Podflow AI by Virtuals (POD)

.@dphnAI $pod v2 is aimed to be released on monday night or tuesday morning

some things that can be expected to be improved in v2 is these

- Auto-update of workers (this is the final launch, no need to turn off / do migrations again)
- Support for NVFP4 on Blackwell gpus (5090
/ B200 / B300 / RTX 6K PRO)
- Higher utilization per GPU
- Improved routing / binary security
- Public API with credits live very soon after launch
- Support for 8x GPU machines

also good calculation from @jkrdoc here as to what we can expect once v2 is rolling 🫡
https://t.co/4xKuij56Yo

A few thoughts on $POD.

I've been trenching for ~2 weeks and can say it for sure that we have several clear meta that normies/retail are about to accept broadly. It is:

> Decentralized AI

We've already crossed the line where AI enjoyers moved back to private inference and been using frontier models while queries are encrypted.

But it's a temporary solution to this problem since frontier models are still limited and censored. Open-source models have solved this issue, however, most just don't have enough capacities to maintain the infrastructure. RAM + energy issues.

That's why we need DeAI. The idea itself is already evolving - numerous GPUs working together on processing prompts sent by users, not knowing the author of initial request (it's called 'private inference').

@dphnAI does exactly this. Distributed network of GPU suppliers are being paid in $POD for processing requests through interface + APIs.

What makes Dolphin different is flywheel tied to the token. The idea of decentralized + private

I don't know if this is a soft shill, but it's basically $POD @dphnAI

> uses peer-to-pool model instead of peer-to-peer to guarantee 100% uptime, something that other networks didn’t and can’t solve (they’re rentals, and gpu's get disconnected, so you cannot achieve massive scale).

> peer-to-pool only works with a solid verification layer that proves nodes are running the requested LLM you are paying for, which the team built as a key innovation (that most are unaware of) and it's likely the most important part of the stack.

The core part of it is called Live-Weight-Proofs or Proof-of-Weight [more info on screenshots].

I've seen less than 5 people talk about this, so if you want to properly understand the moat here, you need to focus on the peer-to-pool + verification stack.

> it has proven demand being the main uncensored model of @AskVenice and team backed by a16z, so yes this is a real company more than a crypto project.

Catalysts include V2 launch + API release + listing in @OpenRouter — which wil