A heavy stream of crypto and AI developments landed between July 23 and July 24, led by BitMEX’s decision to shut down on Sept. 23 and the collapse in its exchange token BMEX, which market data showed fell 99.69% at one point as liquidity tightened. Arthur Hayes later said he was deeply proud of what had been built, while Bubblemaps said 75% of BMEX allocations had never circulated on-chain. Vanar also unveiled a major strategic reset, saying it will move to Base, focus on an AI Organizations economy, migrate existing VANRY holders on a 1:1 basis and lift total supply from 2.4 billion to 10 billion tokens, with 62% still locked during the transition and new allocations vesting over 60 months after a cliff period. Institutional tokenization also moved forward. Mubadala Capital said it is working with Coinbase and KAIO on a blockchain-native version of its long-term private equity fund for qualified investors, with Coinbase set to use Base as one of the token networks and to purchase the token for its own balance sheet. Michael Saylor separately announced the launch of the Bitcoin Security Consortium, backed by a three-year $15 million commitment and founding members including Anchorage Digital, ARK Invest, BlackRock, Block, Blockstream, Coinbase, Fidelity Digital Assets, Galaxy and Strategy. Elsewhere, Ondo Finance said Oasis Pro Markets had received authorization to offer regulated tokenized securities services in the U.S., Uniswap introduced permissioned pools on v4, and a long list of market, policy, security and AI infrastructure updates continued to reshape the day’s flow.

BitMEX said it will shut down on Sept. 23, 2026 at 12:00 p.m. Beijing time after halting new user registrations immediately, with new positions to be restricted from Aug. 26 and remaining positions forcibly closed after the deadline. In China, regulators in Shenzhen said a group of self-media accounts that promoted virtual currency-related business activities to domestic users had been permanently shut down under existing policy rules. The day’s roundup also included fresh market data, with BTC, ETH, SOL, XRP and DOGE all down over the past 24 hours, while tokens such as RE, VELO and WLFI led gainers on OKX data. Macro attention centered on U.S. initial jobless claims, which came in at 187,000 for the week ended July 18, below the 212,000 expectation and marking the lowest reading since the week of Sept. 24, 2022. Elsewhere, the European Union announced sanctions covering more than 100 banks and crypto operators, Argentina moved toward allowing mutual funds to invest in bitcoin and crypto, Grayscale’s Worldcoin ETF filing disclosed concentration and governance details around WLD, and several funding deals and industry statements added to the day’s developments.

Unitree Robotics filed for a $620 million IPO on Shanghai's STAR Market. The prospectus shows 5,500 humanoid robots shipped in 2025—world No.1—but 74% went to universities and research labs, with industrial use at only 9%. Gross margin nears 60% thanks to vertical integration.

A team led by Professor Lei Yinjie at Sichuan University has introduced PolicyTrim, a two-stage post-training framework designed to improve the deployment efficiency of Vision-Language-Action, or VLA, robots without changing model architecture or recollecting expert data. The method focuses on what the paper calls policy efficiency: how many actions in a predicted chunk can be executed reliably, and how many real-world physical steps are needed to finish a task. In the first stage, PolicyTrim expands the reliable execution horizon of action chunks through dynamic execution horizon exploration. In the second, it reduces redundant physical steps with a reward tied to shorter successful trajectories, while using group-anchored regularization to avoid brittle shortcuts. The paper reports tests across LIBERO, ManiSkill, Meta-World and real-world robot tasks, covering architectures including π0.5, OpenVLA-OFT and GR00T. Results cited in the paper include a 3x increase in action chunk utilization, a 51.4% reduction in physical steps, and a peak 5.83x end-to-end speedup for π0.5 on LIBERO while keeping success rate above 98%.

WuBlockchain’s WhiteLine Daily framed the day’s market narrative around a single question: what ultimately justifies a high valuation? The report centered on Changxin Technology’s listing and the commercialization path for embodied AI companies highlighted at WAIC 2026. Changxin is set to list at 8.66 yuan per share, implying a valuation of about 579.2 billion yuan and base fundraising of roughly 57.9 billion yuan. Online subscriptions reached 243.93 times, with a final allotment rate of 0.4714%, and the company is expected to begin trading on July 27. The report argued that the real debate is not the scarcity of a domestic DRAM asset, but which year’s earnings investors should use to value the company. Based on 2025 profit, the implied P/E is above 300x and price-to-book is about 5x. But first-quarter 2026 net profit reached 33 billion yuan, with first-half net profit projected at 66 billion to 75 billion yuan. WhiteLine Daily also pointed to WAIC 2026, where the benchmark for robots has shifted from flashy demonstrations to full-task execution, stable operation, and repeatable deployment, using AgiBot and MagicLab as examples of that shift.

