Big infrastructure and enterprise plays are dominating today. SpaceX and Nvidia are pushing AI compute into orbit, Meta is preparing to charge premium prices for a new AI agent, and Thomson Reuters has built its own legal model rather than keep paying for someone else's. Meanwhile cheap open models are quietly reshaping both productivity and cybercrime.

Summary: SpaceX is working with Nvidia to build 'Starmind' data centres in space, anchored by Nvidia's Vera Rubin NVL72 rack. Each rack holds 72 chips and is redesigned for the radiation and heat of orbit. Elon Musk expects lower costs and higher density over time, with a target of late 2027. Sam Altman had dismissed the idea as ridiculous, but named partners and firm timelines are giving it credibility.

Why it matters: Ground-based data centres face growing local opposition, and orbital compute is being pitched as a way around that limit.

Source: https://nvidianews.nvidia.com/news/spacexai-adopts-nvidia-vera-cpu-to-accelerate-agentic-ai-at-massive-scale

Summary: Mark Zuckerberg is breaking Meta's habit of free services with a new consumer AI agent called Hatch, due in the coming weeks. Hatch can carry out tasks on a user's behalf, and a premium tier may cost up to $200 a month, matching the top plans from OpenAI and Anthropic. Meta also plans to release its next major model, code-named Watermelon, in October.

Why it matters: This is a clear signal that Meta wants to monetise AI directly and compete head to head with the leading labs on price and capability.

Summary: Thomson Reuters has built a proprietary legal AI model on Alibaba's open-source Qwen, using decades of its own legal content. The two-year project cost roughly $40 million across staff and compute. The company says it beats versions of Claude, Gemini and GPT on some tasks, though it is still in internal testing. An open-weights variant is planned.

Why it matters: Large firms are choosing to own and train models rather than pay ongoing API fees, and open Chinese bases are becoming the default foundation.

Summary: A humanoid robot named Tiangong Ultra reportedly ran 100 metres in 9.39 seconds, faster than Usain Bolt's human record of 9.58 seconds. Another robot, Lightning, clocked 9.47 seconds but collapsed straight after. Both struggled to stop cleanly, slamming into barriers and stumbling off the track.

Why it is interesting: It captures where robotics is right now, remarkable raw speed paired with a glaring lack of balance and control.

Summary: LinkedIn's new 'Seems like AI slop' button has been used by more than one million people since it launched. The move followed findings that 41% of long-form posts on the platform were fully AI-produced. LinkedIn updated its detection systems, removed its 'enhance your post' feature, and reports a 40% drop in slop content.

Why it is interesting: It shows a large platform actively pushing back against AI-generated filler and giving users a way to flag it.

Summary: A reader used AI to solve a decades-old question about her grandfather's biological father. She fed DNA matches, family trees, historical documents and interview notes into an AI-assisted workflow. Rather than asking for direct answers, she treated AI suggestions as leads to verify against original records, narrowing many matches down to a few real pathways.

Why it is interesting: It is a practical model for using AI on messy real-world problems, as a lead generator rather than an oracle.

Source: https://www.zdnet.com/article/have-a-genealogy-mystery-how-i-used-ai-to-solve-a-family-puzzle/

Context: Anthropic's ELI5 skill has caught on internally for turning complex material into plain-language explanations. You can replicate the effect with any capable model to make dense information accessible.

Prompt: Explain the following concept as if I am five years old, then explain it again as if I am a smart adult with no background in the field. Avoid jargon. Use one everyday analogy in each version. Concept: [paste the topic or text here].

Example use case: A manager pastes a technical vendor proposal and gets two clear explanations, one to grasp it themselves and one to brief the wider team.

Summary: A report from Taiwanese firm TeamT5 says Chinese state-connected hackers have roughly doubled their attacks after adopting open-source models like DeepSeek. The model offers strong capability with minimal guardrails, helping attackers write malicious code, break into email and map networks. Its low cost and ease of use make it especially attractive.

Why it is important: Cheap, lightly controlled models may pose a bigger security threat than the larger, tightly governed systems everyone watches.

Source: https://www.straitstimes.com/asia/east-asia/chinas-hackers-use-deepseek-for-attacks-researchers-say

Summary: Alibaba announced a $10.2 billion share issue directed entirely at AI investment, though its stock fell on the news. It also released Wan 3.0, a model that generates 30-second video clips from almost any uploaded content, including slides, spreadsheets and web pages.

Why it is important: It underlines how aggressively Chinese tech giants are funding AI and pushing media generation into mainstream use.

Summary: A developer ran a refined Qwen build locally using the Pi coding agent and reported it outperforming Claude Opus 5 High on a benchmark for recent real software bugs. Quantized versions reportedly need only about 18 to 23GB, fitting a 24GB-class GPU. A method from Berkeley researcher Shuo Yang can run fuller checkpoints while preserving accuracy.

Why it is important: If capable models run on consumer hardware, teams gain privacy, cost savings and speed without relying on cloud APIs.

That's it for today's Daily Pulse. Forward this to one person who wants to stay ahead of AI. See you in the next one. - Nicolas