TD Cowen maintained a buy rating on Nvidia despite Google's new AI chips and supply chain issues including Super Micro's lost Oracle contract and B200 inventory buildup. Nvidia announced new partnerships in the UK, China, and automotive sectors.
TD Cowen reaffirms Nvidia buy rating as new deals with BT and Desay SV emerge, while Super Micro loses a $1.1-1.4B Oracle contract and B200 GPU inventories pile up, highlighting growing pains.
Vitalik Buterin proposes Keyed Nonces to enhance Ethereum privacy and state scalability. The plan uses dedicated storage for nullifiers, bloom filters, and sharding to handle billions of nullifiers, preserving dynamic state for DeFi while reducing storage costs.
Vitalik Buterin said keyed nonces could do more than improve privacy on Ethereum. The concept may help shift some use cases into specialized storage, easing long-term state growth caused by nullifiers and supporting more scalable network design.
Vitalik Buterin said keyed nonces may become more than a privacy upgrade for Ethereum, potentially serving as a first step toward specialized state storage to address long-term nullifier growth and scaling challenges.
OPSEC crashed over 88% after OpSec said external attackers stole funds from its staking contract. The team plans a token migration, but community members and prior criticism have fueled doubts over whether the incident was a hack or something more serious.
During the 2008 financial crisis, Suzhou's state-owned investment platform Yuanhe Holdings injected approximately 30 million RMB ($3M) into XuChuang Technology, a fiber optic module startup founded by returning overseas Chinese Liu Sheng. Eighteen years later, after going public via reverse merger as Zhongji Innolight, the company's market cap hit 1.5 trillion RMB, making it one of the top 10 A-share firms. Optical modules are the 'blood vessels' of AI data centers—without them, AI chips are just hot silicon. Suzhou's bet turned into a complete supply chain from optical chips, devices, modules to test equipment, boosting the city's total A-share market cap above 4 trillion RMB. This article traces Suzhou's investment philosophy: 'squatting on the ground' to pick unsexy projects, backed by a system of technology recruitment centers, chain-led funds, and ultra-long-term capital. Now, Suzhou is leveraging its AI+Manufacturing action plan to push industrial output beyond 5 trillion RMB, turning its manufacturing base into a flywheel for the AI era.
Dwarkesh Patel argues that the next generation of AI training should move beyond reinforcement learning from verifiable rewards (RLVR) toward continuous learning in real-world tasks, with experience written back into model weights. He highlights three key paths: malleability (the ability to continuously integrate new experience post-deployment), on-policy self-distillation (OPSD), and "dreaming" simulations. This shift from pre-training to post-deployment evolution could profoundly impact decentralized AI and on-chain agent development.
