24 articles
Index of editorial tags and filtered coverage.
In an interview with the Sources newsletter released this week, Qualcomm CEO Cristiano Amon revealed that several leading artificial intelligence companies are asking for phones able to continuously run models of at least 100 billion parameters by 2028, more than three times what today’s flagship chips can handle.
Ting-Wei “Willy” Sun, described by prosecutors as the “fixer” in a scheme that diverted about $2.5 billion in AI servers with Nvidia chips to China, pleaded guilty on October 8, 2026, in a federal court in Manhattan.
A group of 26 academics led by Will Fithian and Wesley Holliday of UC Berkeley published a study on October 9, 2026, outlining how to pause frontier AI training for at least ten years by replacing training chips with inference-only chips, with mutual verification between countries.
On 30 September 2026, Synopsys and OpenAI signed a multi-year agreement to develop GPT-Synopsys, a specialised model that will operate Synopsys EDA tools for PPA, timing and verification workflows, running on OpenAI-hosted infrastructure under revenue sharing, with early semiconductor customer engagements but no GA date or public benchmarks yet.
The Justice Department and Ars Technica detailed in October 2026 the arrest of Earthmade Computer’s CEO, accused of diverting A100 and H100 GPU servers via Malaysia and Singapore from October 2023 to August 2026 under the Export Control Reform Act.
In early October 2026, the Chips Joint Undertaking opened two calls worth €80 million: €10 million for AI-chip demonstrators already taped out, and €70 million for a shared evaluation platform and pilot procurement path. Both close on 27 October 2026 at 5 p.m. Brussels time.
On 30 September 2026 DeepSeek announced on WeChat, per Reuters and 1 October coverage, a partnership with Huawei to develop and open-source programming infrastructure for Ascend chips, including the TileLang language and compute/communication libraries. The stated goal: an independent GPU software ecosystem that reduces reliance on Nvidia.
The Information and CNBC reported on 27–28 September 2026 that China’s Ministry of Industry and Information Technology asked firms including ByteDance and Alibaba for purchase plans for the RTX PRO 5500—an 84 GB Blackwell workstation chip launched 14 September—and signalled intent to approve. ByteDance is said to be weighing on the order of one million units; Nvidia is said to plan China shipments from late December.
With the Trump–Xi summit closed without an AI pact, attention returns to Congress: the MATCH Act (H.R. 8170) and Remote Access Security Act (H.R. 2683) are the bipartisan vehicles CNAS and analysts flag to close remote access to controlled chips, align allied controls and limit servicing that sustains China's catch-up. As of late September 2026 both remain unfinished.
Per Bloomberg, TrendForce, and The Decoder reports (Sep 2026), Liang Wenfeng is driving a cluster of at least 160,000 Ascend 950DT chips at an Inner Mongolia datacenter, with deliveries spanning Q4 2026–Q1 2027, to train toward 2-trillion then 8-trillion-parameter models—a gradual migration under export controls, not a clean break from Nvidia.
Per Bloomberg (Sep 15) and TrendForce (Sep 16), Meta is testing MTIA 450 (Arke) for roughly H1 2027 deployment and preparing MTIA 500 (Astrid) for late 2027; Broadcom/TSMC inference chips, distinct from the Iris/14 GW story.
ASML EUV lithography demand and HBM/DRAM tightness harden in September 2026 as an infrastructure brake on AI compute; industry coverage describes faster build cycles and memory–logic competition for tool slots.
At the Washington summit in the week of Sep 24, Trump and Xi discussed AI regulation and the chips race; coverage describes tension between safety cooperation and race dynamics, without inventing quotes.
U.S. agencies and firms accuse Chinese labs of industrial-scale API distillation; Beijing denies it. On 23 September the issue crosses chip controls and the Trump–Xi agenda, without reducing Chinese progress to a single theft story.
On 23 September 2026 Germany’s SPRIND and the Netherlands’ NADI opened a ~€40 million challenge to accelerate AI-driven chip design; ~20 months, up to seven initial teams and applications closing 30 November.
CNBC on 23 September 2026 frames the Trump–Xi meeting around AI safety, chips and U.S.–China rivalry; the plausible outcome is a risk-alert channel (Bessent proposal), not a development pause, with Nvidia controls and distillation accusations as friction.
At Apsara Conference 2026 in Hangzhou, Alibaba launched the Zhenwu V900 accelerator (3× M890 performance), clusters up to 500,000 accelerators, 5–10 trillion–parameter Qwen plans, and more than 20 GW of global cloud capacity by 2032.
During the Lee Jae Myung–Macron summit (~8–9 Sep), Samsung announced a strategic partnership and investment in Mistral for customized on-prem models aimed at performance and yields; Jun Young-hyun framed secure AI over fab data as a competitive edge.
Outlets including Mashable report Nvidia could finance or anchor Anthropic’s expected public offering, tying capital, the chip ecosystem and rivalry with the OpenAI–Microsoft axis.
Arthur Mensch confirms Mistral is exploring proprietary ASICs while building a French DC with ~13,800 GB300s (~44 MW) and targeting ~200 MW of European compute by end-2027.
On September 14, 2026, Forbes and RuntimeWire covered Cognichip’s enterprise launch of Artificial Chip Intelligence (ACI), founded by Faraj Aalaei. The system uses native physics-informed models—not LLM wrappers—for microarchitecture, RTL, verification and PPA; one cited case compressed a 55-page spec into days versus 4–5 months. Public evidence remains company and customer benchmarks.
Per an internal memo reviewed by Reuters in July 2026, Meta planned to begin manufacturing its Iris accelerator from September within a four-generation MTIA program, with Broadcom on design and TSMC on fabrication. September coverage indicates production has started; infrastructure ambition is ~7 GW in 2026 and double (14 GW) in 2027, supplementing—not replacing—Nvidia and AMD GPUs.
New accelerators promise up to 4× better energy efficiency for LLM inference.
NVIDIA Blackwell, new Google TPUs, and hyperscaler ASICs defined who trains and infers at what cost.