SpaceX and Nvidia Plan Orbital AI Data Centers, Thomson Reuters Unveils Custom Legal Model, and Hackers Weaponize DeepSeek
Zach Mink
AI-assisted summary of The SpaceX-Nvidia partnership heads for orbit — original source.
Article summary — original
SpaceX and Nvidia Plan Orbital AI Data Centers, Thomson Reuters Unveils Custom Legal Model, and Hackers Weaponize DeepSeek
This edition of The Rundown covers SpaceX and Nvidia's collaboration on space-based AI hardware, Thomson Reuters' $40M internal legal model build, Chinese state-linked hacking groups leveraging DeepSeek, and notable industry funding and legal developments.
SpaceX is collaborating with Nvidia to build orbital Starmind data centers using modified Vera Rubin NVL72 racks, aiming for space deployment by late 2027.
Thomson Reuters trained its own legal AI model on internal content using Alibaba's open-source Qwen, spending $40 million over two years.
Cybersecurity firm TeamT5 reported that Chinese state-linked hacking groups doubled their attack frequency after adopting open-source DeepSeek models.
Perplexity is in discussions to raise Nvidia-led funding at a valuation reaching up to $30 billion.
SpaceX and Nvidia Partner for Orbital Data Centers
SpaceX announced plans to build its space-based Starmind data centers around Nvidia's Vera Rubin NVL72 rack architecture, targeting orbital deployment as early as Q4 of next year. Each rack contains 72 chips, which Nvidia notes deliver up to 25 times the compute of the H100. SpaceX plans to operate Grok and its orbital fleet using the hardware, with Elon Musk stating the space edition is lighter, simpler, and modified for orbital radiation and heat.
Thomson Reuters Launches In-House Legal AI Model
Thomson Reuters introduced a proprietary legal AI model developed by adapting Alibaba's open-source Qwen model and training it on decades of proprietary legal and editorial material. The two-year project cost $40 million in compute and staffing, with recent training runs costing $450,000. Thomson Reuters plans to release an open-weights version for researchers, while noting that training has currently utilized less than 10% of its content archive.
Chinese Hackers Leverage DeepSeek to Expand Cyber Operations
According to Taiwanese security firm TeamT5, state-linked Chinese hacking groups have more than doubled their attack volume by incorporating DeepSeek into their operational workflows. Analysts highlighted that the model is favored due to low operational costs and minimal cyber guardrails, assisting actors with reconnaissance across target addresses, email intrusions, and crafting exploit code.
Industry Quick Hits and Commercial Moves
Perplexity is negotiating an equity round that could value the search firm at up to $30 billion with investment from Nvidia, following revenue growth exceeding $750 million. In automotive technology, Porsche agreed to a five-year, $1.46 billion contract with TCS to incorporate AI into operations. Liquid AI released Pipette, an on-device benchmarking tool verified by Artificial Analysis, while Taiwanese authorities charged an Nvidia manager and two former Super Micro employees for allegedly smuggling 74 restricted AI servers to China.
