Our weekly inference usage has surged 38.86% in just seven days.
Between October 12โ18 and October 19โ25, we recorded a substantial acceleration in inference demand across our infrastructure.
- 5.86 billion tokens processed, up from 4.22 billion
- +1.64 billion additional tokens processed week-on-week
- +38.86% growth in total inference usage
- 837 million average daily tokens, up from 603 million
- 25.1 billion tokens in projected 30-day usage, based on our latest weekly run rate
This acceleration is being driven in part by our latest enterprise deployments, which are bringing increasingly substantial workloads onto Dot's inference infrastructure, alongside significant demand for Dot Motion following its recent launch.
Together, these developments are expanding our inference footprint across both enterprise and consumer applications, demonstrating the growing demand for private, cost-efficient AI infrastructure as we continue to diversify the workloads and revenue opportunities supported by Dot.
Monitor our usage in real-time at: https://t.co/9z6WBI6mGv
Dot Desktop is now fully accessible through remote mobile access, bringing our private, agentic environment another step closer to the capabilities of frontier AI platforms.
Getting connected takes a few steps:
1. Open @BotFather on Telegram and send /newbot to create your own bot.
2. Paste the bot token into Dot Desktop โ Settings โ Telegram.
3. Scan the pairing QR code, tap Start, then confirm your Telegram account in Dot Desktop.
We're continuing to close the gap between Dot and the most advanced agentic environments, while bringing everything together within our own privacy-first ecosystem.
Dot Desktop, your private AI workspace, accessible wherever you are.
Download Dot Desktop at: https://t.co/eFFR0eZBzn
Iโm still firmly in the camp that AI will recover strongly
Anything truly revolutionary has needed its wrinkles ironed out along the way
The best days are still ahead of us
My top pick is base:0x23a2847d772803f9efc64b4277b782b06296fe51 @usedotai
We are actively increasing the agentic capabilities of the Dot platform.
As agents begin to dominate, we will ensure the Dot platform is capable of sustaining agentic demand.
Weโve increased the agentic capabilities of the Dot platform.
Give your private AI a task, define when and how often it should run, and let it handle the research, monitoring and execution on your schedule.
From daily news briefings and model release tracking to privacy monitoring, price alerts and recurring research, Dot works in the background and saves its findings for when you return.
First look below โ
As we expand the Dot GPU Marketplace, we're looking beyond hardware specifications and pricing to optimise real-world inference economics.
This case study present exactly why we're investing in inference optimisation alongside the expansion of our GPU marketplace, and how this makes our GPU marketplace one of the most attractive opportunities for those with compute needs:
On H200 serving Qwen2.5-32B-Instruct, the quantized GEMM execution path materially changes performance, even with identical weights.
AWQ + Marlin: 741.04 output tokens/s.
AWQ without Marlin: 67.73 output tokens/s.
Same quantized checkpoint, same GPU. 10.94ร throughput differential.
GPTQ shows the same pattern: 712.45 tokens/s with Marlin, versus 276.60 tokens/s without it; a 2.58ร improvement.
FP16 baseline: 461.04 tokens/s. AWQ Marlin delivers 1.61ร baseline throughput, despite reducing weight precision by 4ร.
At $6.09576/hour using Dot's GPU the estimated compute cost per million output tokens:
Cost per 1M output tokens =
(hourly rate ร 10โถ) รท (3,600 ร output tokens/s)
- AWQ Marlin: $2.28
- Non-Marlin AWQ: $25.00
Quality: HumanEval pass@1: 56.1% FP16, 51.8% AWQ Marlin, 51.8% GGUF Q4_K_M, 45.7% GPTQ Marlin. AWQ Marlin achieves the highest throughput while tying GGUFโs code-generation score in this test.
By evaluating kernels, quantisation methods and serving configurations, we can extract substantially more performance from the same physical hardware, reducing the effective cost of compute for users while improving infrastructure utilisation.
For our marketplace, this means we're not just offering GPU capacity at competitive hourly rates, but working to make that capacity deliver more tokens, more workloads and more value per dollar.
Better utilisation, higher throughput, lower inference costs, only with the Dot GPU marketplace: https://t.co/TnBUZEMSaQ
44K revenue in October alone.
Usage keeps growing month over month, and itโs all running on $200K worth of owned hardware.
$DOT sits at a $5M market cap. With these numbers it should be at $40M to $50M by now. If youโre still on the sidelines, youโre missing it.
Iโve been in crypto a long time and Iโve never had this much conviction in a project. $DOT
A punchier alternative:
October revenue: $44K
Usage: growing every single month
Infra: $200K in owned hardware
Market cap: $5M
The math doesnโt add up. This should be at $40M to $50M.
Never had this much conviction in a project. $DOT
I like stats.
I like that base:0x23a2847d772803f9efc64b4277b782b06296fe51 is publishing stats. Stats say grow ๐
Markets are down everywhere
Buying growth in private AI platform while cheap.
How can you not love this team????
@usedotai just announced that base:0x23a2847d772803f9efc64b4277b782b06296fe51 Desktop, which was released less than a week and a half ago (and already has 300+ downloads) will now have full AI video and motion design capabilities.
DOT private AI is growing across all surfaces, and is increasing user requested features and functionality all the while.
Try is here: https://t.co/nDUAD8tJgc