MLX Network — node client in development
Every Apple device.
One supercomputer.
MLX pools idle compute from Macs, iPhones and iPads into a single Apple Silicon network. Connect a device, it works while you don't, and you earn MLX for the compute it delivers.
The node client is in active development. Join the waitlist to be first in line for the beta.
LLM inference batch
job_9f3ac1 · 212 devices · 18.4K TFLOPS·h
+1,182.4 MLXrunningEmbedding generation
job_8d12b7 · 486 devices · 9.6K TFLOPS·h
+617.1 MLXrunningModel fine-tune (LoRA)
job_7c90e2 · 64 devices · 58.2K TFLOPS·h
+3,740.9 MLXrunningScientific simulation
job_6f04a9 · 38 devices · 12.7K TFLOPS·h
+818.6 MLXrunning
Example data — live telemetry ships with the node client beta.
Network at a glance
Example data- Devices connected
- 12,847
- Jobs completed · 24h
- 3,912
- Compute delivered
- 2.41M TFLOPS·h
- Epochs completed
- 1,204
Deterministic example values for the pre-launch site. Real numbers arrive when the first node clients report in.
How it works
Your hardware, working in its downtime.
Connect
Sign in with Apple and pair any M-series Mac, iPhone or iPad. The node client installs in under a minute.
Contribute
While your device is idle and plugged in, the node client picks up workloads — inference, embeddings, rendering, simulation.
Earn
Every finished job credits MLX to your account, measured by the compute actually delivered. You stay in control of when your device works.
Devices
Built for Apple silicon.
The node client runs only on Apple silicon — because that's the bet. The most energy-efficient compute most people already own, unified under one scheduler.
Mac
Any M1 or later Mac joins the network at full throughput. Mac Studio and MacBook Pro models are the heavy lifters.
M1 · M2 · M3 · M4iPhone
A-series chips work overnight shifts — plugged in, charging, and idle. Tens of millions of devices, most of them asleep.
A17 Pro · A18iPad
M-series iPads are the quiet workhorses: always plugged in at home, almost never busy when you are not holding them.
M1 · M2 · M4Why MLX
Open-weight models need compute the data center can't sell.
The AI race is becoming a compute race. Frontier labs build gigawatt data centers; open-weight models run everywhere else. MLX is the everywhere else — a unified pool of Apple Silicon that any model, any researcher, any company can rent by the job.
Efficiency
Apple silicon does more work per watt than anything in the data center. Idle devices are the cheapest compute on earth — because the hardware is already paid for.
Privacy
Workloads run in sandboxes on your device. The network sees job results, not your files. You decide when your device works and when it rests.
Capacity
There are over two billion Apple devices in the world. Even a fraction of one percent of their idle hours is a supercomputer the data center cannot match.
Roadmap
What we're building, in order.
Network foundation
Node client beta across Mac, iPhone and iPad. Sign in with Apple. Live telemetry in the Explorer — real devices, real jobs, real MLX.
Workloads
Open the marketplace: inference batches, embeddings, fine-tunes and rendering jobs from partner compute buyers, settled in MLX.
Scale
Developer SDK and job APIs. Larger, longer workloads. A scheduler that treats a million devices like one machine.
Frontier
The largest unified pool of Apple Silicon on earth, priced per job. The open-weight counterweight to the data center.
Get in early
Own the hardware? Put it to work.
The node client beta opens soon. Email us to join the waitlist and you'll get the download link the day it ships — first come, first admitted.
Sign in with a passkey — no password, no email. Your device is the login. No payment required to join the waitlist; the beta is free and you earn MLX for compute your device actually delivers.