
Nvidia puts 128GB local AI desktop power on shelves
RTX Spark compact desktops go on sale with 128GB memory and Blackwell GPUs for agents and creation.
Chips, GPUs, data centres and the physical infrastructure the AI economy runs on.

RTX Spark compact desktops go on sale with 128GB memory and Blackwell GPUs for agents and creation.


Chris Lattner pitches Mojo and Max as write-once AI software. The proof on performance and openness is still missing.

NVIDIA closed one Korean bet and lined up three more, with RF interconnects at the center.

NVIDIA brings Rebellions and SK Telecom into NVLink Fusion as Korea sovereign push moves from chips to deployment.

Nvidia says open RAN and simulation tools passed 1M downloads as labs test robotics, receivers and power control.

NVIDIA says full-stack codesign lets its Vera Rubin NVL72 platform deliver 30x more agentic throughput per dollar with only 3% more HBM.

NVIDIA says Vera delivers 1.6x more agents per watt than the fastest x86 CPU and 2x faster agents than the most efficient x86, with partners reporting 1.5x to 2.2x gains.

Apple introduced the 14-inch MacBook Pro with M6, promising faster CPU and GPU and stronger on-device AI to challenge rival laptop silicon.

Hardware Unboxed finds GPU prices up 8% in September to 34% over MSRP, with Nvidia far above AMD and the RTX 5090 near $6,000.

Ed Zitron tells The Tech Report Microsoft has only 2 GW of AI chips online, leaving tens of billions in GPUs warehoused.

OpenAI's Jalapeño inference chip promises 1.5-1.9x throughput per watt with deployment by year-end, but no prediction market yet prices the rollout.

Broadcom guided AI chip revenue to double in fiscal 2027 and again in 2028, but shares fell 6% on concentration and competition fears.


Arm CEO Rene Haas tells NoPriors why CPUs still anchor AI data centers and why Arm is now building its own AGI CPU with Meta.

Hot Chips 2026 put Arm at the center of four major CPU reveals. The common software foundation is now a competitive lever for agentic AI hardware.

Cerebras showed 4,400 TPS on CS4 and promised 10,000 TPS on CS5, with OpenAI already running its largest model 14x faster.

Nvidia will not compete with customers and is resetting gross margins to 72-73% while building Neocloud competition to protect demand.

Microsoft reframes AI infrastructure around yield, with Azure Maia and Cobalt co-design aimed at agentic workloads that use 3,400x more tokens.




Angelo Zino of CFRA Research discusses Nvidia's strong AI-driven growth, projecting 70% revenue expansion. He covers hyperscaler demand, customer diversification, and competitive threats.



Lam Research CEO Tim Archer discusses the surge in AI-driven chip demand and the company's $3B investment in R&D labs to meet the growing need for advanced manufacturing processes.

AWS and NVIDIA are expanding their partnership, committing to deploy 2 million more NVIDIA GPUs by 2028 and integrating NVIDIA Vera CPUs.

The Maia 200 AI accelerator pioneers SDLA, shifting AI compute to data-movement-centric architectures for superior efficiency and performance.

Nvidia faces a crucial earnings report as its stock aims to rebound. Apple also launches new Macs with upgraded processors, while OpenAI claims its new chips outperform Nvidia's.

OpenAI's VP of Hardware, Richard Ho, discusses the company's new custom inference chip, Jalapeno, claiming it outperforms Nvidia in key benchmarks and highlighting its novel architecture.