asciimike

Not sure why this is getting downvoted, most of the new facilities (e.g. Stargate) are running closed loop systems. PUE goes up a bit if you have to use chillers, but ultimately it's faster/easier to build a large facility in a closed loop fashion than get water (permitting, treatment, etc.). NVIDIA pushing 45C inlet temps is to allow facilities to run dry coolers instead of chillers and make this even easier! Two phase cooling (required above ~3kW TDP) would eliminate water use in the secondary loop (if not the primary, eventually).

One of our engineers built a full data center walkthrough tool on an AVP (complete with data pulled from the building CAD files, as well as our internal DC operations tools): https://www.linkedin.com/posts/fluidstack_our-latest-hackath...

I'm really excited to see AR take off in use cases like this, given how useful it would be to give all our technicians e.g. a legacy Google Glass (not the Intelligent Eyewear thing) and direct them to a rack, overlay information about a failed component, and walk them through the repair SOP.

Fluidstack | SWE, SRE, Infrastructure, Security, Product, Design | SFO, SEA, NYC, ATX (Onsite)| Full-time | fluidstack.io

Fluidstack is building supercomputers for hyperscalers and frontier labs. We design, construct, and operate gigawatt scale datacenter campuses and then deploy, validate, and operate the GPUs and networking equipment inside.

We're deploying a GW of compute in 2026 in partnership with Anthropic (https://www.anthropic.com/news/anthropic-invests-50-billion-...), and are continuing to scale out capacity in the US.

We're hiring product and engineering across the stack: everything from low level systems programming/firmware on XPUs, automating building controls and datacenter robotics, automating compute and network provisioning and recovery from failures, securing our networks and facilities, and software across dozens of verticals (warehousing, logistics, site selection, commercial operations, talent, legal, etc.)

View hundreds of open roles at fluidstack.io/jobs or send an email to careers at fluidstack dot io

Fluidstack | SWE, SRE, Networking, Infra, TPM, PM, and many more | SFO, SEA, NYC, ATX, LON | Full-Time

Fluidstack is a full stack AI infrastructure company building datacenters, deploying GPUs, and operating hundreds of thousands of accelerators for top tier AI labs. We're also partnering closely with Anthropic (https://www.anthropic.com/news/anthropic-invests-50-billion-...) on deploying GW scale clusters.

More info at https://careers.fluidstack.io/jobs or ping careers at fluidstack dot io!

Fluidstack | SWE, SRE, Networking, Infra, TPM, PM, and many more | SF, NYC, ATX, LON | Full-Time

Fluidstack is a full stack AI infrastructure company building datacenters, deploying GPUs, and operating hundreds of thousands of accelerators for top tier AI labs. We're also partnering closely with Anthropic (https://www.anthropic.com/news/anthropic-invests-50-billion-...) on deploying GW scale clusters.

More info at https://careers.fluidstack.io/jobs or ping careers at fluidstack dot io!

Same card worked via Apple wallet for other transactions, and the Visa thing is a known issue historically (https://discussions.apple.com/thread/255018680?sortBy=rank). A local friend confirmed the fix, but I can't seem to find any info from Apple or Visa about it.

I'll bring up a cool additional option in this space: https://gaggimate.eu/

Less fully featured, but more open (Gaggiuino became closed source post people profiting off the OSS).

I'll also pitch another fun OSS project I sponsored a few years back to build an OSS pressure sensor and 0.1g scale: https://github.com/rhit-coultabm/Consistent-Cups-Espresso-Ac...

Since then it looks like Decent has OSS'ed the DecentScale: https://github.com/decentespresso/openscale

Hilariously, the physical cards can only be topped up at add fare machines (cash only) or ATMs (cash only).

If you go down the "add a pasmo/suica on your phone" then you get into licensing issues with the other cabal (credit card issuers): you need to use an Amex to charge it via Apple Pay (IIRC Visa is now supported, but when I tried it last week my card was declined, so... YMMV).

Fluidstack | Forward Deployed Engineer, SWE, SRE, Infrastructure, Networking, Devrel | London/SF/NYC or remote | fluidstack.io Fluidstack builds and operates GPU supercomputers for top AI labs, governments, and enterprises. Our customers include Mistral, Poolside, Black Forest Labs, Meta, and more. Engineers at Fluidstack work on a wide variety of activities, including:

- Deploying clusters of 1,000+ GPUs and operating them with and for customers.

- Validating correctness and performance of underlying compute, storage, and networking infrastructure, and working with providers to optimize these subsystems.

- Migrating petabytes of data from public cloud platforms to local storage, as quickly and cost effectively as possible.

- Debugging issues anywhere in the stack, from “this server’s fan is blocked by a plastic bag” to “optimizing S3 dataloaders from buckets in different regions”.

