Last Week’s AI News #33

From ocean-based data centers to AI systems that begin to improve themselves, the boundaries between software, infrastructure, and intelligence are dissolving. 

Here’s what you need to know from last week in AI:

• Peter Thiel backs $140M ocean-based AI compute startup Panthalassa
• Nvidia brings AI compute into residential homes with Span partnership
• Anthropic signs massive compute deal with SpaceX for Colossus 1
• Anthropic co-founder predicts AI will self-improve before 2029
• OpenAI accelerates AI phone launch to 2027
• Anthropic launches 10 domain AI agents for finance and insurance
• Google launches AI health coach with Fitbit integration
• DeepMind invests in EVE Online as AI training sandbox
• AI system RAVEN confirms 100+ exoplanets from NASA data
• Everything else that happened in AI last week

PETER THIEL BACKS $140M OCEAN-BASED AI COMPUTE STARTUP PANTHALASSA

Peter Thiel has led a $140M Series B round into Panthalassa, an Oregon-based startup building autonomous floating compute systems powered by ocean waves, valuing the company at nearly $1B.

Each 85-meter steel node floats in open ocean, converting wave motion into electricity for onboard AI chips, cooled naturally by seawater. Once deployed, the systems can reposition themselves using hull dynamics alone and transmit results via SpaceX Starlink.

The company is now completing a pilot factory near Portland, with first deployments planned in the Pacific Ocean and commercial rollout expected in 2027.

Why does it matter for businesses?
AI infrastructure is moving outside traditional data centers. Ocean-based compute reduces land constraints, cooling costs, and regulatory friction, while opening a new geopolitical layer to AI infrastructure control.

NVIDIA BRINGS AI COMPUTE INTO RESIDENTIAL HOMES WITH SPAN PARTNERSHIP

California startup Span is partnering with Nvidia to install mini AI data centers directly onto residential and commercial buildings.

The systems, called XFRA nodes, use Nvidia’s liquid-cooled RTX PRO 6000 Blackwell GPUs and are designed to run quietly while tapping unused grid capacity.

Span claims it can deploy these systems 6x faster and at one-fifth the cost of traditional centralized data centers, with pilots already underway in new housing developments.

Why does it matter for businesses?
Compute is becoming distributed at the edge of society, not just in hyperscale data centers. This changes energy distribution, infrastructure planning, and local grid economics.

ANTHROPIC SIGNS MASSIVE COMPUTE DEAL WITH SPACEX FOR COLOSSUS 1

Anthropic has signed a deal with SpaceX to lease its Colossus 1 supercluster in Memphis, bringing over 220,000 Nvidia GPUs online.

Claude usage caps are being expanded, with longer session limits and fewer peak restrictions.

Elon Musk commented that SpaceX will rent compute to AI companies aligned with “good for humanity” goals.

Anthropic is also reportedly committing to a $200B, 5GW compute expansion with Google Cloud.

Why does it matter for businesses?
AI compute is becoming a strategic asset controlled by a small number of infrastructure players, including SpaceX, blurring lines between competitors, partners, and infrastructure providers.

ANTHROPIC CO-FOUNDER PREDICTS AI WILL SELF-IMPROVE BEFORE 2029

Anthropic co-founder Jack Clark estimates a 60%+ probability that AI systems will begin training their own successors before 2029.

He cites rapid capability gains across AI R&D tasks, with models progressing from seconds-long autonomy in 2022 to multi-hour independent work by 2026.

Benchmark data shows coding performance rising from 2% to nearly 94% in under three years.

Why does it matter for businesses?
If AI systems begin improving themselves, model development shifts from linear progress to compounding acceleration, fundamentally changing the speed of technological evolution.

OPENAI ACCELERATES AI PHONE LAUNCH TO 2027

OpenAI is now targeting mass production of its first AI phone in the first half of 2027, a full year earlier than previously expected.

The device will prioritize real-world AI perception, with advanced imaging systems designed for AI agents to interpret environments more effectively.

MediaTek is expected to be the sole chip supplier, using dual AI processors for vision and language tasks.

