This week on Tom's Hardware Premium: October 10, 2026

In case you've been missing out on the action over on Tom's Hardware Premium, we're back with another roundup to fill you in on everything you missed. Central to this week's theme is how Artificial Intelligence and its subsequent buildout are currently impacting all areas of the semiconductor industry.

Headlining this week is our report on how smart software utilization may play a part in the electricity consumption of AI data centers. We sat down and talked with experts in the domain to find out.

• Software could be the easiest fix for hyperscalers' AI power squeeze, researchers say — data center demand is expected to rival Japan's electricity usage by 2030

Reports state that Tencent has reached a deal to lease high-performance AI chips from Oracle to the tune of $7 billion over five years. While the figure on its surface sounds significant, the estimated lease price is around 43% less than an Nvidia H100 leased at $2.80 per hour on a one-year contract, meaning that Tencent may have scored a slight coup in the terms of its lease agreement. While no specific chips have been named, the size of the deal and the prices point toward a Hopper-class GPU. We explored how this rental loophole may be used by Chinese firms to potentially gain access to otherwise export-controlled hardware.

• Tencent scores 100,000 offshore AI chip deal with Oracle for $7 billion despite climbing prices — per-hour costs estimated to be 43% under standard H100 rental rates

If you cast your minds back to last week, Synopsys revealed a slew of new agreements and partnerships; one of the most notable among them was an agreement with OpenAI, unveiling a 'GPT-Synopsys' — a specialized AI model that leverages Synopsys' domain IP knowledge and EDA tools, with OpenAI's capability to design LLMs. GPT-Synopsys is, in all likelihood, a chipmaking-specialized AI model, which may play a part in the development of successors to OpenAI's Jalapeño AI ASIC. As we've previously explored, AI models designing chips will likely run into hurdles surrounding IP usage, which become a non-factor if both AI labs and IP holders decide to team up in a manner such as this, setting the stage for chip design's agentic future.

• OpenAI and Synopsys partner to build "GPT-Synopsys" for autonomous chip design — specialized AI model will operate EDA tools, allowing engineers to deliver more sophisticated chips faster

Neon is a noble gas that's widely used in the DUV chipmaking process, with over 70% of global supply being used by the now-older lithography machines. Crude neon production has usually been focused in Ukraine, and ever since the advent of the country's ongoing conflict with Russia, global supply has been disrupted. Now, companies are racing to bring neon gas recycling systems to market in order to make the most of supplies. Japanese lithography toolmaker Gigaphoton and rival Kanten Techno have both developed machines that recycle the material at varying degrees of success. While DUV lithography remains a primary focus for the Chinese chipmaking ecosystem, the material is not widely used in the newer EUV systems that many leading-edge nodes currently utilize.

• Gigaphoton debuts neon recycling system with claimed 50% recovery rate — systems throw a lifeline to chipmakers that utilize 70% of global neon supply in DUV lithography

Qualcomm has reportedly reached a deal with Huawei to license its patents for the first time. While the official announcement is light on detail and subject to approval, an additional report claims that Huawei's LogicFolding chip-stacking methods are also included in the deal, with products ranging across AI, networking, compute, and other technologies. What makes Huawei's LogicFolding approach so notable is that the technique follows and targets faster data transmission as a foil to not having access to the most advanced EUV machines. Additionally, many of the techniques that LogicFolding incorporates echo what others in the chip industry are already doing, which makes it an eyebrow-raising move on Qualcomm's part. Historically, patent licensing between the two companies has been on the other foot — Huawei licensing Qualcomm patents — not the other way around. We explore the deal in detail in our full article.

• Qualcomm will license patents behind Huawei’s LogicFolding chip architecture, report says — the Kirin 9050 Pro already uses it, with a teardown showing its lower die is mostly cache and I/O

Two former Groq employees and shareholders have filed a proposed class action lawsuit against Groq's former board and CEO, alleging that the company's board breached its duties and cost stockholders billions, following Groq's $20 billion deal with Nvidia in December 2025, according to a new report. The deal between Groq and Nvidia didn't involve an acquisition, and Nvidia is not a named defendant in the case, though the vast majority of Groq's staff have reportedly since joined Nvidia directly. The Groq3 LPU has since been showcased as a part of Nvidia's Vera Rubin ecosystem, which is currently ramping. Nvidia's Groq 3 LPX is scheduled to ship alongside Vera Rubin in H2 2026.

• Former Groq engineers sue board over $20 billion Nvidia deal, saying it handed Nvidia the LPU and the team that built it — plaintiffs allege the board kept billions from other shareholders

The AI train keeps rolling, with a new release from European firm Mistral, named Large 4, ML4, or, if you have seen any of the company's social media posts, Le Chonk. The race towards building capable open-weight models has predominantly been dominated by either U.S. or Chinese labs, meaning a European breakthrough is significant. Testing lab Artificial Analysis scored ML4 a reasonable 38 on its overall intelligence index, and it is currently outrun by numerous Chinese models, including Kimi K3 (max), GLM-5.3-Flash, and MiMo-V2.6.Pro. However, the model excels in cybersecurity tasks, only outmatched by a select few.

So, Europe's sovereign AI efforts appear to still be playing catch-up to Chinese alternatives, despite clear advancements. The global race to deploy capable closed and open-weight models marches on, and Mistral expects that the exact weights for ML4 will be released before the end of October.

• Independent tests rank Mistral’s new trillion-parameter Large 4 the best AI model outside the U.S. and China — but Chinese open-weights still overcome Europe's best efforts

To gain full access to 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.