- Building internal tooling to decrease deployment time and increase cluster reliability, including automation where the customer benefits clearly outweigh the implementation overhead.

We don't require prior experience with GPUs or AI/ML; we require a willingness to learn quickly, work hard, and obsess about customer success.

Apply here: https://jobs.ashbyhq.com/fluidstack or email careers at fluidstack dot io

Drive by comment, I saw: https://unusualwhales.com/news/the-race-to-move-gold-bars-ou...

> The sudden shift is fueled by fears of tariffs as Donald Trump threatens a trade war that could disrupt the gold market. Traders, worried that gold imports may soon face tariffs, are rushing to get their bullion into the U.S. before the policy takes effect.

Fluidstack | Forward Deployed Engineer, SWE, SRE, Infrastructure, Networking, Devrel | London/SF/NYC or remote | fluidstack.io Fluidstack builds and operates GPU supercomputers for top AI labs, governments, and enterprises. Our customers include Mistral, Poolside, Black Forest Labs, Meta, and more.

Engineers at Fluidstack work on a wide variety of activities, including:

- Deploying clusters of 1,000+ GPUs and operating them with and for customers.

- Validating correctness and performance of underlying compute, storage, and networking infrastructure, and working with providers to optimize these subsystems.

- Migrating petabytes of data from public cloud platforms to local storage, as quickly and cost effectively as possible.

- Debugging issues anywhere in the stack, from “this server’s fan is blocked by a plastic bag” to “optimizing S3 dataloaders from buckets in different regions”.

- Building internal tooling to decrease deployment time and increase cluster reliability, including automation where the customer benefits clearly outweigh the implementation overhead.

We don't require prior experience with GPUs or AI/ML; we require a willingness to learn quickly, work hard, and obsess about customer success.

Apply here: https://jobs.ashbyhq.com/fluidstack or email careers at fluidstack dot io

Fluidstack | Forward Deployed Engineer, SWE, SRE, Infrastructure, Networking | London/SF/NYC or remote | fluidstack.io

Fluidstack builds and operates GPU supercomputers for top AI labs, governments, and enterprises. Our customers include Mistral, Poolside, Black Forest Labs, Meta, and more.

Engineers at Fluidstack work on a wide variety of activities, including:

- Deploying clusters of 1,000+ GPUs and operating them with and for customers.

- Validating correctness and performance of underlying compute, storage, and networking infrastructure, and working with providers to optimize these subsystems.

- Migrating petabytes of data from public cloud platforms to local storage, as quickly and cost effectively as possible.

- Debugging issues anywhere in the stack, from “this server’s fan is blocked by a plastic bag” to “optimizing S3 dataloaders from buckets in different regions”.

- Building internal tooling to decrease deployment time and increase cluster reliability, including automation where the customer benefits clearly outweigh the implementation overhead.

We don't require prior experience with GPUs or AI/ML; we require a willingness to learn quickly, work hard, and obsess about customer success.

Apply here: https://jobs.ashbyhq.com/fluidstack/fb45b5dc-614e-40dd-ac5d-... or reach out to mike at fluidstack dot io

Two things:

1. I believe they require that you have an acquired specific landing page (e.g. crusoe.ai/acquired), which we saw directly as conversions through to our waitlist. The volume was lower than other channels, but the signal was much, much higher.

2. To @scarface_74's comment, "you talk to customer and you can ask them where they heard about you from" is mostly why I made the claim. In particular, when fundraising, basically everyone I talked to said, "I heard about you on Acquired, really interesting business model that I hadn't considered, let's talk more about it."

Sponsorship isn't cheap (I would go so far as to call it expensive), but relative to spending an equivalent amount on search or display ads/billboards on 101/etc., I think it was the right choice at the time.

My only analogy for this is what I call "cruise missile marketing" where you're investing a lot of time/money in building something that is very specifically targeted at high value buyers. It works really well for large, infrequent transactions (raising capital, selling GPU clusters, etc.) and less well for commodity SaaS or B2C products where volume >> everything.

Having recently listened to TSMC as well as Costco, LVMH, and Hermes, there's something available to every member of the HN audience in every episode, no matter how "unrelated" you may think it is.

On the other side of the table, as a former sponsor of a season or two: it was the best marketing we (Crusoe) could have done. We had more high quality inbound from our audience than any other channel, by a long shot. They deeply understand their audience and target appropriately. Strongly recommend to anyone trying to reach a decision making audience (customer and investor).

Not sure why it's getting downvoted, definitely experienced Code Yellows, Code Reds and Code Purples at Google. Red is (obviously) worse than Yellow and IIRC was a total code freeze for a period of time. IIRC there was a Code Red around memory at some point because the supply chain was literally so backed up that google couldn't get enough DIMMs and reasonably sized services had to stop deploying because there wasn't enough compute capacity.

