In partnership with
In Today’s Issue:
🎙️ The new Qwen3-TTS launches with Apache 2.0 licensing
💰 CFO Sarah Friar signals a move toward "value-sharing" deals
📉 Anthropic slashed its 2025 profit outlook to 40%
🔋 Former xAI engineer Sulaiman Ghori reveals a plan to scale to 1 million AI workers
✨ And more AI goodness…
Dear Readers,
Today's issue kicks off with Alibaba dropping a bombshell - their new Qwen3-TTS clones any voice from just three seconds of audio, and it's fully open-source, fundamentally shifting the economics beneath companies like ElevenLabs.
Meanwhile, Anthropic's Dario Amodei is at Davos comparing US chip sales to China to "selling nuclear weapons to North Korea" while his own company watches revenue hit $4.5 billion even as margins slip from runaway inference costs.
Add in xAI testing AI workers so convincing employees can't tell they're bots, OpenAI eyeing royalty cuts from drug discoveries, and Google's Gemini API usage doubling in five months - scroll on.…
All the best,
🤖 AI Employees Hit Reality Checks
xAI is testing “human emulator” AI workers that act like real employees—so convincingly that staff sometimes don’t realize they’re chatting with bots, leading to awkward hallucination-driven mix-ups and internal confusion. Former engineer Sulaiman Ghori says scaling this vision to up to one million AI workers will require massive compute, possibly even leasing power from idle Tesla cars, underscoring how manual and messy white-collar automation still is at xAI. Meanwhile, Google is seeing Gemini API usage roughly double in five months, boosting developer momentum - even as enterprise customers struggle with customization.
💰 OpenAI Eyes Royalties, Not Just Ads
OpenAI CFO Sarah Friar says the company wants “value-sharing” deals - taking a cut of profits or IP from discoveries customers make using its AI, starting with drug discovery and potentially expanding to energy and finance. The move signals OpenAI is looking beyond ads toward licensing and outcome-based pricing, joining rivals like Anthropic, Google DeepMind, and Isomorphic Labs in the race to monetize AI-driven breakthroughs. If it works, royalties could one day rival chatbot revenue - but only if OpenAI can prove its models consistently deliver real scientific wins.
🤖 Anthropic Revenue Soars, Margins Slip
Anthropic slashed its 2025 gross margin outlook to 40% as inference costs came in 23% higher than expected, even while revenue is projected to hit $4.5 billion, nearly 12× growth year over year. The AI startup, led by Dario Amodei, is racing to control compute costs through hardware deals as it competes with OpenAI, which expects higher margins but far larger cash burn. Bottom line: AI demand is exploding, but infrastructure economics remain the biggest battleground.
Full conversation with Elon Musk at the World Economic Forum Annual Meeting 2026
Alibaba's 3-Second Voice Cloning Goes Open-Source
The Takeaway
👉 Alibaba released Qwen3-TTS as fully open-source (Apache 2.0), enabling commercial use of voice cloning, voice design, and multilingual speech synthesis without licensing fees
👉 3-second voice cloning works across 10 languages — clone a voice from minimal audio and generate speech in Chinese, English, Japanese, Korean, and six other languages while preserving speaker characteristics
👉 Natural language voice design eliminates technical barriers — describe the voice you want in plain text instead of tweaking SSML tags or acoustic parameters
👉 Performance benchmarks show lower word error rates than ElevenLabs and MiniMax, with streaming latency under 100ms for real-time applications
The race for human-like AI voices just got a massive shakeup. Alibaba's Qwen team dropped the open-source weights for Qwen3-TTS this week, and the capabilities are genuinely impressive: clone any voice from just 3 seconds of audio, design entirely new voices through simple text descriptions, and generate speech across 10 languages with near-human naturalness.
What makes this release stand out? The model lets you describe a voice in plain language — - say, "a warm, energetic narrator with a slight regional accent" - and it creates exactly that. No more fiddling with obscure parameters. The 3-second voice cloning works across all supported languages, meaning you could clone a voice in English and have it speak fluent Japanese while maintaining the original speaker's characteristics.
The technical specs are solid: 0.6B and 1.7B parameter versions, end-to-end streaming generation with latency as low as 97ms, and word error rates that reportedly beat ElevenLabs and MiniMax on multilingual benchmarks. Everything's released under Apache 2.0, making it fully usable for commercial applications.
Why it matters: Open-source TTS reaching commercial-grade quality shifts the economics of voice content creation entirely. This puts powerful voice synthesis tools into every developer's hands without licensing fees or API costs.
Sources:
🔗 https://github.com/QwenLM/Qwen3-TTS
Hiring in 8 countries shouldn't require 8 different processes
This guide from Deel breaks down how to build one global hiring system. You’ll learn about assessment frameworks that scale, how to do headcount planning across regions, and even intake processes that work everywhere. As HR pros know, hiring in one country is hard enough. So let this free global hiring guide give you the tools you need to avoid global hiring headaches.
Amodei Calls China Chip Sales "Crazy"
Anthropic's CEO just made headlines at Davos with some seriously pointed remarks. Dario Amodei compared selling advanced AI chips to China to "selling nuclear weapons to North Korea”.
The context: The Trump administration is loosening export restrictions to let Nvidia sell H200 chips to Beijing. For Nvidia, it's about market access and the argument that China will simply build alternatives anyway. For Amodei, it's a matter of "incredible national security implications.”
What makes this interesting is the timing. While companies like Nvidia and AMD push for broader China access, AI lab leaders are sounding alarms. Amodei isn't new to this position - he urged chip restrictions last year at Davos too, warning about dystopian "1984 scenarios" from unchecked AI proliferation.
The H200, while not Nvidia's latest, would be the most advanced chip legally exportable to China. Meanwhile, Nvidia's cutting-edge Blackwell chips and the upcoming Rubin generation remain restricted.
The AI chip debate isn't just about trade policy. It's a proxy war over who controls the most consequential technology of the decade, with AI lab founders and chipmakers increasingly on opposite sides.
All the news that matters to your career & life
Hyper-relevant news. Bite-sized stories. Written with personality. And games that’ll keep you coming back.
Morning Brew is the go-to newsletter for anyone who wants to stay on top of the world’s most pressing stories — in a quick, witty, and actually enjoyable way. If it impacts your career or life, you can bet it’s covered in the Brew — with a few puns sprinkled in to keep things interesting.
Join over 4 million people who read Morning Brew every day, and start your mornings with the news that matters most — minus the boring stuff.
