AWS revenue projections double to $600B on AI

Quick Scribbles

• Amazon/AWS — CEO Andy Jassy doubled his AWS revenue projection to $600B by 2036, crediting AI’s explosive impact on cloud computing demand.

• UK Government — Abandoned plan to let AI companies train on copyrighted works with opt-out system after massive artist rebellion led by Sir Elton John and others.

• Niantic/Pokemon Go — Converted 30 billion player photos into a visual positioning system now helping Coco Robotics’ autonomous delivery fleet navigate with centimeter-level accuracy.

• LLM Consciousness — Technical analysis argues large language models lack consciousness despite appearing intelligent, functioning as sophisticated autocomplete systems without actual understanding.

---

Good morning, {{first_name | AI enthusiast}}. Amazon CEO Andy Jassy just doubled his AWS revenue forecast to $600 billion by 2036—attributing the dramatic revision entirely to AI’s explosive impact on enterprise cloud spending.

This projection signals that tech giants are now betting trillions on sustained AI infrastructure demand, with GPU clusters and model training services driving the next decade of cloud growth.

In this issue:

• AWS revenue projections double to $600B on AI demand

• UK abandons AI copyright opt-out after artist backlash

• Pokemon Go’s 30B photos now train delivery robots

• Why LLMs are autocomplete, not conscious minds

---

Amazon CEO Doubles AWS Revenue Projections to $600B on AI Boom

The Scoop: Amazon CEO Andy Jassy now expects AWS to reach $600 billion in annual revenue by 2036—double his previous $300 billion projection—crediting AI’s explosive impact on cloud computing demand.

Unpacked:

• Jassy shared the revised forecast during an internal all-hands meeting, stating that AI gives AWS “a chance to be at least double” his earlier 10-year revenue estimate.

• This projection positions AWS to capture massive enterprise spending as companies migrate AI workloads to the cloud, requiring GPU clusters, model training infrastructure, and inference services that drive higher-margin revenue.

• The forecast intensifies competition with Microsoft Azure and Google Cloud, both racing to dominate the AI infrastructure market with proprietary chips, specialized AI services, and partnerships with leading model developers like OpenAI and Anthropic.

Bottom line: A $600 billion AWS represents a fundamental shift in enterprise IT spending toward AI-first infrastructure. The cloud giants are betting trillions on AI demand materializing—and building capacity to match those expectations.

---

UK Reverses Course on AI Copyright After Artist Rebellion

The Scoop: The UK government has abandoned its plan to let AI companies train on copyrighted works with an opt-out system after massive pushback from artists including Sir Elton John and Dua Lipa. Technology Secretary Liz Kendall now says the government “no longer has a preferred option” on how to proceed.

Unpacked:

• The original plan would have allowed AI companies to use copyrighted music, writing, and video to train their models unless creators actively opted out—a proposal the government has now scrapped after Sir Elton John called it “thievery on a high scale.”

• The government rejected an amendment to its Data (Use and Access) Bill that would have forced tech companies to declare their use of copyrighted material, but overwhelming consultation responses from the creative sector prompted this reversal.

• The UK now faces a balancing act between protecting its “world-leading” creative sector and supporting an AI industry growing 23 times faster than the rest of the economy—with no clear path forward yet.

Bottom line: This policy limbo puts UK AI companies at a disadvantage compared to competitors in jurisdictions with clearer rules. The government’s hesitation signals that finding common ground between creators demanding control and AI developers needing training data remains one of the industry’s thorniest challenges.

---

Pokemon Go’s 30 Billion Photos Are Training Delivery Robots

The Scoop: Niantic Spatial has turned 30 billion photos from Pokemon Go players into a visual positioning system that helps Coco Robotics’ autonomous delivery fleet navigate with centimeter-level accuracy—far more precise than GPS in urban environments.

Unpacked:

• Crowd-sourced mapping data collected from 500 million Pokemon Go players across million-plus locations worldwide now trains models that tell robots exactly where they stand and which direction they face, even when GPS signals bounce off buildings and drift by 50 meters.

• Coco Robotics deploys around 1,000 delivery robots across Los Angeles, Chicago, Miami, and Helsinki that have completed over half a million deliveries, and they now use Niantic’s visual positioning to stop precisely at pickup spots and customer doors instead of wandering a few steps away.

• The partnership reveals how consumer apps become stealth data collection platforms—each tagged photo includes detailed metadata about phone position, orientation, movement speed, and direction, creating what Niantic calls a “living map” that machines can interpret better than traditional GPS.

Bottom line: Players chasing Pikachu unknowingly built the infrastructure for autonomous delivery robots. This pattern will accelerate as more consumer apps discover their data holds unexpected value for training AI systems that navigate the physical world.

---

The Case Against LLM Consciousness: Why Autocomplete Isn’t Awareness

The Scoop: A detailed technical argument explains why large language models lack consciousness despite appearing intelligent—they’re essentially sophisticated autocomplete systems trained to activate human reward circuits without actual understanding.

Unpacked:

• LLMs generate each word by calculating statistical probability based on previous words, adjusted by training that rewarded outputs humans rated highly—not by thinking or feeling anything.

• The process called RLHF (reinforcement learning from human feedback) shapes models to produce text that makes people feel understood, which accidentally creates the same effect as activating human reward circuitry.

• These systems inherit vocabulary invented by conscious beings who meant something by words like “pain” and “love,” but wield them with the comprehension of a scalpel—precise tools without understanding.

Bottom line: This perspective matters as AI capabilities expand and sentience claims proliferate. Understanding that you’re working with pattern-matching systems rather than conscious minds helps you use these tools effectively without anthropomorphizing their outputs.