JustSayAIWeeklyReport
AI Industry Shifts From Open Source to Domestic LLMs and Traffic Control
This issue,
what to read.
The timeline on the left mirrors these chapters — read in order, or jump straight to any story.
AI Industry Shifts From Open Source to Domestic LLMs and Traffic Control
The AI race is accelerating away from open-source idealism toward domestic supply-chain self-sufficiency and refined commercial operations. As Gartner projects that half of China's AI accelerators will be domestically produced by 2030, and as DeepSeek deploys price hikes as a traffic throttle while reopening fundraising, even Demis Hassabis at the helm of Google DeepMind has been rumored to be considering departure—signaling that across chips, models, and top-tier talent alike, the global AI landscape is undergoing a violent reordering.
- 01AI industry shifts from open to closed, everyone building their own walled gardens.
- 02The second half is zero-sum: whoever retains traffic, supply chain, and talent wins.
- 03Industry still growing, but smart money's hedging bets.
“This week I had this really strong feeling—that era of open-source sharing and "we all win together" in AI? It's over.”
- DeepSeek's subsidized API pricing became unsustainable amid surging users and compute strain.
- Announcing price hikes amid $8B funding recovery to prove pricing power, not cash-burning for market share.
- Peak pricing reserves scarce compute for paying users, transitioning from subsidies to maturity.
“Right, and there's a detail in that Gartner forecast worth chewing on — it specifically named Kimi K3, GLM-5, DeepSeek V4, those open-source models, not the closed-source ones.”
- Gartner predicts 50% China AI accelerators domestic by 2030, names Kimi K3, GLM-5, DeepSeek V4 open-source LLMs.
- Open-source LLM weights enable flexible deployment, making them the most practical scenario for domestic chip volume growth and ecosystem support.
- Localization targets "good enough" inference and fine-tuning cost parity, not Nvidia-beating performance.
“Back to the US—if this Demis Hassabis leaving Google DeepMind thing actually happens, that's a whole different ball game, way bigger than DeepSeek raising prices.”
- Jeff Dean and Demis Hassabis depart or shift roles, Alphabet faces dual leadership vacuum in infrastructure and AI flagship.
- Demis departure unconfirmed but well-sourced; if true, severely damages Google AI.
- Key departures trigger spiraling brain drain; no one can match both research depth and productization short-term.
This week I had this really strong feeling—that era of open-source sharing and "we all win together" in AI? It's over.
- 01DeepSeek's subsidized API pricing became unsustainable amid surging users and compute strain.
- 02Announcing price hikes amid $8B funding recovery to prove pricing power, not cash-burning for market share.
- 03Peak pricing reserves scarce compute for paying users, transitioning from subsidies to maturity.
Source ↗ smartkarma.com
Right, and there's a detail in that Gartner forecast worth chewing on — it specifically named Kimi K3, GLM-5, DeepSeek V4, those open-source models, not the closed-source ones.
- 01Gartner predicts 50% China AI accelerators domestic by 2030, names Kimi K3, GLM-5, DeepSeek V4 open-source LLMs.
- 02Open-source LLM weights enable flexible deployment, making them the most practical scenario for domestic chip volume growth and ecosystem support.
- 03Localization targets "good enough" inference and fine-tuning cost parity, not Nvidia-beating performance.
Source ↗ finance.biggo.com
Back to the US—if this Demis Hassabis leaving Google DeepMind thing actually happens, that's a whole different ball game, way bigger than DeepSeek raising prices.
- 01Jeff Dean and Demis Hassabis depart or shift roles, Alphabet faces dual leadership vacuum in infrastructure and AI flagship.
- 02Demis departure unconfirmed but well-sourced; if true, severely damages Google AI.
- 03Key departures trigger spiraling brain drain; no one can match both research depth and productization short-term.
Source ↗ telegraphindia.com
The full week — every daily brief's headline, linked to its issue:
08.04This week ran 8 headlines; 3 made the main thread; 13 daily briefs.
“AI Industry Shifts From Open Source to Domestic LLMs and Traffic Control”
Two issues a day. Ten minutes to turn AI noise into judgment — mornings for the world, evenings for China.
Two issues a day — AI noise into judgment.