Jul 15, 2017, 7:44 PM ETConference Presentation, Keynote, Lecture

The End of Cloud Computing

Cloud computing is predicted to reach its functional limit as the dominant model, not because the technology fails, but because the paradigm shifts to "edge intelligence" where processing occurs at the source of data.
The speaker applies a "Forrest Gump rule" for prediction: subtract a current dominant technology (centralized cloud) and replace it to reveal non-sequential future opportunities.
This shift represents a cyclical return to distributed computing, echoing the transition from mainframes (centralized) to client-server (distributed) in the 1980s/90s, now evolving via the Internet of Things (IoT).
Future computing will be defined by a distributed model where endpoint devices (cars, drones, robots) act as decentralized data centers rather than simple terminals.

Device Transformation: IoT objects (self-driving cars, drones, robots) are evolving into "data centers on wheels/wings/arms," capable of hosting hundreds of CPUs internally; a future self-driving car may contain 100–200 separate circuit boards.
Data Volume: Self-driving cars generate approximately 10 gigabytes of data per mile, while advanced sensors like Lytro cameras can produce 300 gigabytes per second.
Latency Constraints: Centralized cloud processing is too slow for critical real-time decisions; a vehicle traveling at speed would crash before a cloud-based decision (e.g., detecting a stop sign) could be processed and returned.
Data Shift: Computing is moving from human-generated text/log data to massive volumes of unstructured, real-world sensory data (vision, acceleration, temperature, gravity) collected continuously by embedded sensors.

The OODA Loop: Edge computing prioritizes agility over raw power, mirroring the military "OODA loop" (Observe, Orient, Decide, Act) where faster processing loops secure victory over slower, more powerful centralized systems.
Three-Step Edge Process:
- Sense: Deployment of ubiquitous sensors (cameras, radar, accelerometers) in diverse objects, including running shoes and infrastructure.
- Infer: Machine learning algorithms run locally at the endpoint to extract relevance and recognition from unstructured data without cloud dependency.
- Action: Immediate physical responses are executed at the edge to ensure safety and responsiveness.
Role of the Cloud: The cloud shifts to a "training center" where curated data from billions of edge devices is aggregated to train global machine learning models.
Feedback Loop: Trained models and updated intelligence are propagated back from the cloud to edge devices, creating a continuous, tightening cycle of improvement.

Supply Chain Commoditization: The scale of producing trillions of IoT devices will drive down sensor and processing costs; LiDAR costs are projected to drop from $75,000 to $500, and eventually to 50 cents.
Market Expansion: The total addressable market expands beyond the 7 billion human population to "trillions of devices," creating a massive new ecosystem for management and security.
Talent Transformation: The industry will shift from "data-centric programming" (mathematicians and data analysts) to traditional "logic-based" (if/then/else) coding, necessitating new programming languages specific to data analytics.
IT Sector Disruption: IT managers and CIOs face an expansion from managing billions of mobile devices to coordinating trillions of interconnected peer-to-peer endpoint devices across all industries.

Networking and Security: New challenges and opportunities will arise from connecting trillions of devices in a peer-to-peer network without centralized data pools.
Cost and Power Trajectory: Processing power will increase by several orders of magnitude while costs plummet, enabling sensor integration into everyday items like footwear, glasses, and clothing.
Industry Applications: Disruption will span all sectors, including insurance (drones for house inspection), healthcare (remote surgery via robots), and consumer tech (smart footwear with gait analysis).
Programming Evolution: New software languages will emerge specifically designed to handle the complexities of distributed data processing and edge analytics.