Article summary — zh
SpaceX携手英伟达筹建太空数据中心，多项AI行业动向同步汇总
本期简报涵盖多项AI领域重大动态：SpaceX联合英伟达布局太空轨道AI算力；汤森路透基于阿里开源模型自研法律AI；中国黑客团队被指利用DeepSeek大幅增加攻击频次；另附多项投融资与行业速讯。
SpaceX计划采用英伟达Vera Rubin NVL72机架构建太空数据中心，目标最快2027年发射入轨
汤森路透耗资4000万美元基于阿里Qwen开发首款自研模型，旨在摆脱对外部昂贵API的依赖
台湾研究机构TeamT5指出，黑客团队借助开源模型DeepSeek使网络攻击数量增长超一倍
英伟达洽谈参投Perplexity估值或达300亿美元；保时捷与印度TCS签署14.6亿美元AI合作协议
SpaceX与英伟达联合推进轨道数据中心
SpaceX宣布将围绕英伟达Vera Rubin NVL72机架打造其“Starmind”太空数据中心，马斯克透露了针对轨道辐射与散热环境精简的硬件细节，目标在2027年第四季度发射首批机架。尽管目前轨道算力成本被估算为地面计算的4倍以上，但SpaceX计划全线采用英伟达硬件支撑从Grok到轨道机群的算力需求。
汤森路透自研法律AI模型以降低API依赖
汤森路透基于阿里巴巴开源模型Qwen及其数十年专有数据，历时两年耗资4000万美元打造首个内部自研模型，最新一次微调训练成本为45万美元。该模型训练覆盖了不到10%的内容库，内部测试显示其部分领域跑分超越Claude Opus 4.8与GPT-5.5，未来将面向研究人员开放权重版本。
研究称黑客利用DeepSeek加剧网络攻击
台湾安全公司TeamT5在彭博社报道中披露，有政府背景的中国黑客团伙在将DeepSeek等开源模型整合入攻击手段后，实施攻击的频率较以往翻了一倍以上。分析指出，DeepSeek由于成本低且网络安全防护限制较少，被用于编写渗透代码、窃取企业邮件及目标侦察。
行业快讯：投融资、合作与合规审查
Perplexity正洽谈以最高300亿美元估值引入英伟达股权投资，其年营收已突破7.5亿美元；保时捷与印度IT巨头TCS签署为期5年、价值14.6亿美元的AI整合协议；AI研究员Luke Metz从OpenAI离职加入Meta；英伟达一名经理及两名前超微电脑员工在台湾面临起诉，涉嫌经由印尼、日本和香港向中国非法走私74台受管制的AI服务器。
Saved original article
[Read Online]({{live_url}}) <sup>|</sup> [Sign Up](https://www.therundown.ai/subscribe?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) <sup>|</sup> [Advertise](https://therundownai.typeform.com/to/kraZ1TSO?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922)
**Good morning, {{ first\_name | AI enthusiasts }}, and welcome to our 3,218 new readers.** Sam Altman called data centers in space "ridiculous." Elon Musk just answered with a launch timeline and a big-time hardware partner.
SpaceX and Nvidia are officially teaming up on Musk's orbital data centers, aiming for the first racks in space by late 2027. With U.S. data centers piling up opposition, the best option left might be one without any neighbors (yet).
**In today’s AI rundown:**
Nvidia joins SpaceX’s orbital AI build
Thomson Reuters builds its own legal AI model
Build a reusable AI design system with Open Design
Chinese hackers double attacks with DeepSeek
**LATEST DEVELOPMENTS**
Image source: Nvidia
**The Rundown:** SpaceX just announced it will [build](https://nvidianews.nvidia.com/news/spacexai-adopts-nvidia-vera-cpu-to-accelerate-agentic-ai-at-massive-scale?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) its space-based Starmind data centers around Nvidia’s Vera Rubin NVL72 rack, with Elon Musk [revealing](https://x.com/elonmusk/status/2091939113008238838?s=20&utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) new details on a slimmed-down space version of the system he wants in orbit by Q4 next year.
**The details:**
Each rack packs 72 chips working as one big computer, with Nvidia saying each chip puts out up to 25x the computing power of its older H100.
Musk said the space rack is “significantly simpler, lower cost, denser and lighter” than typical hardware, tweaked for orbit’s radiation and heat.
Musk has called Vera Rubin "the best AI computer," with SpaceX reportedly set to run everything from Grok to its orbital fleet on Nvidia alone.
Analysts estimate orbital compute at more than 4x the cost of ground compute today, a gap Musk [claims](https://www.npr.org/2026/04/03/nx-s1-5718416/ai-data-centers-in-space-spacex-elon-musk?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) will flip in its favor in the next few years.
**Why it matters:** Sam Altman called space-based data centers ‘ridiculous’ earlier this year, but both the timelines and companies lining up behind them are getting very real. With the negativity around the ground buildouts reaching a boiling point, the orbital solution can’t come soon enough for those hoping to continue scaling the AI boom.
**The Rundown:** Most companies bolt on AI governance after the fact and hope regulators don't notice the gaps. Dataiku builds it directly into how models and agents are built, deployed, and monitored, so compliance isn't a scramble; it's the default.
