Meta enters the frontier model race, chip stocks wobble as capital questions mount, Nvidia cedes ground to Huawei in China, and Microsoft bets big on its own silicon.
Meta’s Frontier Bid: Meet Muse Wonder
On Monday, Meta’s Superintelligence Lab released its first frontier models, called Muse Wonder and Muse Wonder Full — the company’s first direct assault on OpenAI and Anthropic. For years, Meta was content to play the lightweight lane with open models like Llama. That era appears to be over.
The pitch is all about memory. Meta calls it “active memory”: the idea that the model retains what matters to you over days, weeks, and months. On a benchmark called MemoryBench, Meta says Muse Wonder outscores rival models including Opus, Astra, and Fable.
Independent testing tells a more measured story. Artificial Analysis gives Muse Wonder Full a 56 on its Intelligence Index, behind Claude Opus 5.5 at 58, while the standard Muse Wonder scores 52 — just under GPT-6 Astra at 53. This looks like parity with the middle of the pack, not a leapfrog moment. Still, Mark Zuckerberg called the release “reaching the frontier,” and Meta’s stock jumped more than three percent.
The pricing is premium: $4 per million input tokens and $20 per million output tokens through the Meta Model API, right in line with the top Anthropic and OpenAI offerings. And the most capable version, Wonder Full, won’t even appear in the API — it goes only to paying Muse Code subscribers. The message is clear: Meta is a frontier lab now, and if you want its best model, you come to it directly.
The “Brutal Session”: AI Stocks Hit the Skids
Tuesday was, in Reuters’ words, a “brutal session” for chip stocks — the strongest sign yet that the mood around the AI trade of the decade is shifting. In Seoul, the KOSPI fell more than six percent, its fifth straight day of selling in chipmakers. Samsung dropped 10 percent, SK Hynix eight percent. In Tokyo, memory maker Kioxia plunged 13 percent. In the United States, the Philadelphia semiconductor index fell 2.7 percent on Monday.
It’s not just prices that are moving — it’s the cost of money. BlackRock’s research arm just downgraded its outlook on U.S. stocks to neutral, citing potential volatility around the AI buildout. The International Monetary Fund warned Tuesday that the consequences of an AI correction could be global.
The trigger is the capital question: who pays for all this compute? Reuters points to Firmus shelving a five-billion-dollar data center IPO in Australia, credit investors getting pickier about AI debt, and Broadcom’s CEO saying capital is now the biggest constraint on growth. When the people selling you the shovels say the money is the problem, that’s a signal.
Not everyone’s bailing. One fund manager told Reuters this is likely the next phase of the AI cycle, where markets focus on monetization and returns on invested capital — not the end of the trade. Korea, he said, could be an early warning of a broader shakeout.
Nvidia Loses Ground in China as Huawei Closes In
Over in China, the chip map is being redrawn. IDC data reported by Reuters shows Nvidia’s share of China’s data center AI chip market fell to 34 percent in the first half of this year. U.S. export controls have choked off Nvidia’s supply: the company shipped about 678,000 AI accelerators in China in the first half, down 44 percent from a year earlier.
Huawei is right behind at 28 percent, shipping 572,000 accelerators — up from about 300,000 last year. Nine of the eleven Chinese chipmakers IDC tracks grew shipments. Alibaba’s chip unit T-Head holds 7 percent, Cambricon 5 percent, and Baidu’s Kunlunxin 4 percent. Jensen Huang has been warning about exactly this: that export curbs would force a U.S. company to hand China a fifty-billion-dollar market to Huawei. The data says it’s happening.
Tencent Borrows Big to Keep Spending
Meanwhile, Tencent is borrowing big to keep spending. On Tuesday, Tencent sold U.S. dollar and offshore yuan bonds in a deal that could raise at least $4.5 billion, with investor demand reportedly near four times that amount. It’s the latest in a wave of technology firms borrowing to fund AI infrastructure — even as investors grow more selective about which companies deserve the money.
Microsoft’s Maia 300: If You Can’t Beat the Cost, Build the Chips
On Monday, Microsoft unveiled Maia 300, its third-generation custom AI chip, and said it plans to deploy more than a million of them next year. The chip uses chiplets built on TSMC’s two-nanometer process, and Microsoft says it will deliver about 40 percent more useful computing per megawatt than leading GPU-based systems. The chips go into racks of 96, drawing about 167 kilowatts each.
The message is aimed straight at Nvidia’s grip on Microsoft’s costs. The previous generation, Maia 200, already runs Microsoft’s in-house models and some OpenAI workloads on Azure at 30 to 40 percent lower cost than Nvidia hardware. Microsoft says it will mainly rely on its own silicon for incoming AI requests, with multi-gigawatt deployments starting in 2027. Google and Amazon have been doing this for years. Microsoft is catching up fast.
Quick Hits
- The AI search squeeze on media is getting real. In the U.K., Reach — the publisher of the Daily Mirror — says its digital revenue fell more than 20 percent last quarter, blaming Google’s AI Overviews for draining traffic from publishers. The company’s overall revenue fell 14 percent, and shares dipped in early trade.
- DigitalOcean wants your AI agents to improve themselves. The company announced Agent Learning Services — Insights, Signals, and Evaluations — plus Agent Data Services, a feedback loop that lets customers’ AI agents record what they do, flag where they fall short, and test fixes before customers see them. The bet: launching an agent is easy, but keeping it from quietly plateauing is the hard part. DigitalOcean will demonstrate the services live at the Open Intelligence Summit on October 13.
Sources
- Meta releases first frontier models Muse Wonder and Muse Wonder Full: TechCrunch
- “Brutal session” AI trade selloff rocks chip stocks: Reuters
- Tencent sells dollar and dim sum bonds to raise at least $4.5B amid AI spending: Reuters
- Nvidia’s China AI-chip share falls to 34% as Huawei reaches 28%: Reuters
- Microsoft unveils Maia 300 custom AI chip, plans 1M+ unit deployment: Reuters
- Daily Mirror publisher Reach hit by Google AI Overviews traffic slump: Reuters
Adapted from the October 13, 2026 episode of Daily Tech and AI.
Daily Tech and AI is AI-voiced and produced by Angelo Ruiz.
Leave a Reply