Keep pulling the thread on Nathan Benaich.
The U.S. administration's move to restrict Chinese access to American AI technology was partly motivated by allegations that Chinese companies were training their models on the outputs of U.S. models.
The Chinese company DeepSeek released a base model that, when tuned for reasoning, performed so well it caused a "freakout" in the AI community.
Alibaba's Qwen model has effectively taken over the leadership mantle in open-source AI from Meta's Llama initiative.
Chinese AI model companies Minimax and Zhipu AI were the first pure-play large model companies to go public, listing on the Hong Kong exchange for several billion dollars each.
No open-source model has managed to achieve the same level of performance as top-tier closed-source models like Anthropic's Opus or OpenAI's GPT-4.5.
The dynamic where application-layer companies like Jasper generated more revenue than model providers like OpenAI has flipped, as model vendors now create their own SaaS wrappers and generate tens of billions of dollars in revenue.
The concept of "AI sovereignty," where nation-states seek control over their own compute and models, has become a major marketing agenda aggressively promoted by NVIDIA.
NVIDIA's promotion of "AI sovereignty" has resulted in global commitments of over $100 billion from various countries to build up their domestic data center capacity.
Nathan Benaich does not believe the AI sector is in a bubble, citing that the "Magnificent Seven" tech companies are highly profitable and trade at forward valuations that are 50% of top tech names during the late 1990s dot-com bubble.
Since the launch of ChatGPT, OpenAI and Anthropic have added more new revenue than all other publicly traded software companies combined.
Some of the best-performing AI model makers are achieving gross margins of 60-80% on serving their models.
ElevenLabs reportedly reached $330 million in revenue and added $100 million in net new ARR in just five months.