PLUS: NVIDIA's true moat and AI's workplace takeover
Good morning
Open-source AI models from Chinese firms are quietly being integrated into major US tech platforms. Companies like Pinterest and Airbnb are leveraging these models for their top-tier performance and significantly lower cost compared to proprietary US alternatives.
This trend marks a significant shift in the AI landscape, moving away from the dominance of closed systems from US labs. But as this new competition pushes the industry toward a more open future, what are the long-term strategic implications for the US tech sector?
In today’s Next in AI:
Chinese open-source AI powers US tech giants
NVIDIA's real moat is software, not silicon
AI's rapid growth in the American workplace
A developer’s take on AI coding tribalism
China's AI Goes Global
Next in AI: US tech giants like Pinterest and Airbnb are increasingly adopting open-source AI models from Chinese firms like DeepSeek and Alibaba. These models offer top-tier performance at a fraction of the cost of their US-based, proprietary counterparts.
Decoded:
Pinterest reports that training its own models using open-source techniques is 30% more accurate and up to 90% cheaper than using leading off-the-shelf alternatives.
This trend is shaking up the leaderboard, with Alibaba’s Qwen surpassing Meta as the most downloaded model on Hugging Face following an underwhelming Llama 4 release.
While US firms like OpenAI feel the pressure to monetize by turning to ads, China’s government support for open-source AI is accelerating its global adoption.
Why It Matters: The rise of high-performing, accessible AI from China marks a significant shift away from the closed-system dominance of US labs. This new competition is pushing the entire industry toward a more open and democratized future.
NVIDIA's Real Moat Is Software, Not Silicon
Next in AI: While new chips challenge NVIDIA on speed, the company's true defensible advantage isn't just its hardware. It's the massive CUDA software ecosystem built over two decades by more than four million developers.
Decoded:
The AI industry is rapidly shifting focus from training models to running them, with the inference market projected to triple to $100 billion by 2028.
In response to specialized competitors, NVIDIA's next-gen Blackwell architecture is designed to deliver a 30x improvement in inference performance, aiming to keep its hardware competitive.
The CUDA platform's deep integration into tools like PyTorch and TensorFlow creates immense switching costs, representing a more significant barrier to entry than simply building a faster chip.
Why It Matters: Losing some market share isn't a death sentence in a sector growing this quickly. The company's biggest long-term threat comes from its largest customers—hyperscalers like Google and Amazon—developing their own custom silicon.
AI in the American Workplace
Next in AI: A new Gallup poll reveals that AI is rapidly becoming a staple in the American workplace, with nearly a quarter of U.S. workers now using AI tools on a frequent basis to handle daily tasks.
Decoded:
AI usage isn't uniform, with the tech sector leading the charge—6 in 10 employees there use it frequently, followed by significant adoption in finance and education.
Professionals are using AI for practical tasks like drafting communications, synthesizing large documents, and generating ideas, turning to tools like ChatGPT and Google's Gemini to boost daily efficiency.
While many AI-exposed jobs are held by adaptable workers, research has identified 6.1 million workers in roles like administrative support who are more vulnerable to displacement.
Why It Matters: AI is no longer a niche tool; it's a mainstream productivity partner for a growing portion of the workforce. This rapid integration signals a fundamental shift in how daily work gets done, pushing professionals to adapt and leverage these tools or risk falling behind.
Developer AI Tribalism
Next in AI: A software developer details his journey from being a staunch AI skeptic to having AI author 90% of his code, arguing that developers should focus more on curiosity than picking a side in the increasingly polarized debate.
Decoded:
Lawson notes that a quality threshold was crossed in 2025, where AI tools became faster and more effective than manual coding for many tasks. He credits his shift to being able to work with Claude in plan mode to write a spec and have the model handle the busywork.
He highlights an emerging trend of using AI to fix AI, where systems like Cursor Bugbot works its magic by finding and suggesting fixes for bugs in AI-generated code. This creates a powerful feedback loop where multiple agents build on each other's work.
The author's core message is a call for curiosity and experimentation over dogma. He observes many developers are "burying their heads in the sand" and argues that nobody truly knows where this is heading, making hands-on tinkering the most valuable approach right now.
Why It Matters: The practical integration of AI into coding workflows is now moving faster than the debates about its potential. For developers, the most insightful commentary comes not from think-pieces, but from firsthand experience building with these new tools.
The Shortlist
UCLA developed an AI tool that analyzes medical records to identify previously undiagnosed cases of Alzheimer's disease, capturing 80% of potential cases and aiming to reduce health disparities.
Translators are facing significant disruption from AI, with a survey revealing over a third have lost work and many are now asked to post-edit machine translations, a practice one translator called "digging your own professional grave."
An AI-generated character named 'Amelia', originally created for a UK government-funded anti-extremism game, has been co-opted and turned into a viral far-right meme across social media.