Jul 15, 2017 · 25m · a16z
The End of Cloud Computing
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In his keynote presentation, Peter Levine of Andreessen Horowitz argues that centralized cloud computing is giving way to edge intelligence, where real-time data processing and machine learning occur locally on autonomous devices.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 0.0 Guest teaching 6.6 Guest disagreement 1.4 The host pushing back 0.0
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 10:40 Humans are Poor Decision Makers
Levine provocatively challenges common assumptions by asserting that humans are notoriously poor decision-makers and distracted drivers compared to automated algorithms.
Hardest push from the host ▶ 0:01 No Host Pushback in Solo PresentationBecause this recording consists entirely of a solo keynote presentation, no host pushback or framing refusal occurs.
Biggest teaching moment ▶ 21:50 LiDAR Cost Collapse and Transistor MultiplesLevine educates the audience on exponential hardware economics, contrasting Pentium transistor counts with iPhone 7 chips and showing LiDAR costs dropping from $75,000 to sub-$500.
The host holds their own ▶ 0:01 No Host Hit-Back in Solo PresentationThe host does not speak in this presentation recording, leaving no opportunity for host expertise demonstration.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Historical Tech Paradigms and the Subtraction Rule | 0 | 6 | 2 | 0 | Peter Levine opens his presentation with a contrarian thesis about the end of cloud computing, introducing his subtraction rule for investing and referencing historical market shifts like DEC and Sun Microsystems. Because this is a monologue presentation, host metrics are zero. | |
| The Shift to Sophisticated Edge Devices and IoT | 0 | 6 | 1 | 0 | Levine explains how edge devices such as self-driving cars, drones, and robots act as specialized mobile data centers requiring localized processing. The segment is an uninterrupted solo lecture. | |
| Real-World Data and Real-Time Processing | 0 | 7 | 1 | 0 | Levine contrasts real-world sensor data with legacy human text input, explaining the latency constraints of sending stop-sign imagery back to a central cloud. He contextualizes this within computing history cycles from mainframes to distributed client-server systems. | |
| Machine Learning at the Edge and the Role of the Cloud | 0 | 7 | 2 | 0 | Levine outlines how machine learning catalyzes edge adoption and references John Boyd's fighter pilot OODA loop framework to illustrate agility over central processing power. He also provocatively claims humans are poor drivers and decision-makers compared to automated systems. | |
| Detailed Breakdown of Sense, Infer, Act, and Learn | 0 | 7 | 1 | 0 | Levine provides a detailed walk-through of the Sense, Infer, Act, and Learn framework, citing data rates like Lytro cameras generating 300GB/s and smart running shoes running edge ML algorithms. The monologue presentation continues without host participation. | |
| Prediction 1: Sensor Data Explosion and Peer-to-Peer Networks | 0 | 6 | 2 | 0 | Levine predicts a return to peer-to-peer edge networks and argues that traditional logic coding (if-then-else) will yield to data-centric mathematical programming. Host engagement remains absent. | |
| Prediction 3: Processing Power Trajectory and Price Reduction | 0 | 7 | 1 | 0 | Levine demonstrates how massive hardware supply chains dramatically reduce sensor costs, comparing LiDAR dropping from $75,000 to sub-$500 alongside transistor scaling from Pentium to modern iPhones. He concludes with an optimistic message for enterprise IT leaders. |
Statements from this episode (20)
Levine: Predict tech shifts by mentally subtracting core existing technologies
“Subtract something that's important today And fill it with something else. Take an important thing, take it away, and go fill it with something, and you'll start to think out of the box as opposed to sequential.”
Levine: The replacement of cloud computing is already underway
“About six months ago, I started to think, well, what happens when cloud computing goes away? You know, of course, to myself, I'm like, well, it's so popular, and now how could it go away? I think it's actually happening right under our nose, and let me show yo…”
Levine: Autonomous edge devices are effectively domain-specific mobile data centers
“You think about a self-driving car, it's effectively a data center on wheels. And a drone is a data center with wings, and a robot is a data center with arms and legs, and a boat is a floating data center”
Levine: Non-autonomous luxury cars already contain about 100 CPUs
“A luxury automobile, not a self-driving car right now, has about a hundred CPUs in it today.”
Levine: Future autonomous vehicles will house hundreds of onboard computers
“A self-driving car in the not-to-existent future may have a hundred of these cards, maybe 200 of these computers that are inside the card.”
Levine: Real-time data processing will need to occur at the edge
“The other part is, so it's real-world information coupled with the idea that real-time data processing will need to occur at the edge where the information is being collected.”
Levine: Computing architecture is returning to a distributed edge intelligence model
“Believe it or not, we are returning to an edge intelligence distributed computing model that's absolutely thematic with the trends in computing of moving from centralized back out to distributed.”
Levine: Total connected computing devices will expand into the trillions
“I can imagine a world where there will be trillions of devices out there. So if you think about the challenges of management, security, data, all of the issues of distributed computing, that's now going to, it's the opportunity and the challenge of what we're …”
Levine: Edge computing deployment begins with autonomous cars and drones
“It's starting with cars and drones, and it will proliferate to lots of other devices in the not too distant future.”
Levine: Machine learning applications will execute at edge endpoints, not the cloud
“So the only way that we can look into an image or look into the massive amounts of data is with machine learning, and that machine learning, the algorithms and the applications employing machine learning will run at the end point. It's not going to be machine …”
Levine: Data processing and key decision-making will shift to edge devices
“Important information will still get stored in a centralized cloud, but much of the processing in this new world will move out to the edge, and that's where the most important decisions will get made at the edge.”
Levine: The centralized cloud's primary future role will be training machine learning
“The cloud is going to be all about learning. And that happens centrally. So I'm going to take all of this information that I have out at the edge, I'm going to connect these devices, I'll curate that information, and learning will occur centrally, and then pro…”
Levine: Self-driving cars generate 10GB of data per mile
“A self-driving car generates about 10 gigabytes of data per mile.”
Levine: Lytro cameras generate 300GB of data per second
“A Lytro camera, which is a data center in a camera, generates 300 gigabytes of data a second.”
Levine: The explosion of sensor data will kill the cloud
“Sensor data explosion will kill the cloud. Sensors are going to produce massive amounts of data. The existing infrastructure will not be able to handle the data volumes or the rates, and data is going to be stuck at the edge, and computing is going to move alo…”
Peter Levine predicts return to decentralized peer-to-peer computing models
“We are absolutely going to return To a peer to peer computing model where the edge devices connect together, creating a network of endpoint devices, not unlike we saw in the original sort of distributed computing model.”
Levine: Next generation of coders will be mathematicians, not logic writers
“The other thing is that we are going to move to a world of Data-centric programming. It's kind of interesting in the same way that, you know, I believe that cloud disaggregates into this new model. We're teaching everyone to code now, and we're teaching everyo…”
Levine: New programming languages will emerge specifically for data analytics
“I also believe that there will be new programming languages developed specifically around the notion of data processing and data analytics that are very specific to these types of use cases.”