The Intelligence Explosion: Anthropic’s 2029 Warning, SpaceX’s Power Play, and the Rise of Oceanic Compute
OpenAI’s secret AI phone timeline, the Pentagon’s new classified AI club, and Figure’s massive humanoid production ramp.
The velocity of technological progress has shifted from a steady climb to a vertical ascent. This week, the industry didn’t just witness new models; it stared into the face of a self-evolving future. With Anthropic’s leadership projecting a world where AI trains itself within three years and the physical infrastructure of the internet moving into the open ocean, the “frontier” has expanded far beyond the screens on our desks.
We are no longer just building tools; we are architecting an autonomous layer for reality. From OpenAI’s accelerated path toward a dedicated AI phone to the Pentagon’s selective “AI-first” fighting force, the stakes have shifted from corporate market share to national security and planetary-scale engineering.
In this edition, we break down the logic behind the “self-improving AI” timeline, the surprising alliance between Elon Musk and his rivals, and the daring hardware bets that seek to turn the Pacific Ocean and residential walls into the next great data centers.
1. The AI Vanguard: Autonomy & Alliances
The “2029” Threshold: AI Research Goes Auto-Pilot
Anthropic co-founder Jack Clark has published a thesis that changes the math for AGI. He places 60%+ odds on AI systems training their own successors before 2029, citing public data that shows AI is already devouring core R&D tasks at an exponential rate.
The Data: METR benchmarks show AI’s independent task capacity jumped from 30 seconds in 2022 to 12-hour stretches in early 2026. Projections suggest 100-hour autonomous runs by year-end.
The Coding Leap: On the SWE-Bench (real GitHub coding), models moved from a 2% success rate to 93.9% in under three years.
Why it matters: Once AI can reliably build and train its own next generation, model development moves from human-speed to machine-speed. We are approaching the “intelligence explosion” faster than the most aggressive models predicted.
The SpaceX-Anthropic Alliance: Pragmatism over Rivalry
In a stunning reversal of public rhetoric, Anthropic has signed a deal with SpaceX to lease the “Colossus 1” supercluster. Despite Elon Musk previously labeling Anthropic as “misanthropic,” he is now providing the compute backbone for Claude’s expansion.
The Muscle: Anthropic will utilize 220,000 Nvidia GPUs within a 300+ MW Memphis cluster.
The Payoff: Claude Code’s usage caps have immediately doubled for paid tiers, and peak-hour restrictions have been eliminated.
Why it matters: This is “the enemy of my enemy is my friend” in action. Musk is patching the compute hole for OpenAI’s biggest rival while SpaceXAI pivots toward becoming a dominant infrastructure provider for the entire sector.
The Rise of “AI-Native” Consulting
The barrier to AI adoption is no longer the model; it’s the installation. Anthropic just formed a $1.5B services company with Blackstone and Goldman Sachs, while OpenAI is reportedly raising $4B for its own “Deployment Company.”
The Strategy: These ventures pair frontier engineers with mid-sized companies to build custom, integrated workflows that traditional IT departments can’t handle.
Why it matters: Frontier labs are effectively turning into elite consulting firms to ensure their models actually get “installed” into the messy reality of global business.
Source: Anthropic Enterprise Services
2. Deep Dive: Agentic Reality & Medical Breakthroughs
OpenAI’s Vision: The 2027 AI Phone
Supply chain analyst Ming-Chi Kuo reports that OpenAI has moved the mass production timeline for its first AI-native phone up by a full year.
The Specs: Aiming for 2027, the device will feature a specialized Image Signal Processor (ISP) to help AI agents “see” the world in high-dynamic range, alongside dual processors for simultaneous vision and language tasks.
Source: Ming-Chi Kuo on OAI Hardware
Doctor AI: Outperforming the ER
A Harvard study in Science put OpenAI’s o1-preview through 76 real ER cases. The AI diagnosed patients more accurately (67.1%) than two attending physicians (55.3% and 50.0%).