Robinhood Chain’s early meme-token launch market is already shifting. NOXA, one of the chain’s first breakout launchpads, paused new token issuance on July 11, later moved users to a static interface, and on July 15 said it would stop taking future trading fees and pass all trading revenue to creators. Its rise had been fast, driven by early traction around CASHCAT and added visibility from Robinhood’s attention to native projects. Its retreat has been just as fast. CoinW Research argues the market has now entered a high-supply, low-conversion phase. Dune data cited in the report shows 42,709 new tokens were created on Robinhood Chain on July 16 alone, with Pons.family and Flap accounting for 50.30% of that total. Yet only 18 tokens across the chain had market capitalizations above $1 million at the time of writing, and most of those came from NOXA and Virtuals rather than the newest issuance leaders. The report says the handoff in launch volume has not produced a new dominant platform. Pons leads in issuance, but the report points to suspected bot-driven activity and a raw graduation rate of about 0.51%. Flap has scaled distribution quickly, though it had no $1 million tokens in the cited data set. Meanwhile, Uniswap may be the clearer downstream beneficiary as more launchpads route graduated liquidity into public pools, even if value capture for UNI still depends on protocol-fee activation on Robinhood Chain.

Walden Robotics, a startup spun out of Toyota Research Institute in January, emerged from stealth on July 15 with a $300 million seed round and an $1.1 billion valuation. The financing was co-led by Toyota and Deviation Capital, with participation from NVIDIA, Boeing, Samsung Ventures, Prologis Ventures and CoreWeave Ventures. The company reached unicorn status in about six months. Led by co-founder and CEO Russ Tedrake, a Massachusetts Institute of Technology professor who previously led TRI’s robotics and machine learning team for nearly a decade, Walden is built on research that includes Diffusion Policy, Universal Manipulation Interface, Large Behavior Models, OpenVLA and the open-source simulator Drake. Its founding team also includes CTO Ben Burchfiel, COO Kerri Fetzer-Borelli, Chief Product Officer Dave Johnson, Chief Strategy Officer Adrien Gaidon, Chief Architect Siyuan Feng and AI lead Rares Ambrus. Walden’s pitch centers on commercializing TRI’s robotics work inside real production environments, starting with Toyota’s manufacturing system. The company says its Large Behavior Models are designed to let robots learn and transfer skills across tasks instead of being programmed one workflow at a time. TRI’s prior research used nearly 1,700 hours of robot data, 1,800 real-world tests and more than 47,000 simulation tests. Walden is combining autonomous operation with remote human assistance as it pushes those systems into factories and warehouses.

Embodied AI company Xingdong Jiyuan announced on July 6 that it had secured a new RMB 1 billion financing round led by Chengtong Fund, with participation from multiple state-backed investors, financial institutions, industrial backers, and existing shareholders. The latest deal lifts the company’s fundraising over the past two months to RMB 2.5 billion, underscoring strong investor appetite for embodied intelligence platforms that can combine model development, hardware integration, and real-world deployment. According to the company, its core edge lies in a full-stack strategy spanning data, foundation models, motion control, dexterous hands, and robot bodies. Xingdong Jiyuan has positioned world models such as PAD, Ctrl-World, and VLAW at the center of that stack, while using self-developed dexterous hands as a key source of high-quality physical interaction data. It says its dataset now includes more than 12 million teleoperation clips, with over 1.5 million clips from dexterous-hand teleoperation, covering more than 100 real-world scenarios and 1,000-plus manipulation tasks. On commercialization, the company said it has already reached scaled deployment in logistics, working with clients including SF Express and China Post across more than 10 logistics centers in five provincial-level regions. It is also expanding into high-end manufacturing and commercial service use cases with partners such as Samsung, Lenovo, Haier, and Geely. The latest financing highlights how capital markets are increasingly prioritizing deployable embodied AI systems over concept-stage narratives.