Purple is/was "developer experience is so bad we need to stop developing new functionality and make the current functionality usable."

> Can confirm: there are about two dozen bottom bracket tools in the drawer. In fairness, bottom brackets have been a pain in the ass for decades. Even on old ones, there are a couple different hook spanners and pin spanners you might need for the lock ring and adjustable cup and a couple other weird-ass wrenches that you need from time to time. Shit's usually tight AF too, and the various tools that were fine for manufacturing a bike get a little iffy when everything's good and seized after 20 years of neglect.

I agree the various (totally random) BB standards are a pain; every bike build I've done has meant that I CADed and then 3D printed the tool (100% infill PETG, takes ~2 hours on a Prusa MK3S).

Curious how long it is before we get 3D printing tech easy enough to where shops can have a printer, download any special tools for a bike assembly, and make them in a few hours.

Hope is a great example of a manufacturer who does this today: https://www.hopetech.com/open-source-tools/

No, it's still illegal

Location: Austin, TX

Remote: Yes, open to hybrid or in person

Willing to relocate: Yes (SF, Seattle, Denver/Boulder)

Technologies: Product leader, platform builder, API enthusiast, AI, Cloud, IaaS/PaaS

Résumé/CV: linkedin.com/in/asciimike

Email: hn at mikemcdonald dot co

Experienced product manager looking for a group PM/director role in AI infrastructure, platforms, or tooling. I build platforms that developers love, historically at Firebase, Google, GitHub, and most recently, Crusoe, where I built a GPU IaaS platform that powers many quickly growing AI startups and an increasing number of enterprises.

Possibly a dumb question, but why can't we grow synthetic quartz in the same way we grow silicon wafers? Or can we and it's just not cost effective vs mining?

Cunningham's Law: "the best way to get the right answer on the Internet is not to ask a question, it's to post the wrong answer"

Where/how did you learn welding? Have been considering doing a community college welding course (as they have all the equipment, instruction, etc.).

I'm scared to ask, but how big is said CPU (and how many kW is it putting out)?

I'd love to bring a Japanese beverage vending machine and drop it off in the middle of a reasonably populated US city and see how quickly it gets used (or, unfortunately, destroyed). Would only accept Yen (and ideally Suica/Pasmo, though unsure if those would work properly?).

Where do you source vintage audio gear from?

Oof, the short answer is Windows + virtualization + GPUs is hard for anyone to support. Two main difficult problems:

- Licensing Windows is really, really annoying (and expensive), and BYOL is something people seem oddly reticent to do - Installing (and updating) NVIDIA GPU drivers on Windows is something that requires GUI access (at least the first time)

Paperspace was going to be my answer, but I guess DO didn't like those problems either! NVIDIA has an RTX Cloud, though I admit I'm struggling to find mention of it on their website, maybe something like: https://www.nvidia.com/en-us/data-center/free-trial-virtual-...

Mind sharing the resource material? I've looked into this briefly, seems like a common-ish path for flat bed trucks is using a surplus S-280 (e.g. https://www.ramims.com/products/S-280C-G).

I really like both:

- https://www.dirttrailswanted.com/overland-peanut (S-280 route)

- https://wabisabioverland.com/ (custom build, both in house and contracted through: https://totalcomposites.com/)

Two things:

- Practicing CAD and 3D printing: building a go pro mount on my bike computer mount for a light (https://twitter.com/asciimike/status/1836892842716750221), an oil filter wrench (https://twitter.com/asciimike/status/1840051449625018870), etc. I like the idea of needing a specialty car/bike/etc. tool and being able to quickly model and build it in hours vs having to order an often fairly expensive tool that takes days to show up.

- Import/export (specifically from Japan to the US): In particular, importing kei vehicles >25 years old as well as specialty roasted coffee. Both of which have a lot of specific regulations and intricacy, and while I've done a lot of reading on how to do them (e.g. this doc on importing a kei truck: https://wittymelon.wordpress.com/portfolio/diy-how-to-import...). If anyone has connections to people who have imported kei vehicles to the US, I'd love to chat!

The short answer here is that NVIDIA doesn't like Cloud Service Partners using RTX cards, as they are "professional" cards (they are also significantly cheaper than the corresponding data center cards). IIRC, A40, L40, and L40S have ray tracing, and might be more available on CSPs. Otherwise, the GPU marketplaces that aren't "true" CSPs will likely have RTX cards.

Paperspace (now DO), Vultr, Coreweave, Crusoe, should all have something with ray tracing.

Awesome, thanks for confirmation!

To your point, I _very much do not_ want to stick my nose into complex financial tools, hence my desire to use a more simple tool to get the exact value of "give me the thing I generally want, minus the one or two things I don't." Sounds like this is one option for doing so!