The Catch: In one case, the AI caught a rare infection 12 to 24 hours before the human doctors did.
Why it matters: As models become generations more advanced, the role of the physician shifts from primary diagnosis to high-level oversight of AI-driven care.
Source: Harvard: AI Outperforms Doctors
3. Robotics, Hardware & The Tech Frontier
The Ocean Frontier: Wave-Powered Data Centers
Peter Thiel has led a $140M round for Panthalassa, a startup building autonomous, floating compute nodes powered by the motion of the ocean.
The Tech: 85-meter steel nodes convert wave energy into electricity for onboard AI chips, with seawater providing natural liquid cooling.
The Deployment: These nodes steer themselves via hull shape and beam data back via Starlink, opening up the ocean as a solution to the terrestrial grid crisis.
Source: Panthalassa: Powering AI at Sea
Span & Nvidia: The “Porch” Data Center
While Panthalassa goes to sea, Span is going to your backyard. Teaming with Nvidia, they are installing mini data centers (XFRA) on the exterior walls of homes to tap into unused electrical capacity.
The Efficiency: Span claims it can install 8,000 units 6x faster and at 20% the cost of a traditional 100MW data center.
Figure’s 24x Production Jump
Humanoid startup Figure has ramped its “BotQ” factory from one robot a day to one per hour. With 350+ units shipped, they are now targeting 12,000 robots per year.
The Data Engine: Every robot shipped is a data collector for “Helix,” Figure’s AI model, creating a fleet-wide feedback loop for learning tasks.
Source: Figure Robot Production Update
4. Industry Briefs: What Else is Moving?
AI & Software Strategy
GPT-5.5-Instant: OpenAI has begun rolling out the newest iteration of GPT-5.5, offering significantly more personalized memory and concise responses for ChatGPT users.
Realtime-2: New API models from OpenAI allow voice agents to reason, use multiple tools, and “talk while they think,” effectively ending the awkward “turn-based” era of voice AI.
The Pentagon’s Club: The DoD added OpenAI, SpaceX, and Nvidia to its classified networks, while blacklisting Anthropic due to “supply-chain risk”—a move that complicates the government’s access to the powerful Mythos model.
DeepMind x EVE Online: Google DeepMind is using the 23-year-old space game EVE Online as a “social sandbox” to train agents in long-term memory and complex economic reasoning.
Hardware & Tech Shifts
GameStop’s $56B eBay Bid: In a massive “meme-stock” power move, GameStop has made an unsolicited offer to buy eBay to create a dominant secondhand marketplace.
Tesla’s 10 Billion Mile Milestone: FSD (Supervised) has hit the data threshold Elon Musk set for “unsupervised” safety. The tech is there; the regulatory approval is the next hurdle.
The Mac Mini Shortage: Apple confirmed that the demand for local AI agent development has caused a global shortage of Mac minis and Mac Studios.
Uber’s Sensor Grid: Uber is turning its human-driven fleet into an “AV Cloud,” selling sensor data from everyday trips to autonomous vehicle companies to train their models.
The Final Word: The Speed of Self-Evolution
The narrative of 2026 has shifted from “what can AI do?” to “how fast can AI build itself?” As Anthropic prepares for “fire drills” in capability surges and Figure ramps up robot production to an hourly cadence, the window for human-led design is closing. We are entering an era of automated research, decentralized compute, and agentic hardware that operates while we sleep.
Efficiency is no longer a goal; it is the baseline. The companies and individuals who thrive in the next twelve months will be those who stop treating AI as a tool and start treating it as a collaborator that is learning to improve itself in real-time.
Stay curious, stay grounded.
Which update do you think will have the biggest impact on your industry by 2027? Join the conversation below.
#AI #Robotics #Hardware #TechTrends #FutureOfWork #Innovation #WeeklyUpdate