**With Dataiku, you can:**
Confidently navigate real-world AI regulation across models and agents
Embed governance into every stage, from analytics to deployed AI systems, like Beinex, DHL Aviation, and OHRA already do
Join some of the world's top global companies already using Dataiku
Image source: Thomson Reuters
**The Rundown:** Thomson Reuters, the legal research giant behind Westlaw and the Reuters newswire, [introduced](https://www.thomsonreuters.com/en-us/posts/innovation/how-we-built-thomson/?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) its first homegrown AI model, built by retooling Alibaba's open-source Qwen and training it on decades of the firm's content.
The two-year effort cost $40M in staff and computing power, with the company’s latest training run coming in at $450k.
Thomson benchmarked the model ahead of Claude Opus 4.8, Gemini 3.1 Pro, and GPT-5.5 in certain areas, though it’s only been tested internally so far.
CTO Joel Hron compared renting AI to renting a house, [telling](https://www.businessinsider.com/thomson-reuters-builds-ai-model-rely-less-on-anthropic-2026-8?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) Business Insider you end up with no "equity that compounds into something valuable" over time.
An open-weights version of Thompson will be available for researchers, with its training so far covering under 10% of the company’s overall content library.
**Why it matters:** $40M might sound like a lot, but to a giant company previously on expensive models, it's closer to a year of API invoices than a moonshot budget. Expect more orgs with deep content and knowledge bases to run the same math, especially as training costs become even more accessible on top of strong open base models.
**The Rundown:** In this guide, you will learn how to use a free AI design tool to extract your brand's design rules, letting you easily create consistent design assets using AI.
**Step-by-step:**
Download [Open Design](https://open-design.ai/?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) and connect an existing AI agent such as Codex CLI
Click "Create New Design System,” paste in your website, click Add, then Continue. Add a logo, images, or a [DESIGN.md](http://DESIGN.md?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) if it missed anything
Select the new design system and give it a concrete brief like "Build a 10-second animation explaining our product”
For v2, give feedback like you would to ChatGPT. You can also annotate on the design or circle things you want changed
**Pro tip:** Shape the idea and prompt in ChatGPT first, then build it in Open Design's visual, Figma-like workspace.
**The Rundown**: AI transformation fails when you DIY. StackAI gives you one platform to build, deploy, and scale, backed by a team of dedicated AI experts who know what works. IT retains 100% control over access, tokens, and usage.
**Why enterprises choose StackAI:**
No-code platform for rapid building and deployment
300+ native integrations, no LLM lock-in
Global governance and advanced analytics
Image source: Images 2.0 / The Rundown
**The Rundown:** China's state-tied hacking crews are [pulling off](https://www.bloomberg.com/news/articles/2026-08-24/chinese-hackers-use-deepseek-to-boost-attacks-researchers-say-mt7o4205?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) over twice as many attacks as they did before integrating open-source AI models like DeepSeek into their playbooks, Taiwanese research firm TeamT5 revealed in a new Bloomberg report.
The groups reportedly favor DeepSeek's low cost and loose guardrails, with researchers yet to catch pricier models like Kimi K3 in an attack.
"DeepSeek is the AI of choice for Chinese hackers because it's relatively powerful with very low cyber guardrails," TeamT5 chief analyst Charles Li said.
The tracked groups put DeepSeek to work writing break-in code, raiding a Taiwanese firm's email, and mapping a target via 1,000 addresses.
**Why it matters:** The AI Security Institute [warned](https://www.aisi.gov.uk/blog/how-fast-is-autonomous-ai-cyber-capability-advancing?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) in May that AI's cyber skills are doubling every few months, and the numbers show the attacks themselves are now doubling right alongside it. While fears center on top models like Mythos, cheap, downloadable ones with fewer guardrails might be the harder security problem to solve.
