Welcome back, Spectrum Visionaries.
London's medieval streets are about to meet Silicon Valley's robotaxis. While Waymo prepares to navigate British roundabouts, OpenAI is quietly building the computing backbone for the most expensive sense: Vision.
This issue covers why teaching machines to see is 100x more pricey than language but Big Tech is doing it anyway. Plus, we’ll show you how to make AI avatars that actually sells products.
THE AI UPDATE
Waymo's Robotaxis Are Coming to London and Vision Is the Deciding Factor:
Waymo announced its autonomous taxi service launches in London in 2026, its first European bet and easily its hardest. The expansion requires one thing above all: ultra-robust visual navigation to test whether it can navigate medieval streets, left-side driving, and London's notorious taxi culture.
The RNIB's Head of Inclusive Design Robin Spinks, blind since birth, sees these developments as liberation: 'I've long hoped for technology that enables spontaneous autonomous travel.' If AVs deliver, they could revolutionize mobility for millions with disabilities and make roads safer for everyone.OpenAI Expands US Data Centers with Oracle, SoftBank:
The GPT creator is building 5 new AI data centers across the US, expanding Stargate to 7 gigawatts. Oracle is handling three sites in Texas, New Mexico, and the Midwest, while SoftBank takes Ohio and Texas.
Why so much power?
Because OpenAI is scaling the infrastructure required to run vision workloads at a national scale, from camera streams to satellite imaging to clinic-level diagnostics. Vision requires more compute than language, and Stargate is designed to feed that appetite.Robots are finally doing the dirty work
In Vegas, Oto the humanoid concierge handles midnight check-ins in 50+ languages, freeing hotel staff for meaningful guest interactions. DFKI's underwater robot pulls 250 kg of hazardous ocean waste from depths unsafe for divers. And the UK Army's 50 new bomb-disposal bots mean no soldier has to approach live explosives.
Robots are taking on work that's either mind-numbingly repetitive or life-threatening for humans, thanks to Vision Intelligence.
THE DEEP DIVE
Why Vision Is the True Cost Center of AI
We think language makes us intelligent, but our brains spend 5x more energy on vision. Now AI is discovering the same expensive truth.
Brain’s Hidden Power Bill
The brain runs on just 20 watts, about the same as a dim LED bulb. Yet nearly 30% of that tiny power budget goes to vision, while language uses barely 5%.
Why the imbalance? Because vision isn’t symbolic. It’s continuous, high-bandwidth input. Our retina streams 100-1,000× more data than our speech system. Every frame, the visual cortex reconstructs motion, depth, edges, and lighting, all before you consciously “see” anything.
AI Runs Into the Same Physics
LLMs stay efficient because words are compressed but Vision isn’t. One image equals thousands of text tokens, and a minute of video carries as much data as an entire book.
Before an AI can identify anything, it must decode millions of pixels into edges, objects, depth, and motion. In the human brain, that makes vision 3-5× more energy-intensive than language. In AI Vision systems, the gap is even wider: vision routinely demands 10-100× more compute per sample.
This is why models like Sora, Midjourney, and Gemini 2.5 are computationally heavy—they're reconstructing reality frame by frame.
The Takeaway?
Vision is the heaviest workload in intelligence for both brains and machines. Giants like Nvidia and OpenAI are investing aggressively because AI Vision is the foundation for autonomous cars, robotics, medical imaging, and every real-world AI product of the next decade. The breakthrough will be whoever makes it cheap enough to run everywhere.
IN THE LOOP
What’s trending in the Space
US researchers built bat-inspired robots that navigate in darkness using high-speed AI Vision.
Meta is dropping $1B on a new Wisconsin AI data center, its biggest US bet and a clear signal that hyperscalers are scaling for multimodal workloads.
Google launches Private AI Compute for regulated industries. A fully isolated AI environment for banks, hospitals, and government agencies.
The US has officially blocked Nvidia’s high-end AI chips from going to China, reshaping the global GPU market and tightening America’s grip on AI leadership.
What did you think of today’s issue?
Hit reply and tell us why; we actually read every note.
See you in the next upload
