus batteries, the choice others are making in West Texas.
2. Grid connection drastically changes economics, but pipelines for gas are cheap or something, so the massive cost and delay from grid interconnection simply isn't worth it
3. There are particular political favors going on with Chevron, e.g. Chevron wants gas in the area and is willing to increase MS's turbine allocation if they do it in west Texas, or Chevron is helping get around pesky local political approvals for data centers, or something like that.
The cost of gas does not seem like a justification for this, though.
They will also want reliable power, which seems like justification enough to keep their options open. A “phased, modular approach” sounds like it would give them options. Maybe Microsoft could install solar and/or batteries later and use their natural gas turbines for backup?
With enough batteries they might get through the night, but seasonal shortages are much harder to handle that way.
I bet a carbon tax on data centers would be popular if the Democrats get back in.
Natural gas is cleaner to burn than coal, if I had to guess this is lazy half-done greenwashing, i.e. 'at least it's not coal!' Similar to California's measures for cleaner power generation.
There is also probably a balance of payments aspect of this. Chevron uses a ton of Azure. They probably asked for Microsoft to use some amount of Chevron in exchange. This is probably that deal.
The cost of gas is completely irrelevant. In order to bring a new large load onto the grid, you have to coordinate with ERCOT, and they just made that process more tedious. Once you get your grid connection approved and built out, you have to source your own power anyway, and frankly there just isn't enough power on the market to realize the stated datacenter buildout goals in this country.
In short, you're going to have to build your own power plant anyway, so why bother with the grid? Gas is the cheapest, fastest zero-to-production choice for onsite power generation, and has been for a long time. Unless you're dealing with nuclear, the fuel cost just doesn't matter compared to the rest of the buildout, and gas wins because you can take off-the-shelf turbines and bolt them down.
You can only get away with it in places that don't care about environmental regulations, which are the places most likely to approve new buildout of gas infrastructure anyway. Nobody in the northeast is going to approve the creation of a brand new carcinogen factory.
> Gas is the cheapest, fastest zero-to-production choice for onsite power generation, and has been for a long time.
Is it, though? There's a ton of projects out there waiting to get their solar on the grid in west Texas, partnering with one of them and launching the project early while waiting on the interconnection queue, and adding enough batteries, gives a more robust solution right now without the SPOF nature of large single generators.
Add to that the long backorder list for gas turbines right now, with no end in sight, and I'm surprised that Microsoft would power this particular location with turbines because it's probably their best chance to do off-grid massive solar projects.
Massive off-grid solar is what China is choosing for some absolutely massive new industrial projects. Nuclear is a no-go because it takes so long to deploy, but solar + batteries are cheap COTS and available in abundance, unlike gas turbines.
> Is it, though? There's a ton of projects out there waiting to get their solar on the grid in west Texas,
gas turbines run at night too so there's no storage/backup supply issue to consider. They also take up significantly less space than wind or solar and those data centers are already gigantic.
There is no point in waiting for interconnection when you can just... not do that, and do all your generation behind the meter, with complete control of the generation to match your load. Solar wants an interconnect so they can sell off surplus; with gas you just turn the dial down to meet the load and walk away.
The clock is running on the datacenter goldrush. 70-90% of the capex window is going to be soaked up just with construction time. Introducing a capricious ERCOT permit process and shopping around for friendly solar projects to hop in bed with makes no sense when you can just write a check and solve the problem forever.
I'd bet the deal with Chevron was to enable Microsoft to hop the queue here and get those GEV turbines soonest.
I'm not saying wait for an interconnection, I'm saying take a project that already has a fully developed plan, that's waiting for interconnection, and build the DC right there next to it, beef up the battery component of the plan and go to town. Interconnection happens some time in the future, giving far more financial opportunities for everyone, but in the interim the electricity goes fully to the DC.
Because I expect the economics of enough batteries to carry a data center’s worth of demand through the night and morning is too expensive compared to essentially free natural gas.
Data centers would ideally run at ~100% utilization, so any drop in solar output needs to be fully met by batteries.
Good luck. The renewables overbuild required to firm baseload demand for a datacenter is insane. You need something like 4x solar and 1.5x storage in a 10hr battery.
Respectfully, the laws of NPV and IRR hurdles don't matter in Chinese infrastructure.
Whatever you choose in the US, it's not cheap, and developers crawl over each other to sign up hyperscalers. Race to the bottom.
Cheap gas is great, but 2.67GW of new build natural gas in this market will cost $6-8 billion in fixed costs. You need wholesale pricing of ~$50-60/MWh ... OVER 25 years! ... to recover just the fixed costs.
For West Texas, prices averaged mid-$30s over the last year.
Microsoft has all of the leverage here, and Chevron wants a big announcement in an area where they don't have a lot of experience.
I don't understand why datacenters especially shouldn't be able to run mostly on renewables.
This won't apply to every datacenter, but the AI inference ones especially, should be seeing most demand during the day. So what's built in north America is used when it's daylight there?
If so, isn't that a perfect case for solar?
To be clear, I'm not saying it can power down, but at night it should be able to scale down significantly?
