This week in Tom's Hardware Premium: November 14, 2026 — Mac Studio versus DGX Spark, AMD's MEXT cuts DRAM requirements in half, and the Huang saga reaches critical mass

Seismic appears to set the tone for this week's events in the hardware industry, and we've covered the biggest beats, including deeper dives on hardware and AI topics over on Tom's Hardware Premium. Stories this week are colored by the fact that Nvidia founder Jensen Huang has stepped away from his post as CEO. The ramifications are wide and far-reaching, as the company transformed from a gaming GPU manufacturer to an AI business over the past five years.

First up, it's not only hyperscalers and neoclouds that have their eyes on AI. An increasingly popular segment among enthusiasts is PCs dedicated to running local AI models. With so many choices on the market currently, from the recently released RTX Spark to AMD's Strix Halo platform, we take a look at two unlikely contenders that are now going head-to-head: Nvidia's DGX Spark and Apple's Mac Studio. We ran a suite of real-world and synthetic benchmarks across both devices and tested large models on both machines. With prices at eye-watering levels, it's more crucial than ever that you make the correct choice for your intended workloads.

• Going deep on local AI performance benchmarking with Apple’s Mac Studio with M5 Ultra taking on Nvidia DGX Spark — inference engine recipes mandatory for bigger models as MLX stack matures

DRAM capacity constraints have been an industry-wide sticking point as the AI buildout accelerates demand for dies. With costs spiraling across both client and enterprise hardware, AMD has shared further detail about MEXT, a predictive memory engine which aims to offer 99% of a server's performance while running on half the DRAM. Powered by a transformer model and clever use of flash storage, AMD is already running 145 servers with the system in place, with a view to expand that figure to 2,000 next year.

• AMD shares how it can retain 95% of server performance with just half the DRAM, sidestepping memory shortages with MEXT — tech powers over 100 servers already running, with plans to expand to 2,000 servers in 2027

Next, we sat down with Cognichip's Chief Product Officer in an extensive interview to explore how the company is rolling out AI-powered chip design tools. Back in September, the company revealed a suite of proprietary AI models, named Artificial Chip Intelligence (ACI), which are specifically developed for semiconductor engineering. While the tools have yet to be used in a shipping product, it could substantially improve chip development productivity by automating labor-intensive tasks.

• Cognichip reveals how its new AI-assisted program can slash chip design times from months to days — automating labor-intensive tasks cuts design time down, Chief Product Officer explains

The Jensen Huang scandal deepens, as the now-former Nvidia CEO has become embroiled in a series of alleged extramarital affairs and has since resigned from Nvidia. While much of the past week's stories have centered on his relationship with alleged Chinese Spy Alice Huang, we explored the former CEO's unlikely relationship with Kang Wei, Ms. Huang's husband, who recently landed in hot water with China's Securities Regulatory Commission.

• How Shanghai fund manager Kang Wei introduced his wife Alice Huang to Jensen and built a $7.5 billion Nvidia AI GPU smuggling operation — Mistress's husband joked she was the Nvidia CEO's 'personal assistant'

Nvidia-based compute remains difficult to find in China, especially as AI labs line up to order racks and newer cards like the RTX 5500 Pro, with over 6 million units reportedly ordered for the export-compliant GB202-based AI accelerator. But the Chinese AI labs that are chasing after Nvidia's tech may still have to pay a premium and ensure that they're running enough homegrown silicon to not fall afoul of Beijing's rules. We explore how some companies have now moved to renting compute in Southeast Asian countries like Malaysia and Singapore in an attempt to get around export controls.

• Nvidia cracks down on overseas cloud AI GPU rentals for China — partners face 51% non-Chinese quota as black market B300 rack prices surge to $2.8 million

Google's one of the biggest competitors to Nvidia, especially in the AI domain. The company's 8th-gen TPU lineup has now expanded even further, with two new chips designed for long-context agentic AI and security. These two new chips, TPU 8i Max and TPU 8tx, signal a further diversification across Google's eighth-generation TPU stack. While TPU 8i Max doubles the specs of the original TPU 8i, TPU 8tx makes a more interesting change to AMD's EPYC to serve its use in specialized security workloads.

• Google adds two new specialized chips for AI agents and security to 8th-gen TPU lineup — TPU 8i Max puts two chips on one package for long-context work, TPU 8tx swaps Arm hosts for AMD EPYC

Lastly, a new proposed framework aims to track how easy it might be for an AI model to break through its safeguards and generate potentially harmful content. The Frontier Model Foundation has proposed a Jailbreak Rating Framework which aims to classify the type and severity of a possible LLM jailbreak. With four scoring components in hand, the foundation aims to identify possible jailbreak vectors and inform labs of them before they happen in the real world, with potentially disastrous consequences.

• New framework grades AI jailbreaks across capability gains, scope, and how easy they are to deploy — Frontier Model Forum proposes CVSS-style rating system for new AI models

To read any of the articles we mention above, as well as unlock Bench, you'll need to purchase a Tom's Hardware Premium subscription. Your support goes a long way in helping us to publish deeper, more insightful articles that might not otherwise be covered.