Alibaba's video AI just hit number one on day one

Alibaba's new video generation model landed at number one on a major global benchmark on its first day out, per Bloomberg - beating OpenAI's Sora and Google's Veo in head-to-head quality rankings.
Why it matters: The "Chinese AI labs are behind" narrative just took a serious hit, and the generative video race - which is shaping up to be the next major commercial battleground - now has a new frontrunner you probably weren't watching. (source)

Anthropic is exploring custom chip development, per Reuters, to reduce its dependence on Nvidia and Google TPUs - following the same playbook already run by Google, Meta, and Amazon.
Why it matters: When even the safety-first AI lab decides hardware control is a strategic necessity, it confirms that owning your compute stack isn't a luxury anymore - it's existential, and Nvidia's moat is getting crowded with challengers. (source)

Google and Intel are expanding their AI infrastructure partnership, with Intel Gaudi accelerators and Intel's foundry services taking a larger role in Google's compute stack as both companies push back against Nvidia's market dominance.
Why it matters: Google diversifying its chip supply chain - and Intel landing a marquee AI customer - is exactly the kind of competitive pressure that drives down prices and speeds up innovation across the whole industry. (source)

SiFive, a RISC-V chip design company backed by Nvidia itself, closed its latest round at a $3.65 billion valuation - building open-architecture AI chips as an alternative to proprietary ARM and x86 ecosystems.
Why it matters: The fact that Nvidia is bankrolling a company that could eventually compete with its own hardware tells you everything - open silicon is no longer a hobbyist bet, it's a serious hedge on the next wave of AI infrastructure. (source)

AI systems are rapidly gaining persistent memory - retaining context, preferences, and history across sessions - but per PitchBook, the industry has not settled on whether that data belongs to the user, the platform, or becomes a proprietary asset the company monetizes.
Why it matters: Memory is what makes AI genuinely useful over time, but it also makes switching platforms painful - and whoever controls your AI's memory controls your habits, your history, and your lock-in. (source)

Pony.ai launched PonyWorld 2.0, a self-improving autonomous driving engine that generates synthetic training scenarios and updates its own models without requiring new real-world data collection - backed by real validation data from its active robotaxi operations in China and the US.
Why it matters: A self-improving simulation loop could cut the cost and timeline of autonomous vehicle development dramatically, and if PonyWorld 2.0 delivers, it reframes how the entire industry thinks about training at scale. (source)

That's a wrap for today. The AI world doesn't sleep, and neither does this newsletter.

Hit reply and tell us which story surprised you most - we actually read every one.

Ready to try every AI model in one place? One subscription gives you access to ChatGPT, Claude, Flux, Sora, ElevenLabs, and 20+ more.

Like what you read?