Instant Inference

Lightning fast private inference. Home of the ORIGINAL Instant Inference subnet.

Visit us at

[Instant Inference @instantsubnet]

Miners and Validators can now join Instant!

Documentation is live and we are actively onboarding participants.

The future is fast... the future is Instant.

[Instant Inference @instantsubnet]

@ivanfioravanti Are you suggesting Apple might win the AI race?

[Instant Inference @instantsubnet]

@araseb_ I can generate a zero knowledge proof of myself

[Instant Inference @instantsubnet]

.com

[Uday👨‍💻 @uday_devops]

Which domain name would you choose for your project?

- .In
- .ai
- .io
- .app

[Instant Inference @instantsubnet]

Bittensor is the way

[Jessie Dong @jessiedong_]

inference companies are taking on a lot of risk right now

they're signing longer contracts for compute because they never want to have customers and run out of compute at the same time. it'd be a big problem if users are growing, but they have to pause the product

it's hard to get compute fairly right now, so this is all they can do. inference companies are in a particularly tough spot because they work on a pay-as-you-go model, where customers pay per token and can leave whenever. clearly, this is a problem when you buy gpus on fixed contracts, because you'll pay for them even if customers don't use them

inference companies are simply stuck with no other choice. changing the way compute is bought will also make inference better!

[Jessie Dong @jessiedong_]

inference companies are taking on a lot of risk right now

they're signing longer contracts for compute because they never want to have customers and run out of compute at the same time. it'd be a big problem if users are growing, but they have to pause the product

it's hard to get compute fairly right now, so this is all they can do. inference companies are in a particularly tough spot because they work on a pay-as-you-go model, where customers pay per token and can leave whenever. clearly, this is a problem when you buy gpus on fixed contracts, because you'll pay for them even if customers don't use them

inference companies are simply stuck with no other choice. changing the way compute is bought will also make inference better!

[Instant Inference @instantsubnet]

@JosephJacks_ This is one of the signs that LLMs have a ceiling. They can’t adapt on the job like humans can.

Fixing this problem would require a full AI stack pipeline and architectural redesign.

[Instant Inference @instantsubnet]

Let’s super charge this 💪

[Richard @RMladek]

🚨 THE ECONOMICS OF AN INFERENCE COMPANY 🚨

demand for tokens is exploding. exponentially.

an inference provider is loosely defined as a company that gets access to gpus, runs an inference stack on them and makes money selling tokens.

but price per token keeps falling and gpu prices keep rising. so how do the economics work?

this is a brand new company category. 0 ipos. nothing to analyze.

...except minimax and zhipu (Z ai). the chinese companies behind the minimax and glm model families. both listed in hong kong in january.

so we finally have audited numbers. one of them went from losing money on every token to a 24.6% margin in 18 months. the other lost 75% of its openrouter volume in ten weeks.

let's go through the financials in detail. starting with minimax 🧵⬇️

[Instant Inference @instantsubnet]

@0xVirtanen Absolutely! They need to also have the ability to LEAD and know how to keep the talent they have.

[Instant Inference @instantsubnet]

True! I wish more subnet investors knew this going in. Lots of subnets out there with non-technical leaders thinking they can vibe their way to success!

[Yash @YashHustle_22]

Hot Take:

Vibe coding doesn’t work at scale.

[Instant Inference @instantsubnet]

Never sleep

[Nitin @nitin_builds]

Before sleeping,

fix the bug.

[Instant Inference @instantsubnet]

Who is going to teach the AI the next evolution of programming?

When will it code natively in assembly or binary for us?

[Daniel @growing_daniel]

I wish I learned more about programming before it ended

[Instant Inference @instantsubnet]

Come join us on Discord! https://t.co/bGlvcLTOgv

[Instant Inference @instantsubnet]

@sebyrubino https://t.co/fERrtW0lcP

[Instant Inference @instantsubnet]

Instant is leaving SN46.

This probably will not surprise those following the Discord. I have been silent publicly for about a month. Behind it have been serious disagreements about Instant’s future, its identity, and the terms under which we can move forward.

We set out to build an inference company. I do not want its future to depend on repeated disputes over control of the very identity we are working to establish. I would rather address that uncertainty now than allow another year of engineering, investment and effort to make it harder to resolve. The same thing that has happened about a month ago is happening again.

Contractual confidentiality obligations limit the details I can share. Counsel is involved, and we are finalizing steps concerning our departure.

I have formally requested that SN46 stop using the Instant name and branding.

To everyone asking what happens next: we are continuing to build and evaluating where Instant can best deliver its vision. All code, and IP, including MVNIR, will be leaving SN46.

It is sad that it has come to this, I hope to work together with SN46 in the future.

Bittensor should be a place where builders can build. That is what we came here to do, and that is what we will keep doing.

Dan

Instant CEO

[SEBY @sebyrubino]

Wave 1: Neoclouds

Wave 2: Accelerated neoclouds (we are here)

Wave 3: Liquid, permissionless, and accelerated neoclouds (inevitable)

[Instant Inference @instantsubnet]

Everyone talks about serverless inference and large data center builds.

Few are talking about analog inference and its potential to bring frontier intelligence closer to the consumer, without the need for data centers.

The future is fast. The future is Instant.

[Anissa Gardizy @anissagardizy8]

SCOOP: Anthropic signed a $35B cloud deal to rent GPUs from Lambda—at a Texas data center Nvidia leased from Hut 8.

🤯 hard to keep that one straight

More details on the arrangement in my latest for @WSJ: https://t.co/jzaLZ5OiW0

[Instant Inference @instantsubnet]

@opendansor The longest lever!

⚡️����

[opendansor @opendansor]

Launch is imminent.

Bittensor has yet to fully unlock accelerated, specialized inference.

Instant is here to change that. From trading and video models to faster validation cycles across the network.

More inference. More loops. More innovation.

This is how Bittensor moves faster.

[Instant Inference @instantsubnet]

What wasn’t possible yesterday is possible today.

Accelerated inference changes what can be built.

Instant is almost here.

[fal @fal]

Introducing https://t.co/4xpxWRD3VV

A new platform for infinite, interactive AI livestreams. Pick a channel, prompt what happens next, and watch it generate in real time.

You aren't just watching the show. You're directing it.

[Instant Inference @instantsubnet]

@Sebtensor 🔥