Anthropic flags AI-run cyberattacks, Microsoft plans 38 GW of data center capacity by 2032, and OpenAI pauses new $200 Pro sign-ups

Anthropic reports AI-run cyberattacks, alleged model theft and bioweapons misuse

Anthropic published its September 2026 threat intelligence report on September 10, covering misuse it says it detected and disrupted between December 2025 and August 2026. One campaign, tracked internally as GTG-20006, pushed multi-agent AI workflows through much of the attack chain and matched the tradecraft of Midnight Blizzard, the Russia-linked group that U.S. authorities have tied to Russia's SVR. Anthropic said the operation touched more than 20 organizations, mainly in Ukraine and Europe, among them government ministries, defense and intelligence bodies, diplomatic missions and defense supply-chain companies. Autonomous agents checked whether security products flagged the group's malware, then rewrote and rebuilt the code until it slipped past detection, the report said. It also documented five cases in which scientists used Claude in ways that could support biological weapons development, and six involving software for conventional weapons such as firearms, missiles and armed drones. On the commercial side, Anthropic alleged that seven China-based labs, including Alibaba, Moonshot, DeepSeek and Xiaomi, ran unauthorized distillation campaigns against Claude, and attributed the largest to operators linked to Alibaba, with more than 151 million exchanges between May and July 2026. Anthropic said it banned the accounts and tightened its safeguards. China's foreign ministry said it was not aware of the report. (TechNode Global / Reuters, Sept 11, 2026)

Microsoft plans to more than triple data center capacity, to over 38 GW by 2032

Microsoft intends to expand its global data center capacity from roughly 12 gigawatts today to more than 38 gigawatts by 2032, Bloomberg reported on September 10, citing people familiar with the plans. The roadmap covers company-owned and leased facilities and excludes capacity rented from so-called neocloud providers such as CoreWeave. About 2 GW of current capacity is dedicated to AI chips; roughly a third of the planned 38 GW is expected to be AI-specific. The buildout follows a stretch of acute shortages that pushed some workloads toward rivals: retailer Temu signed a major cloud deal elsewhere after Microsoft could not supply the resources it wanted, and GitHub suffered an outage of nearly eight hours in August that was attributed in part to capacity limits. Microsoft added 88 data centers in fiscal 2026 and carries lease commitments above $329 billion. CFO Amy Hood said in April that the company expected to stay capacity-constrained at least through 2026. A key expansion site is the East US 3 cluster near Atlanta. Microsoft has not confirmed the 38 GW target in a filing or press release. (Bloomberg via Financial News, Sept 10, 2026)

OpenAI pauses new sign-ups for its $200 ChatGPT Pro plan as Astra demand strains capacity

OpenAI has stopped accepting new subscriptions and upgrades to its $200-a-month ChatGPT Pro plan, saying demand for the GPT-6 Astra model is exceeding its capacity. Thibault Sottiaux, the OpenAI executive who leads Codex, wrote on September 10 that the $200 tier "puts the most strain on our systems" and that pausing it was the smallest step that would preserve broad access while the company adds capacity. Existing Pro subscribers keep their accounts and limits, and OpenAI's other plans and its API remain on sale. Help documentation states that canceling or downgrading during the pause blocks re-subscribing until it lifts. OpenAI has not given a reopening date. GPT-6 Astra shipped on September 3 and rolled out across ChatGPT Plus, Pro, Business and Enterprise, the API, Microsoft Azure and Amazon Bedrock, at $10 per million input tokens and $50 per million output tokens. (The Verge / TechCrunch via AI Tech Daily, Sept 10-11, 2026)

Pentagon in talks to lend about $5 billion to AI cloud startup Fluidstack

The U.S. Department of Defense is negotiating a loan of roughly $5 billion to Fluidstack, an AI cloud infrastructure company, which would be the largest ever made by the department's Office of Strategic Capital. Fluidstack was founded in London in 2017 and has moved its headquarters to New York while expanding across the United States. It has secured multiyear data center leases, raised $1.5 billion from Jane Street and works with Google and Anthropic. Erebor Bank, backed by defense-tech entrepreneur Palmer Luckey, is advising on the potential financing, according to the reports. The loan would put the U.S. military directly into a capital-intensive part of the AI supply chain: the data centers, chips and electrical systems used to train and run advanced models. (The Wall Street Journal / Reuters via Tech Startups, Sept 11, 2026)

Huawei and Cambricon raise AI chip quotes as HBM costs climb

Huawei has lifted the indicated price of its Ascend 950DT accelerator by 20% to 50% over two months to more than 250,000 yuan (about $37,255), while Cambricon has raised quoted pricing for its upcoming 690 chip by 20% to 30%, Reuters reported on September 10, citing three people with direct knowledge of the matter. The driver is high-bandwidth memory, where a global shortage has been compounded for Chinese vendors by U.S. export controls on advanced memory, pushing them toward grey-market procurement that carries a premium. The Ascend 950DT is scheduled to launch in the fourth quarter of 2026 with 144 GB of Huawei's HiZQ 2.0 memory and 4 TB/s of bandwidth. Cambricon told 21st Century Business Herald on September 11 that it had not issued any price-increase announcement. (Reuters, Sept 10, 2026)

Enflame Technology jumps on its Shanghai STAR Market debut

Enflame Technology (688801) began trading on Shanghai's STAR Market on September 11, opening at 410 yuan against an IPO price of 142.18 yuan, a gain of 188.37%. The stock rose as much as 234% intraday to 475 yuan before closing up 179.2% at 397 yuan, giving it a market value of about 170.9 billion yuan. Tencent is Enflame's largest customer, and sales to it accounted for 83.79% of 2025 revenue. First-half 2026 revenue reached 1.12 billion yuan, up 279.08%, with a net loss of 632 million yuan. The listing completes the group of four leading Chinese cloud-AI chip developers — Moore Threads, MetaX, Biren and Enflame — on public markets. (Securities Times, Sept 11, 2026)

EU cyber reporting rules for connected products take effect

Manufacturers placing products with digital elements on the EU market became subject to mandatory cybersecurity reporting on September 11, under Article 14 of the Cyber Resilience Act. Companies must file an early warning within 24 hours of learning of an actively exploited vulnerability or a severe incident, a fuller notification within 72 hours, and a final report within 14 days after a corrective measure is available for a vulnerability or within one month for a severe incident. Reports go through the Single Reporting Platform run by the EU cybersecurity agency ENISA. The obligations apply to products already on the EU market as well as new ones, while the Act's main technical requirements apply from December 11, 2027. (European Commission, Sept 11, 2026)

Xiaomi plans 200 billion yuan in R&D over five years

Xiaomi founder and CEO Lei Jun said on September 10 that the company will invest 200 billion yuan in research and development over the next five years. The statement followed a Xinhua report on Xiaomi's in-house chip and vehicle programs, including its XRing processors and the Pengcheng series of cars. (Xinhua via IT Home, Sept 10-11, 2026)