StartupFortune

Wall Street finally caught up to what DDR5 spot prices have been screaming for months: memory is the AI boom's new bottleneck, and Micron is the easiest way to own it. Morgan Stanley upgraded Micron to Overweight on October 6 and lifted its price target to $220 from $160, pointing to DRAM pricing that keeps beating even bullish forecasts.

J.P. Morgan helped shape it. A top-five U.S. bank spent months teaching a public blockchain its own playbook for how institutional trades should settle, closing the days-long gap where a deal can still collapse after both sides agree. The Solana Foundation on October 6 released Solana DvP, an open-source escrow program that gives banks and asset managers a standardized way to settle tokenized trades atomically on-chain, built with input from J.P. Morgan, according to CoinDesk and Decrypt.

Alibaba's DAMO Academy built an AI agent that screened 2.4 million crystal structures in 28 GPU hours and walked away with four real, lab-confirmed superconductors. That speed is the actual story. On July 3, 2026, Alibaba's DAMO Academy unveiled Elements Claw, an AI agent built with Renmin University of China and the University of Chinese Academy of Sciences to hunt for new superconducting materials.

Seven rounds of training. Nearly 1,000 Elo points. A model a thousand times smaller than frontier AI systems now plays chess above 2,600, and it hadn't plateaued when the researchers pulled the plug. Princeton researchers Adithya Bhaskar, Jeffrey Cheng and Danqi Chen built a 4-billion-parameter language model that plays chess at a strong Grandmaster level and, unlike a traditional engine, can tell you why it picked a move.

GLM-5.3 is now one of the models you can call inside Amazon Bedrock, giving a Chinese open-weight system the one thing it couldn't buy on its own: a front door inside the cloud platform most American enterprises already trust with their data. AWS said on October 5 that it had added Z.ai's GLM-5.3 to Amazon Bedrock, giving enterprise customers managed access to a 744-billion-parameter mixture-of-experts model built by Zhipu AI, one of China's best-funded AI labs.

Marvell's stock has nearly tripled this year on the bet that hyperscalers want an alternative to Nvidia and Broadcom for custom AI silicon. Today in New York, the company has to show the math behind that bet actually works. Matt Murphy walked onto a stage in New York on Tuesday carrying a $200 billion market cap and a stock that's climbed more than 160% in 2026.

Four world-model startups, one backer. Nvidia's venture arm keeps writing checks across the category instead of picking a single winner. Reactor announced on October 5 that NVentures, Nvidia's venture arm, and Sapphire Ventures had joined its Series A, pushing the young startup's total raised to $74 million. The San Francisco company had only come out of stealth in May with a $59 million round led by Lightspeed Venture Partners, backed by WndrCo, Amplify Partners, Sky9 Capital and FPV Ventures.

Anthropic's answer to a flaw that lets malicious MCP servers run arbitrary commands: working as intended. OX Security puts the exposure at 200,000 vulnerable instances tied to a supply chain of 150 million-plus downloads. Should a tool-calling standard let a subprocess run whatever commands it wants, unsanitized?

Vercel and Render both added spend caps and usage alerts after AI compute spikes blindsided customers' bills. Pricing teams rarely see the resulting churn coming.

Wikipedia's parent foundation just told the world that AI agents running wild cost it four days of degraded service, and it's pointing squarely at OpenAI. The Wikimedia Foundation disclosed on October 5 that an internal investigation had tied "rogue" OpenAI agent activity to a partial outage of its Wikidata Query Service that ran from May 7 to May 11.