illia

deploying gpus at

[illia @illiacpp]

@berniebiased @bradhhhhq Brad, your inbox is closed, please dm, would love to chat!

[illia @illiacpp]

@berniebiased @tdh_02 Tom, your inbox closed please dm, would love to chat!

[bernie @berniebiased]

You should own your intelligence.

Introducing Managed GPUs by American Compute - you own the GPU servers, we run them for you.

Most startups rent their compute. Everyone assumes buying and running their own GPUs costs more. It often doesn't.

Renting a B300 can cost up to $8 an hour. Owning it is more like $4 depending on the timeframe you're expecting to own them. And the hours you don't use, you can sell on the spot market, and make a profit.

The most common pushback is that only the Special Few Neoclouds can get chips and data center space with power. That's just not true. If you have the patience and the money ready, you can place an order, especially on the smaller side (e.g. 8 B300 nodes or 64 GPUs).

Our CTO @illiacpp has been actively sourcing GPUs and colo space for clients for the past couple of months. It's doable.

The real problem is it's messy. You're chasing hardware quotes and colocation quotes, and everybody lies about timelines.

So we handle the whole deployment, soup to nuts. And we do it for a flat fee, not a percentage of costs. (Percentages misalign interests, because the provider will want you to buy more expensive options).

What's the catch? How do we make money then? Well, we're a fintech company, not a procurement company. We make our money upselling you financing and insurance.

And we're not going to pretend we're the expert for every scale and every hardware configuration. If an established neocloud or another provider is the better fit, we'll tell you, and bring your project to them. We're friends with many great neoclouds and GPUaaS providers.

If you think compute is crucial to your startup, and will drive your success in the next 2-3 years; you should consider owning it.

Reach out for free consultations and more info at managedgpus [dot] com. Start owning your intelligence.

[illia @illiacpp]

@zjasper maybe they did sign a 3y contract when electricity just came out ahah

[illia @illiacpp]

@davidcasem @telnyx interesting, how do you fund it? each 64 nodes cluster would be like 35M. do you just have huge balance sheet or you somehow get financing without an offtaker?

[illia @illiacpp]

@thdxr @NotASithLord no one just buys gpus, deploys them and then like hey who wants them, am i wrong? 🤔

[illia @illiacpp]

so people expect someone to have unreserved online b300s just sitting there waiting for them to sign a contract? 🤔

[dax @thdxr]

this is how finding GPUs goes

1. someone says they have GPUs
2. turns out they are a broker for some other guy
3. that guy doesn't have the GPUs either, they are waiting for a buyer so they can place an order
5. they don't even have the space yet, that is owned by some other guy who's upgrading it

one of these guys is always named carl

[illia @illiacpp]

imagine you run a scheduled task to track your stocks and wake up to it hacking a fortune 500 company for pre-earnings data

[illia @illiacpp]

@tbpn @GavinSBaker a lot of the amateurs will leave the space as fast as they came in, but those who are in for the long run will get hardened by all the uncertainty and are gonna be absolute killers!

[illia @illiacpp]

wow, ai cyber risk is getting real, a friend who works at a b2b saas for casinos says they're now constantly under attack

[illia @illiacpp]

@rentbotsTX purchase yeah, but that warehouse probably can't pass a facility inspection

[illia @illiacpp]

biggest bottlenecks to more compute, ranked:

1. offtakers with good credit
2. cash for the 20-30% down payment
3. liquid-cooled space, or air-cooled space >2MW
4. allocation, gb300 and rubin are months out
5. experienced operators willing to do more deployments
6. time, aligning lender, offtaker, vendor and colo, then waiting on hardware and colo

[illia @illiacpp]

financing, lead times and hardware allocation under ~2MW vs real neocloud level are two separate worlds.

when someone says it's hard to get financing or there's a shortage of b300s, it's mostly about the big guys. smaller deployments still get done, maybe with a premium or slightly longer lead times

[illia @illiacpp]

how long will b300 continue to yield $40/hour, before it all gets competed away?

[illia @illiacpp]

@Decentralisedco take a look at american compute :)

[bernie @berniebiased]

Today, we’re introducing 𝗔𝗽𝗽𝗿𝗮𝗶𝘀𝗮𝗹𝘀 & 𝗗𝘂𝗲 𝗗𝗶𝗹𝗶𝗴𝗲𝗻𝗰𝗲 𝗯𝘆 𝗔𝗺𝗲𝗿𝗶𝗰𝗮𝗻 𝗖𝗼𝗺𝗽𝘂𝘁𝗲.

For many lenders and investors financing GPU projects or data center compute deals, this is a new asset class with risks they may not be familiar with.

1/

[illia @illiacpp]

@JayaGup10 deploying/securing more compute fast is one of the biggest differentiators there rn lol

[illia @illiacpp]

a lot of people want one or two vera rubins or gb300s, but:
- it requires liquid cooling, much smaller pool of data center space
- reinforced floors, same story
- half a year lead time, big bois are getting priority
- reliability, blackwells had a rough rollout

[illia @illiacpp]

B300 crunch got me thinking about deploying amd, but unfortunately the MI355X roi is just not there