Export Controls AI News & Updates
Huawei Developing Advanced AI Chip to Compete with Nvidia's H100
Chinese tech company Huawei is making progress developing its new Ascend 910D AI chip, which aims to rival Nvidia's H100 series used for training AI models. This development comes shortly after increased US restrictions on AI chip exports to China and could help fill the resulting void in the Chinese AI market.
Skynet Chance (+0.04%): The development of advanced AI chips outside of US regulatory control increases the potential for divergent AI development paths with potentially fewer safety guardrails, while also making powerful AI training capabilities more widespread and harder to monitor globally.
Skynet Date (-1 days): Huawei's chip development could accelerate the timeline toward advanced AI risks by circumventing export controls intended to slow capabilities development, potentially creating parallel advancement tracks operating under different safety and governance frameworks.
AGI Progress (+0.03%): While the chip itself doesn't directly advance AI algorithms, the proliferation of computing hardware comparable to Nvidia's H100 expands the infrastructure foundation necessary for training increasingly powerful models that could approach AGI capabilities.
AGI Date (-1 days): By potentially breaking hardware bottlenecks in AI model training outside of US export controls, Huawei's chip could significantly accelerate the global pace of AGI development by providing alternative computing resources for large-scale model training.
Anthropic Proposes National AI Policy Framework to White House
After removing Biden-era AI commitments from its website, Anthropic submitted recommendations to the White House for a national AI policy focused on economic benefits. The recommendations include maintaining the AI Safety Institute, developing national security evaluations for powerful AI models, implementing chip export controls, and establishing a 50-gigawatt power target for AI data centers by 2027.
Skynet Chance (-0.08%): Anthropic's recommendations prioritize national security evaluations and maintaining safety institutions, which could reduce potential uncontrolled AI risks. The focus on governance structures and security vulnerability analysis represents a moderate push toward greater oversight of powerful AI systems.
Skynet Date (+1 days): The proposed policies would likely slow deployment through additional security requirements and evaluations, moderately decelerating paths to potentially dangerous AI capabilities. Continued institutional oversight creates friction against rapid, unchecked AI development.
AGI Progress (+0.01%): While focusing mainly on governance rather than capabilities, Anthropic's recommendation for 50 additional gigawatts of power dedicated to AI by 2027 would significantly increase compute resources. This infrastructure expansion could moderately accelerate overall progress toward advanced AI systems.
AGI Date (+0 days): The massive power infrastructure proposal (50GW by 2027) would substantially increase AI computing capacity in the US, potentially accelerating AGI development timelines. However, this is partially offset by the proposed regulatory mechanisms that might introduce some delays.
Anthropic CEO Calls for Stronger AI Export Controls Against China
Anthropic's CEO Dario Amodei argues that U.S. export controls on AI chips are effectively slowing Chinese AI progress, noting that DeepSeek's models match U.S. models from 7-10 months earlier but don't represent a fundamental breakthrough. Amodei advocates for strengthening export restrictions to prevent China from obtaining millions of chips for AI development, warning that without such controls, China could redirect resources toward military AI applications.
Skynet Chance (+0.03%): Amodei's advocacy for limiting advanced AI development capabilities in countries with different value systems could reduce risks of misaligned AI being developed without adequate safety protocols, though his focus appears more on preventing military applications than on existential risks from advanced AI.
Skynet Date (+1 days): Stronger export controls advocated by Amodei could significantly slow the global proliferation of advanced AI capabilities, potentially extending timelines for high-risk AI development by constraining access to the computational resources necessary for training frontier models.
AGI Progress (-0.01%): While the article mainly discusses policy rather than technical breakthroughs, Amodei's analysis suggests DeepSeek's models represent expected efficiency improvements rather than fundamental advances, implying current AGI progress is following predictable trajectories rather than accelerating unexpectedly.
AGI Date (+1 days): The potential strengthening of export controls advocated by Amodei and apparently supported by Trump's commerce secretary nominee could moderately slow global AGI development by restricting computational resources available to some major AI developers, extending timelines for achieving AGI capabilities.