Wow, not one mention of the env vars that have a far greater influence on how the models actually work under the hood - https://code.claude.com/docs/en/env-vars
Very important for bedrock deployments and other not-as-standard deployments
Key for how I've deployed it - disable adaptive thinking, max thinking tokens, disable telemetry, etc
GPT-5.5 is a solid leap with Codex or other harnesses. Opus 4.7 I still don't understand how people use... I tried it for a day or two, have tried it for a few hours every week or so since release, and still use 4.6 as daily driver (with xhi thinking).
As with these daily opinion threads, ymmv. I find GPT's code to be competent, but its voice isn't great. If Claude can be a little too cool, GPT-5.x often reads like 90s era movie hacker technobabble. This has got to be RLHF/alignment and the sort of tone that people like. Also anecdotally I used xhigh for a while and turned it down to medium because it would take so long to do even simple jobs. The instruction following is quite good with 5.5 so there isn't too much need to let it wander off.
It could be an overflow but related with the frequency at which the register was increasing, rather than the max value of te register. E.g. +1 this uint16 (65535) once every 500,000 cycles on this 32 Mhz chip, that previously was a 1 Mhz chip and never had a problem.
apparently you can straight up duplicate/add/rearrange layers without changing any of the weights and get better results as well - https://dnhkng.github.io/posts/rys/
> This is probably due to the way larger numbers are tokenised, as big numbers can be split up into arbitrary forms. Take the integer 123456789. A BPE tokenizer (e.g., GPT-style) might split it like: ‘123’ ‘456’ ‘789’ or: ‘12’ ‘345’ ‘67’ ‘89’
xVal basically says "tokenizing numbers is hard: what if instead of outputting tokens that combine to represent numbers, we just output the numbers themselves, right there in the output embedding?"
It works! Imagine you're discussing math with someone. Instead of saying "x is twenty five, which is large" in words, you'd say "x is", then switch to making a whistling noise in which the pitch of your whistle, in its position within your output frequency range, communicated the concept of 25.00 +/- epsilon. Then you'd resume speech and say "which is large".
I think the sentiment is that today's models are big and well-trained enough that receiving and delivering quantities as tokens representing numbers doesn't hurt capabilities much, but I'm still fascinated by xVal's much more elegant approach.
There have been a couple "studies" and comparing various frontier-tier AIs that have led to the conclusion that Chinese models are somewhere around 7-9 months behind US models. Other comment says that Opus will be at 5.2 by the time Qwen matches Opus 4.5. It's accurate, and there is some data to show by how much.
Personally was always a fan of just going with the largest fans possible - surprised we don't see more cases designed around 140mm and larger. 200mm is much less common but has a more pleasing noise profile
I'm also a fan of that sort of setup. A Fractal Meshify 2 XL will fit a bunch of 140mm fans, or you can get the Torrent which is smaller but has 2x 180mm fans up front. I have both and would recommend them, though the Torrent is a tight fit for a big board, and the shield on the back of the Asus W790 motherboards interferes with the cable routing grommets on the motherboard tray, so you have to remove them.
Noctua makes really good fans, I'm told. Want to get on their level and make a similar amount of money? In a world of slop, quality engineering is valuable.
Using this with tmux and various VPN tech. Main issue is scrolling. Termius + tmux don't scroll very well. And I've been led to believe tmux is necessary to keep sessions open when I turn off my phone screen
Scrolling is quite jenky with Termius - I thought there's a way to keep sessions going when there are intermittent drops in connection via Termius, but for how I've been building, when I lose connection I just restart claude and reexplain the context of the task.
I’m the Blink developer and really curious. Which way do you think Blink is inferior? I think the ssh/mosh tooling is way more powerful, keyboard config, etc. But would love to improve what I can. Currently working on new UI and better access to hosts, so it is a good time.
Sorry to have missed this. First, props that Blink is indeed superior in that it actually works.
My recollection is that the primary way Termius "feels better" is the pre-terminal UI (configuring hosts, etc).
But I'm still a daily Blink user, so I don't 100% recall what Termius did to make me think it was nicer, I just remember that scrolling was fucked so I gave up on it.
Thanks! How I eventually found it was stripping stuff back layer by layer. And by that I mean I started with the raw camera feed and got to where things worked well in a different swift view. And then from there, peeled stuff back from the main process feature by feature. And then bam, aircraft were exactly where they should be (minus the compass inaccuracy). I even had stuff like drawing mountain peaks (I live near Denver) as "aircraft" to figure things out, determining different FOV at different zoom levels (a lot of AI keyed in where the boxes would move in one direction at low zooms, be completely correct at some middle zoom, and then in the opposite direction at high zoom).
And that peeling back was me looking at each function to see what it did (I am a dev, but not for SwiftUI). So yep, can't vibe code it all!
There is no free gas pipeline capacity to get gas out of West Texas. Any time new pipelines are built, they are filled within months.
This makes a ton of sense for oil producers (which are also gas producers) who can sell their gas for less of a loss (potentially a profit!) and also for MSFT who can lock in long term contracts for minimal cost. I'd guess these contracts are for $1-2/MCF which is win/win for the oil companies in the area and MSFT.
reply