**QUICK HITS**
▸ [Florence built an AI detective for a decades-old family mystery](https://app.therundown.ai/community/posts/4f96e696-3874-41d3-a9c7-9e3f4b25380e?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922)
Today’s workflow comes from reader **[Florence Locheron](https://www.linkedin.com/in/florence-locheron/?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922)**:
“I used AI to help investigate a family mystery that had remained unresolved for decades: identifying the biological family of my maternal grandfather. I had DNA matches, family trees, names, dates, historical records, old photographs, obituaries, Facebook genealogy groups, and conversations with possible relatives. The difficult part was connecting all these scattered clues without jumping to conclusions.
I built a research workflow in which AI acts as an investigation partner. Instead of asking AI, 'Who was my grandfather's biological father?', I asked it to analyze possible family connections and identify which hypotheses were compatible with the available evidence. I treated AI-generated connections as leads rather than genealogical proof.
When a promising connection appeared, I returned to the original genealogy and DNA sources to verify it. This gradually turned a long list of DNA matches into a smaller number of research paths.”
*See Florence’s workflow here. How do you use AI? Tell us for a chance to be featured.*
🎬 [Wan 3.0](https://www.therundown.ai/tools/wan-3-0?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) - Alibaba’s latest AI video model, now generally available
📱 [Antigravity](https://www.therundown.ai/tools/antigravity?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) - Google’s agentic development tool, now with remote control
🎶 [Firefly Audio](https://www.therundown.ai/tools/firefly-audio?utm_campaign=slack-turns-coding-into-a-group-project&utm_medium=referral&utm_source=therundownai.beehiiv.com) - Create commercially safe music, voiceovers, & sound effects
🤖 [Apodex 1.1](https://x.com/Apodex_AI/status/2091916791308313018?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) - Open models that coordinate parallel AI teams on tasks
**Perplexity** is [in talks](https://www.theinformation.com/articles/nvidia-discusses-perplexity-investment-30-billion-plus-valuation-considered-tech-licensing-deal?utm_campaign=Editorial&utm_content=Article&utm_medium=organic_social&utm_source=threads%2Ctwitter&rc=a8sc5q) to raise funding from Nvidia in an equity round at a valuation as high as $30B, with the company’s revenue more than tripling to $750M+ this year.
**Porsche** [signed](https://www.reuters.com/world/india/indias-tcs-buy-porsches-it-unit-373-million-2026-08-24/?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) a 5-year, $1.46B deal with Indian IT giant TCS to help integrate AI across the automaker’s customer, factory, and engineering operations.
**Liquid AI** [launched](https://www.liquid.ai/blog/pipette-on-device-ai-benchmarking-by-liquid-ai?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) Pipette, a new open-source suite that benchmarks AI models on phones and laptops, with methodology independently verified by Artificial Analysis.
**AI researcher Luke Metz** reportedly [joined](https://www.axios.com/2026/08/24/meta-hires-openai-luke-metz?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) Meta, ending a two-year loop that took him from OAI to Thinking Machines, back to OAI, and now to Alexandr Wang's team.
**An Nvidia manager and two ex-Super Micro staffers are** [facing](https://apnews.com/article/taiwan-china-us-nvidia-ai-server-chip-illegal-export-511e9ae69d517c49f19512d45b1a8b0c?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922) charges in Taiwan over sneaking 74 banned AI servers into China via Indonesia, Japan, and Hong Kong.
Read our last AI newsletter: [A mystery challenger at the AI frontier](https://therundownai.beehiiv.com/p/a-mystery-challenger-at-the-ai-frontier?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922)
Read our last Tech newsletter: [The mRNA cancer vaccine finally delivers](https://tech.therundown.ai/p/the-mrna-cancer-vaccine-finally-delivers?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922)
Read our last Robotics newsletter: [Humanoids beat Bolt’s 100m record](https://robotnews.therundown.ai/p/humanoids-beat-usain-bolt-100m-record?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922)
Today’s AI tool guide: [Build a reusable AI design system with Open Design](https://app.therundown.ai/guides/build-a-reusable-ai-design-system-with-open-design?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922)
RSVP to next workshop on Aug. 27: [Build your first AI Eval](https://app.therundown.ai/live/how-to-build-your-first-ai-eval?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0922)
See you soon,
*Rowan, Zach, Shubham, Jennifer, and Nate — the humans behind The Rundown*