Nvidia scales back funding guarantee for Ohio OpenAI data center, WSJ reports ↗
Nvidia is reportedly shrinking its financial backstop for OpenAI’s huge Ohio data-center project to less than $120 billion, down from the previously discussed $250 billion. The guarantee would initially cover only the project’s first phase.
Investors had apparently grown nervous about just how much financing risk Nvidia was taking on. OpenAI is still discussing a binding lease for the full 10-gigawatt site - so the mega-build isn’t dead, just looking a little less financially bonkers. (Reuters)
Apple trains its own AI model for China market with Alibaba's support, sources say ↗
Apple has reportedly trained a China-specific large language model with support from Alibaba, moving beyond its earlier strategy of leaning primarily on local third-party models.
The approach could give Apple considerably more control over AI features on Chinese devices while navigating Beijing’s regulatory requirements. It’s an intriguing dual-track setup - Apple gets its own model while still weaving Alibaba technology into the wider Apple Intelligence offering. (Reuters)
US to tell partners they must pick sides in AI race with China ↗
Washington is preparing to tell dozens of countries that participation in its AI coalition may be incompatible with joining China’s competing framework. A draft message targets 35 countries aligned with the US-led effort around AI models, semiconductors and critical minerals.
Kazakhstan is already caught in the middle after joining initiatives connected to both sides. AI geopolitics is starting to look less like ordinary tech competition and more like choosing which railway track your entire digital economy runs on... slightly dramatic, except apparently not. (Reuters)
Republican senator urges Trump to back US open-weight AI models amid industry debate ↗
Senator Jim Banks wants the US government to create incentives for American companies building open-weight AI models, while reducing dependence on increasingly popular Chinese alternatives.
That drops Washington straight into an industry argument. Meta and Nvidia have pushed for keeping open models competitive, while Anthropic CEO Dario Amodei has warned that releasing powerful model weights can create security risks that are harder to monitor. (Reuters)
Goldman in talks with investors on Nvidia financing deal after landing prized role, sources say ↗
Goldman Sachs is sounding out banks, insurers, asset managers and other investors for Nvidia’s enormous $500 billion AI infrastructure financing initiative.
Nvidia has said it could backstop up to $125 billion of potential deals, while Goldman can provide private credit, junior capital and debt placement. Basically, AI compute is beginning to look like its own Wall Street asset class - GPUs meet structured finance, and the complications practically write themselves. (Reuters)
Google will now allow users to remove visible watermark from its AI generations ↗
Google is making visible watermarks optional on AI-generated images, video and music produced through models including Nano Banana, Omni and Lyria.
The important bit: Google isn’t removing the invisible provenance layer. SynthID watermarks and C2PA-related metadata remain, meaning generated media can still carry machine-readable signals even when the obvious visual label disappears. Surprisingly practical... and inevitably controversial. (TechCrunch)
FAQ
Why is Nvidia reducing its financial guarantee for the OpenAI Ohio data center?
Nvidia is reportedly reducing its financial backstop to less than $120 billion, down from the previously discussed $250 billion. The smaller guarantee would initially cover only the first phase of the project. The change appears intended to limit Nvidia’s financing exposure while OpenAI continues discussions over a binding lease for the full 10-gigawatt site.
What is Apple’s China-specific AI model and how is Alibaba involved?
Apple has reportedly trained its own large language model specifically for the Chinese market, with support from Alibaba. This could give Apple greater control over the AI features offered on devices in China while helping the company operate within local regulatory requirements. Alibaba technology is also expected to remain part of the broader Apple Intelligence approach in the country.
How could the US-China AI race affect other countries?
The US is preparing to tell countries participating in its AI coalition that joining China’s competing framework may be incompatible with continued participation. The dispute spans strategically important areas, including AI models, semiconductors and critical minerals. Countries working with both sides, such as Kazakhstan, could therefore face growing pressure to choose between competing technology ecosystems.
Why are open-weight AI models becoming a US policy issue?
Open-weight AI models are becoming politically significant as US policymakers consider how to keep American models competitive with increasingly popular Chinese alternatives. Senator Jim Banks has called for incentives to support US developers. The debate also involves security concerns, since releasing powerful model weights can make models easier to distribute and modify while potentially making their use more difficult to monitor.
How is Goldman Sachs involved in Nvidia’s AI infrastructure financing plans?
Goldman Sachs is reportedly speaking with banks, insurers, asset managers and other investors about Nvidia’s $500 billion AI infrastructure financing initiative. Nvidia has said it could backstop up to $125 billion of potential transactions. Goldman’s role could include private credit, junior capital and debt placement, illustrating how large-scale AI computing projects are increasingly relying on sophisticated infrastructure financing structures.
Can Google AI images still be identified after visible watermarks are removed?
Google is making visible watermarks optional for AI-generated images, video and music created with models including Nano Banana, Omni and Lyria. Removing the visible mark, however, does not necessarily remove provenance information. Google is retaining technologies such as SynthID and C2PA-related metadata, which can provide machine-readable signals indicating that media was generated or processed using AI.