Asus launches ProArt laptops in Australia that run 120-billion-parameter AI models offline

The 120-billion-parameter figure applies to one memory tier only. Asus ties it to configurations with 128 GB of unified memory ^[15]. The lower ProArt P16 variant has 24 GB ^[5], and the ProArt P14 range starts at 32 GB ^[6]. Those entry machines have about 19% and 25% of the top tier's memory ^[12][13]. A team that wants large-model work kept off the cloud is buying at 128 GB.

Asus aimed the launch at creative professionals who want AI workloads to run locally ^[3]. Every example of agent work it gives is a media task ^[10]. The data it says buyers keep control of is raw assets and intellectual property ^[2].

The demand case rests on the Telsyte trust figures, which show a 42-point gap between Australians who use AI and Australians who trust technology companies with their data ^[14]. itwire reported that Asus is looking to capitalise on that gap ^[8]. A national sentiment survey is not a procurement signal. It measures unease about where data goes, and testing whether a laptop answers that unease for a particular workload is a separate job for the buyer.

Asus sells the machine as a hybrid. "Australian creators now have the flexibility to run autonomous AI agents directly on the device or seamlessly tap into leading cloud services when they need them," said Bradley Howe, head of consumer at Asus Australia ^[9]. The Australian bundle includes three months of Google AI Pro, six months of 500GB Dropbox storage and a year of Asus Secure Auto-Backup at 200GB ^[11]. So a device bought to keep data local arrives with several ways for files to leave it.

The record here does not include Australian prices for the 128 GB machines, or any figure for how quickly a 120-billion-parameter model runs on them. Without both, a per-seat hardware line cannot yet be compared with a cloud bill. The laptops are on sale now ^[1]. It will take several refresh cycles to learn whether local inference wins a real share of sensitive AI work from cloud services.

I would expect a sensible first purchase this quarter to be a few 128 GB units, assigned to one named workload and timed and costed against the same work in the cloud. That choice shapes next quarter's decision. If the test holds, local AI becomes a capital line tied to the laptop refresh cycle. If it fails, the spend stays a cloud subscription.