NVIDIA says Jetson Orin Nano 2 delivers 78 TOPS of AI compute with 8GB of memory in the same form factor as its predecessor, with availability expected in the first half of 2027. [Image courtesy of NVIDIA.]
NVIDIA just announced the Jetson Orin Nano 2, a new computer platform for robotics and edge AI that the company says doubles inference performance compared with the Jetson Orin Nano Super. It can also run at 40% less power at equivalent performance. NVIDIA has not yet disclosed pricing, and says the module and developer kit are expected to become available in the first half of 2027.
Compact AI systems are already finding their way into research labs. Duke biomedical engineers, for example, have used the Jetson Orin Nano to control and process images in a compact optical coherence tomography system and to process holographic cytometry data in real time, while researchers developing the Chemist Eye platform use the same hardware for vision-based safety monitoring and robot decision-making in self-driving laboratories.
Meanwhile, Apple, which has emerged as an unlikely beneficiary of the local AI boom, announced a new Mac mini, offered with the M6 or M5 Pro. Preorders opened today, with delivery beginning September 22. The M6 is Apple’s first chip built on a 2 nm process.
In a press release, Apple says the new Mac mini offers “a massive leap in AI performance,” and is designed with “always-on agentic computing” in mind. Specifically, Apple says the M6 Mac mini delivers up to four times faster AI performance than the M4 model, including up to 4.8 times faster LLM prompt processing in LM Studio. M6 configurations have 16GB of unified memory as standard, are configurable to 32GB, and provide up to 170GB/s of memory bandwidth. Whereas Apple originally positioned the Mac mini as an affordable entry point for consumers switching to Mac, and at times went years between major updates, over roughly the past six months the diminutive desktop has gained traction among developers and AI enthusiasts running local models and autonomous agents such as OpenClaw, with Reuters reporting that demand depleted inventories at several Apple stores.
The new Mac mini launch comes amid criticism over delays to its Apple Intelligence features, including a more capable Siri. Apple has agreed to a $250 million settlement over claims that it advertised enhanced Siri and AI capabilities that had yet to arrive.
Apple says its new Mac mini, powered by the M6 and M5 Pro chips, delivers up to 4x faster AI performance and expanded memory options for local agentic workflows. [Image courtesy of Apple.]
Apple and NVIDIA are targeting an increasingly crowded and varied market for compact AI computing. NVIDIA also unveiled RTX Spark in May, a Windows platform purpose-built for personal agents. NVIDIA says compact PCs using it will offer up to 128GB of unified memory and up to one petaflop of AI compute, with systems from ASUS, Dell, HP, Lenovo, Microsoft and MSI expected this fall.
NVIDIA’s own DGX Spark, for example, which began shipping in October 2025, is a purpose-built desktop AI system based on the Grace Blackwell GB10 chip. NVIDIA currently sells its 4TB Founders Edition for $4,699, with 128GB of unified memory, 273GB/s of memory bandwidth and up to one petaflop of FP4 AI performance. NVIDIA also offers a two-system bundle, and the systems can be linked over 200 Gb/s ConnectX-7 networking for distributed workloads.
ASUS, Dell, Lenovo and other manufacturers sell systems based on the same GB10 platform, with differences in storage, cooling, support and other system-level features.
Meanwhile, AMD’s Ryzen AI Max platform takes another approach, combining x86 CPUs, integrated Radeon graphics and large pools of shared memory in compact workstations from companies including HP and Framework.
As the table below shows, the compact AI computing landscape spans markedly different use cases, from low-power systems built into robots to deskside machines for developers, researchers and users running local models and autonomous agents. Their performance figures also require some caution: vendors cite a mix of TOPS, FP4 petaflops, memory bandwidth and application-specific benchmarks that are useful within context but cannot be compared directly across systems.
System
Basic positioning
Memory
Claimed AI compute / relevant spec
Current price
Source
NVIDIA Jetson Orin Nano 2
Robotics / edge deployment
8GB
78 TOPS; 2x inference vs. Orin Nano Super, or same performance at 40% less power in 15W mode
Not announced; module and developer kit expected H1 2027
NVIDIA
Apple Mac mini M6
General desktop / local agents
16GB standard; configurable to 32GB
Apple claims up to 4x faster AI vs. M4 Mac mini and 4.8x faster LM Studio prompt processing; 170GB/s
$899+
Apple
Apple Mac mini M5 Pro
Higher-end local AI / dev
Up to 64GB
Apple claims up to 4x faster LM Studio prompt processing vs. M4 Pro; 307GB/s
$1,699+
Apple
NVIDIA RTX Spark PCs
Windows personal agents / local AI
Up to 128GB
Up to 1 PFLOP AI compute
TBD; systems expected fall 2026
NVIDIA
Apple Mac Studio M5 Max
AI/pro workstation
Up to 128GB
Apple claims up to 3.9x faster LLM prompt processing in LM Studio vs. M4 Max; up to 614GB/s
$2,499+
Apple
Apple Mac Studio M5 Ultra
Large local models
Up to 512GB (512GB configuration expected late October 2026)
1.2TB/s memory bandwidth
$5,499+
Apple
NVIDIA DGX Spark
Dedicated desktop AI development
128GB
Up to 1 PFLOP FP4; 273GB/s
$4,699
NVIDIA
ASUS Ascent GX10
DGX Spark-class AI workstation
128GB
1 PFLOP FP4
$4,699
ASUS eShop
Dell Pro Max with GB10
Enterprise local AI workstation
128GB
GB10 Blackwell, 1 PFLOP FP4
Configuration-dependent; verify current listing
Dell
Lenovo ThinkStation PGX
Enterprise AI workstation
128GB
Up to 1 PFLOP FP4 with sparsity; 273GB/s
Varies by storage config; verify current listing
Lenovo Press
Framework Desktop, Ryzen AI Max+ 395
x86/local AI PC
Up to 128GB
Radeon 8060S; 256GB/s
Configuration-dependent; 128GB mainboard $3,149
Framework specs; mainboard
Prices as of Aug. 25, 2026, and subject to change amid ongoing DRAM supply constraints.