As Chinese tech firms moved from counting tokens burned to measuring AI returns, employees started performing efficiency rather than achieving it. Researchers and workers describe a year of AI FOMO in which looking productive became safer than being honest about what the tools actually do.
NVIDIA's Earth-2 open models are being deployed across Europe for real-world weather and climate prediction, with banks and payment firms also showing production-scale AI built on accelerated computing and open-weight models.
Elon Musk says SpaceX raised $40 billion in debt to buy GPUs arriving in the first half of 2027, a sign AI build-outs now lean on borrowed money. In the same breath he warned that more than 200 neoclouds face consolidation as debt markets turn harsh.
At its 1024 Developer Festival in Beijing, iFlytek unveiled the Spark X2.6 model, a Token-as-a-Service platform and a new voice model, pitching AI capacity as everyday industrial infrastructure rather than a demo.
Nvidia chief executive Jensen Huang has dismissed warnings that AI will destroy humanity as fabricated hype built for clicks. His remarks land as AI safety scares multiply, from jailbroken frontier models to AI-driven cyberattacks on small firms.
At an internal staff meeting, OpenAI chief Sam Altman laid out three possible endings to the global AI race, including one where a breakthrough lets a company cripple rivals' hardware. The options range from a global regulator to a merger of top labs.
DeepSeek released DeepInfer, the inference framework that runs its large-scale chat service, under the Apache 2.0 license, plus a companion token-encoding toolkit. It is the company's latest move to open up its full stack, from models to the software that serves them.
Tencent is considering its first major offshore bond in years, a dual-currency deal worth up to $5 billion, as China's tech giants race to bankroll AI computing. The cash-rich gaming and cloud group is stockpiling long-term capital before the spending peaks.
You've reached the end
