8/13: DeepSeek-V4 Pro Official Release

It’s Thursday. Between a founder building autonomous ocean vehicles, one helping make SF YIMBY again, the CEO of Grindr, one of OpenAI’s most internet-famous researchers, and more, today was a fantastic day for guests. Be sure to monitor with us live on X and YouTube, and follow us on Instagram.

We like to joke on MTS that our favorite company is China’s largest memory manufacturer, ChangXin Memory Technologies (长鑫存储). Today, CXMT surpassed Tencent to become the most valuable company in China, with a market cap of $555 billion. So how did we get here?

CXMT was founded in 2016, less than a decade ago, in the Tier 3 city of Hefei, Anhui Province, by an engineer named Zhu Yiming. Zhu was born in Jiangsu Province in 1972. He studied physics at Tsinghua University, one of China’s top colleges, and became deeply interested in semiconductor design and manufacturing. At the time, the industry was deeply centered in the US, and so Zhu moved overseas in 1994 to pursue a physics PhD at Stony Brook University on Long Island. Not wanting to miss out on the 90s tech boom in Silicon Valley, he left Stony Brook early (after getting a master’s in electronics engineering) to work in the memory industry.

In 2005, frustrated at his company undervaluing his ideas on chip design, he left to found GigaDevice, which built NOR flash memory and microcontrollers. Zhu saw the future of the memory industry moving to Asia, so he moved himself and his company to Beijing. GigaDevice became one of China’s leading semiconductor companies, and went public in 2016. Today, it’s worth over $40 billion.

The same year GigaDevice went public, Zhu partnered with the municipal government of Hefei on a new project: building China’s first domestic manufacturer of DRAM, dynamic random-access memory. At the time, the global DRAM market was concentrated almost entirely into three companies: South Korea-based Samsung and SK Hynix, and US-based Micron. GigaDevice had no cell design or manufacturing process for DRAM, so they had to look elsewhere: to Qimonda, a defunct German DRAM manufacturer that had gone bankrupt in 2009. CXMT acquired many of Qimonda’s patents and recruited many former engineers. With that done, they shifted focus to mass production.

Fab 1, backed by the Hefei government, was completed in early 2018 and began production in 2019. By 2020, it had done something no Chinese company had ever done before: sustained mass production of DDR4 and LPDDR4 DRAM chips, which were used mostly for budget Chinese smartphones and PCs. Over the following years, CXMT advanced its process technology further, producing DDR5 and LPDDR5, used for more high-end devices.

In 2024, it began producing small amounts of older HBM2, the technology used in AI chips, and in 2025, it released DDR5 server memory. This would turn out to be incredibly useful — around that time, rising AI demand caused the memory market to skyrocket. In July 2026, CXMT went public, and its stock nearly quintupled on the first day of trading. Today, it produces nearly 10% of the world’s DRAM supply, and its memory ends up in mainstream Chinese smartphones, PCs, and data centers. It’s had an incredible arc. If anyone knows where to get CXMT merch, please let us know!

• DeepSeek raises prices and launches V4 Pro and DeepSeek Harness v0.1. Though DeepSeek-V4-Pro-0813 began rolling out yesterday, today was the official announcement. Consensus is that it’s not much more impressive than DeepSeek-V4-Flash-0731, a much smaller model. DeepSeek is finally getting into agentic coding with the first version of DeepSeek Harness, its competitor to Claude Code and Codex, built with extensible plugins in mind. Also, the company will be substantially raising prices, starting Sunday: Flash will go from $0.14/$0.28 to $0.22/$0.66 during off-peak hours and $0.44/$1.32 during peak hours1. Pro will go from $0.435/$0.87 to $0.66/$1.98 off-peak or $1.32/$3.96 peak. Still, this is much cheaper than other models, like Kimi K3 ($3/$15), Sol ($5/$30), and Fable ($10/$50).

• Google releases Gemini 3.7 Flash. It performs somewhat poorly on Agent Arena, behind GPT-5.6 Terra and Luna and DeepSeek-V4 Flash. Gemini 3.5 Pro, originally slated for release in May, is still nowhere in sight. On the bright side, 3.7 Flash is incredibly fast, at 340 tokens per second (compared to ~60 for Sol and Fable).

• OpenAI’s revenue run rate tops $40 billion. Codex, subscriptions, API, and a small amount of advertising revenue contributed. OpenAI most recently reported $25B of run-rate ARR in February, and $19B in December of 2025. Anthropic reported $47B of run-rate ARR in May, but it’s possible the companies measure revenue differently.

• Anthropic may buy Decart for $6 billion. Decart builds world models — like Oasis for physical AI and Lucy for real-time video generation — as well as highly optimized software for inference. This would be Anthropic’s largest acquisition ever by far.

• Databricks raises $5 billion at a $190 billion valuation. The company, founded in 2013 by a group of researchers at UC Berkeley, builds cloud platforms for data analytics and AI. The round would be Databricks’ Series M, and make it one of just a handful of private companies valued above $100B (along with Anthropic, OpenAI, ByteDance, and Stripe).

• X open-sources the For You timeline algorithm, including the weights on post interactions. A like is +0.5, a retweet is +1, a reply or quote is +5, a reply or quote from a mutual is +20, and a share via copy link is +20. Block is -31.2, “Not interested” is -43.2, and a report is -234 — so one report erases 468 likes.

• Anthropic investors expect the IPO valuation to exceed $2 trillion. This would make it the largest IPO of all time, bigger than SpaceX ($1.77T) and Saudi Aramco ($1.88T). Investors are projecting $100-120B of revenue by the end of the year.

• Vantage Data Centers may IPO at a $100 billion valuation. Denver-based Vantage, founded in 2010, is one of the world’s largest providers of data center infrastructure. It plans to raise around $10 billion at the IPO.

• Flock Safety expands its privacy controls. Default data retention settings will increase, searches of data will require a case code to justify them, and Flock will be able to block users from accessing data if their usage patterns are abnormal.

• Demis Hassabis met with government officials before his ouster. Hassabis met with Treasury Secretary Scott Bessent and OSTP Director Michael Kratsios to discuss his plan for a self-regulatory body for AI modeled after FINRA.

• OpenAI Chief Revenue Officer Denise Dresser leaves. Dresser joined OpenAI just nine months ago, and was previously CEO of Slack. She will be replaced by Dali Rajic, previously president and COO of cybersecurity company Wiz.

• The Trump administration is using AI for tariff enforcement in an early case of the government using AI.

• OpenAI previews GPT-5.6 Sol Ultrafast, a version of Sol powered by Cerebras allowing speeds up to 750 tokens per second. It’s invite-only for now.

• Mike Flanigan (Seasats)

• Barak Gila (Sway)

• Bryan Bischof (Theory Ventures)

• Jade Furuyabashi and Frank Liu (MongoDB)

• George Arison (Grindr) and Jayden Clark (MOTS Podcast)

• Rayan Krishnan (Vals AI)

• Aidan McLaughlin (OpenAI)

• Apoorva Joshi and Mikiko Bazeley (MongoDB)