Why does it matter for businesses?
AI is moving from software into hardware ecosystems. Whoever controls the device layer may control the primary interface for agent-based systems.

ANTHROPIC LAUNCHES 10 DOMAIN AI AGENTS FOR FINANCE AND INSURANCE

Anthropic has introduced 10 production-ready AI agents designed for financial services and insurance workflows.

These agents handle tasks such as pitchbook generation, KYC review, earnings analysis, and valuation workflows, while integrating directly into enterprise systems.

They can be customized to internal risk policies and deployed across Claude platforms or enterprise connectors.

Why does it matter for businesses?
AI is shifting from general-purpose models to vertically integrated labor systems. Entire job functions in finance and insurance are becoming modular AI workflows.

GOOGLE LAUNCHES AI HEALTH COACH WITH FITBIT INTEGRATION

Google has launched its AI health coach powered by Gemini, integrating Fitbit, Health Connect, medical records, and wearable data into a unified health system.

A new $99 screenless wearable tracks biometric signals and feeds data directly into the AI system.

The platform can interpret medical records, suggest workouts, and analyze diet from images.

Why does it matter for businesses?
AI is becoming a personal health operating layer. Control over biometric ecosystems may become as valuable as control over mobile operating systems.

DEEPMIND INVESTS IN EVE ONLINE AS AI TRAINING SANDBOX

DeepMind has taken a stake in Fenris Creations, using EVE Online as a long-term AI training environment.

AI systems will operate in an offline version of the game to test memory, planning, and long-term reasoning in complex systems.

Why does it matter for businesses?
AI is moving from static benchmarks into living simulated environments, closer to real-world unpredictability than traditional training datasets.

AI SYSTEM RAVEN CONFIRMS 100+ EXOPLANETS

Researchers at the University of Warwick used an AI system called RAVEN to analyze NASA TESS data and confirm over 100 exoplanets.

The system also identified more than 2,000 additional candidates, including planets in previously unexplained orbital zones.

Why does it matter for businesses?
AI is now accelerating scientific discovery at scale, not by new instruments, but by extracting more value from existing data.

EVERYTHING ELSE THAT HAPPENED IN AI LAST WEEK

• OpenAI rolled out GPT-5.5-Instant to all ChatGPT users, improving performance, memory, and response personalization. This makes advanced AI more usable in daily business workflows without prompt engineering.
• Microsoft expanded Copilot Cowork to iOS and Android, adding task-specific skills and business data integrations. This pushes AI assistants directly into mobile-first business operations.
• Perplexity launched Computer for Professional Finance, combining licensed financial data with 35 AI workflows for analysts. This automates routine financial analysis tasks that previously required junior staff.
• Anthropic introduced new Managed Agents features including dreaming, outcome evaluation, and multi-agent orchestration. This allows AI systems to self-review work and coordinate complex business tasks.
• Perplexity also rolled out its Personal Computer feature for Mac, enabling AI to take actions across local files, apps, and browsers. This moves AI from assistant to active operator inside business environments.
• Spotify launched Personal Podcasts, allowing users to turn documents, notes, and briefings into audio content. This automates content repurposing for internal communication and learning.
• Coinbase announced a 14% workforce reduction as it shifts toward AI-native teams and agent-driven workflows. This signals real organizational restructuring around automation rather than augmentation.
• Sierra raised $950M at a $15B valuation, with its AI customer experience platform already serving 40% of Fortune 50 companies. This shows enterprise customer support is rapidly being centralized through AI systems.
• Scale AI secured a $500M Pentagon contract for large-scale data analysis. This highlights growing institutional reliance on AI systems for high-volume decision workflows.
• Subquadratic introduced SubQ, a model with a 12M token context window and significantly lower cost for long tasks. This enables SMBs to process large-scale documents and workflows without enterprise infrastructure.

Infrastructure is becoming distributed. Models are becoming increasingly autonomous. And industries are shifting from human-executed processes to AI-coordinated operations.

The companies that adapt early won’t just use AI, they will operate inside systems defined by it.

And that gap is only going to widen.

👉 Subscribe to the newsletter to stay ahead of what’s coming next.