Chipmakers and AI agents dominated the last 24 hours. AMD became the 14th U.S. company to cross a $1 trillion valuation on a rally analysts tied to AI agent compute demand. Meta emergency-patched a zero-day in its new Muse AI assistant hours after disclosure. The EU fined Google €403 million over location-data handling, and Alibaba unveiled a new AI chip alongside plans for 5–10 trillion-parameter models. Xiaomi launches its 18 Pro series today.
AMD crosses $1 trillion on the AI agent trade
Advanced Micro Devices soared past $1 trillion in market capitalization for the first time on Monday, September 21, closing around $610–$613, up around 9–10%, after an intraday high of $613.92. AMD is the fourth U.S. chipmaker to reach the mark, after Nvidia, Broadcom, and Micron — and worth four times its $255 billion valuation a year ago.
Mizuho analyst Daniel O’Regan attributed the surge to AI agents: Meta’s recently launched Muse assistant, released September 8, runs constantly in the background and handles multiple tasks, which he said has “much bigger compute implications than a simple chatbot query,” driving a “full-blown feeding frenzy” in chip stocks. Arm closed up 17.1%, Intel up 12.1%, and Qualcomm up about 4% the same day.
| Figure | Detail | Source |
|---|---|---|
| $1 trillion | AMD’s first crossing, September 21 | Reuters |
| ~$610–$613 | Close, up ~9–10% | Reuters |
| $255 billion | AMD’s valuation one year earlier | Morningstar |
| $5 trillion+ | Nvidia’s valuation, for comparison | Reuters |
| +17.1% / +12.1% | Arm / Intel on the same day | Morningstar |
Meta hotfixes a zero-day in its Muse AI agent within hours of disclosure
Meta pushed an emergency hotfix to its Muse macOS app on September 22 after macOS security researcher Patrick Wardle publicly disclosed a zero-day on September 21. His proof of concept, titled “not-a-mused,” showed how an unprivileged local process could redirect the app’s dictation traffic to an attacker-controlled server and silently hijack the agent.
The flaw centered on an undocumented internal setting, endo_voyager_dictation_endpoint, which controls where the app routes audio captured when a user clicks Muse’s microphone button. Meta designed Muse to send dictation to its cloud servers for transcription rather than processing it on-device, and the setting was modifiable by any process under the logged-in user’s account. No administrator password, no permission prompt.
David Singleton of Meta Superintelligence Labs confirmed the patch shortly after midnight, calling it “a local privilege escalation attack, not a remote exploit” requiring malicious code already on the machine. Wardle confirmed the fix at 06:36 UTC with a “Hooray, hot-fixed!” The concern that remains: Muse asks for sweeping permissions across email, calendar, payments, and health data, and Wardle noted a ClickFix-style social engineering trick could supply the local foothold a remote attacker needs. As agents take over more of the user’s digital life, their attack surfaces grow with them.
EU fines Google €403 million over location data
Ireland’s Data Protection Commission fined Google €403 million ($463 million) on September 21 for unlawfully processing users’ location data under the GDPR between May 2018 and February 2020. The watchdog found Google did not lawfully process location data in its Web & App Activity and Location History settings, or in the Location Accuracy feature of Android.
The investigation, opened six years ago, produced the fourth-largest privacy fine the Irish regulator has issued, plus an order for Google to reach GDPR compliance within six months. Google called the case “historical policies that have since been updated.”
Worth noting for Android developers: retention “for longer than necessary” aggravated the finding, and the regulator cited location data’s power to “influence [users] with ads or to infer their interests.” It also lands amid a broader enforcement wave: the EU fined Google $1 billion in July over alleged Digital Markets Act violations, with payment due September 25.
Alibaba unveils the Zhenwu V900 AI chip and plans 5–10 trillion-parameter models
At its Apsara Conference in Hangzhou on September 22, Alibaba rolled out the Zhenwu V900, which CEO Eddie Wu called “the most powerful AI chip in China today,” delivering three times the performance of its predecessor, the Zhenwu M890. V900-based clusters can scale to 500,000 accelerator cards, with mass production planned for early 2027.
Alibaba also announced plans to train a new Qwen successor at 5–10 trillion parameters, up from the 2.4 trillion of its current flagship Qwen3.8-Max, and a target of more than 20 gigawatts of global AI data-center capacity by 2032. The company’s ADRs rose 3.3% in premarket trading. The timing is pointed: the announcement came days before Chinese leader Xi Jinping’s scheduled state visit to Washington, where AI and trade are expected to be on the agenda.
Xiaomi 18 Pro series launches today
Xiaomi launches its 18 Pro series in China today, September 23, at 19:00 local time. Headline specs, largely leaked ahead of the event: the Pro Max carries an 8,500mAh silicon-anode battery (32% silicon content), both models get dual 200MP Leica cameras (main plus periscope telephoto) and 100W wired charging, and both are expected to run Qualcomm’s Snapdragon 8 Elite Gen 6 with HyperOS 4 based on Android 17.
Official pricing is unannounced, but the direction is up: the Xiaomi 17 Pro launched at CNY 4,999, and memory costs for a 12GB/512GB phone have risen about CNY 1,500 since early 2025, according to Xiaomi president Lu Weibing. Treat the leaked specs as expected until confirmed on stage.
Key takeaways
- Agentic workloads are repricing chips. AMD’s $1 trillion crossing was tied by analysts to AI agents, not chatbot queries — always-on agents demand more compute per user.
- Agents are the new attack surface. Meta patched Muse’s zero-day within hours, but the pattern — cloud-based audio, broad permissions, permissive local settings — deserves scrutiny across every vendor.
- EU enforcement compounds. Google’s €403M GDPR fine arrives weeks before a $1B DMA payment deadline — Europe’s regulatory pressure is a recurring operating cost now.
- The compute race is geopolitical. Alibaba’s V900 and 10T-parameter plans are the hardware side of China’s push for AI self-sufficiency ahead of high-stakes trade talks.
References
- Reuters via Channel News Asia — “AMD joins $1 trillion club as chipmakers rally on AI-driven demand” — channelnewsasia.com
- Morningstar (MarketWatch) — “AMD hits a major milestone as AI stocks blaze higher” — morningstar.com
- Associated Press via Indian Express — “Google fined $463 million for breaching EU rule on location data” — indianexpress.com
- TRT World (AP) — “EU fines Google $463M over location data practices” — trtworld.com
- Unite.AI — “Meta Hot-Fixes Muse Zero-Day That Let Attackers Hijack the AI Agent” — unite.ai
- iPhone in Canada — “Muse Mac App Security Flaw Allows Hackers to Exploit the AI Agent” — iphoneincanada.ca
- Associated Press via ETV Bharat — “China’s Alibaba Unveils New Powerful Chip And Ambitious AI Model Plans” — etvbharat.com
- Associated Press via Deccan Chronicle — “China’s Alibaba Unveils New Powerful Chip and Ambitious AI Model Plans” — deccanchronicle.com
- NokiaPowerUser — “Xiaomi 18 Pro Series Launching Sept 23: Specs, Release Timeline & Expected Prices” — nokiapoweruser.com
- TechCabal — “Xiaomi 18 Pro: Price, release date, specs and features” — techcabal.com
Was this useful?
Thanks for the feedback.
