25 September 2026 — Apple is pitching its new Mac Studio, priced from $5,499, to businesses seeking to run AI models without paying cloud providers per token, Benzinga reported on 22 September.
Key points
- Apple hardware chief Johny Srouji says businesses can reuse the Macs without a per-token charge.
- A Mac Studio with 256GB of memory and 16TB of storage costs $18,299.
- Apple demonstrated four Mac Studios running a trillion-parameter model together.
- Nvidia’s data-centre revenue reached $89 billion last quarter, up 117% from a year earlier.
Srouji’s case for owning a Mac Studio
Apple hardware chief Johny Srouji put the distinction in terms of repeated use: “There’s no cost per token. You’re just using the machine again and again,” he said. Under Apple’s pitch, a business buys the computer upfront and runs suitable AI models locally instead of paying a cloud provider or AI company each time it uses them.
The purchase price rises with the configuration. A Mac Studio with 256GB of memory and 16TB of storage costs $18,299, against the $5,499 starting price for the M5 Ultra model. Apple raised Mac prices in June, citing higher memory and storage costs.
The upgraded Mac Mini and Mac Studio began shipping on 22 September and are being pitched for corporate work including software development and complex business tasks, NDTV Profit reported. Customers pay upfront for the Macs. Apple pitches local inference as an alternative to recurring cloud usage charges.
Four Mac Studios run a trillion-parameter model
Apple demonstrated four connected Mac Studios running a trillion-parameter AI model. The cluster could operate from a single wall outlet, Benzinga reported. In the demonstration, the model was used to identify and repair a graphics coding problem, according to NDTV Profit.
Apple’s unified memory design allows a Mac’s CPU and GPU to use the same pool of memory, making it easier to run large models locally. The company has also added chip-to-chip networking using RDMA over Thunderbolt, a connection intended for demanding AI workloads.
The demonstration used multiple machines for one model, rather than one desktop handling the task alone. Apple is selling the Macs for workloads that businesses can place on their own equipment. Apple’s own most advanced cloud model reportedly runs on Nvidia GPUs through Google Cloud.
Nvidia’s $89 billion data-centre business
Nvidia’s data-centre revenue reached $89 billion last quarter, up 117% from a year earlier. Cloud providers and AI companies continued buying computing infrastructure during the period. Data centres accounted for $89 billion of Nvidia’s $96.2 billion in total quarterly revenue, GuruFocus reported.
Nvidia chief executive Jensen Huang described the value of usage in the company’s latest earnings release: “Its tokens are productive and profitable. Now, compute is revenue,” Benzinga reported. Nvidia also sells products intended to run AI models locally, including DGX Spark and RTX Spark.
Apple’s machines target some of the inference work for which customers might otherwise rent data-centre capacity. Nvidia’s desktop products put it in that local market too, alongside its much larger data-centre operation.
Apple’s 4.6% enterprise desktop share
Apple held 4.6% of the enterprise desktop market, compared with 91.3% for Windows, according to IDC figures reported by NDTV Profit. That existing desktop footprint is the market into which Apple is pitching the upgraded machines for corporate AI work.
Microsoft is also pursuing AI that runs on devices. Its chief executive, Satya Nadella, has described that approach as “unmetered intelligence.” Nvidia, meanwhile, is expanding its local AI hardware range alongside its sales to data-centre customers.
Apple’s 512GB Mac Studio configuration, intended to handle larger models, is due in late October.
