SpaceX and Nvidia Join Forces for Orbital AI Data Centers Targeted for 2027

SpaceX and Nvidia Target 2027 for Space-Based AI Compute

SpaceX has officially partnered with Nvidia to construct its upcoming “Starmind” orbital data centers. Built around custom versions of Nvidia’s Vera Rubin NVL72 rack architecture, the initiative aims to deploy initial computing clusters into orbit by late 2027.

The move comes as terrestrial data center construction faces mounting power, cooling, and regulatory bottlenecks across the United States. Elon Musk revealed that the space-ready version of the hardware has been re-engineered to be lighter, denser, and more resistant to extreme orbital heat and radiation.

• Next-generation computing: Each rack combines 72 interconnected chips operating as a single unit, with Nvidia projecting individual processor performance at up to 25 times that of previous H100 hardware.

• Proprietary fleet integration: SpaceX plans to run its broader AI workloads, including xAI's Grok model and orbital operations, exclusively on Nvidia-powered hardware.

• Cost projections: Although current estimates place orbital computing at over four times the expense of ground facilities, SpaceX anticipates this cost ratio will invert in favor of space over the coming years.

Thomson Reuters Unveils Custom Legal AI Model

Thomson Reuters has introduced its first proprietary legal AI model, developed by fine-tuning Alibaba’s open-weights Qwen architecture on decades of proprietary legal analysis and editorial records.

• Investment scale: The two-year project required roughly $40 million in staffing and compute infrastructure, with its latest single training run costing approximately $450,000.

• Internal benchmarks: Initial internal tests showed the domain-specific model outperforming broader frontier systems such as Claude Opus 4.8, Gemini 3.1 Pro, and GPT-5.5 on specialized legal tasks.

• Open-weights plan: An open-weights variant will be released to academic and independent researchers, while current training data represents under 10% of the publisher's total content archive.

Cybersecurity Report: DeepSeek Utilized to Double Attack Frequency

A recent report from cybersecurity intelligence firm TeamT5 indicates that state-linked hacking operations in China have more than doubled their attack cadence by integrating accessible open-source models like DeepSeek into automated offensive workflows.

According to researchers, threat actors favor the model due to its high performance, low operating expense, and minimal built-in guardrails. Documented applications include automated reconnaissance across thousands of network targets, automated breach script generation, and rapid corporate inbox exfiltration.

Additional AI Industry Developments

• Perplexity Funding Talks: Search engine startup Perplexity is reportedly in discussions to secure an investment round from Nvidia at a valuation approaching $30 billion, bolstered by annual revenues climbing past $750 million.

• Porsche Signs Multi-Year AI Contract: Automaker Porsche has entered a five-year, $1.46 billion partnership with Indian IT services provider TCS to implement AI systems across engineering, factory automation, and retail operations.

• Liquid AI Launches Pipette: Liquid AI released Pipette, an open-source evaluation suite designed to benchmark on-device AI performance across mobile phones and personal computers, with validation methodology verified by Artificial Analysis.

• Taiwan Charges Tech Staff Over Chip Smuggling: Prosecutors in Taiwan have charged an Nvidia employee alongside two former Super Micro personnel for allegedly routing 74 export-controlled AI servers to China through intermediate shipments across Asia.