The New Engine

Abstract: This essay is part of a series on why the content industry is broken and what would fix it. The Well We Never Tapped examined the starvation of the content pipeline. The Heat Death of Your Sci-Fi Universe examined the strip-mining of major franchises. The New Contract and The New Engine examine the broken relationship between platforms and audiences, and propose a new model built from the consumer up.

Disclosure: This reflects my personal experience and interpretation of publicly available information. It represents my views alone—not any employer or organization—and is not professional advice.

The trust dividend

Part 1 rebuilt the consumer contract. Honest pricing. Audience-validated greenlight decisions. Hard walls between tiers. That work produces something beyond a better consumer experience. It produces trust.

Trust matters here because the infrastructure model that can save the streaming industry requires consumer participation at a level no platform has ever asked for. It requires people to opt in. To share resources. To engage with the platform as a partner rather than a vending machine.

That only happens if the contract is clean first.

The waste economy

Think of a datacenter as a giant warehouse full of computers. These warehouses deliver your streaming video. They run 24 hours a day. They consume enormous electricity. They require massive cooling systems. They consume millions of gallons of water.

Most of those computers are barely doing anything. The average server runs at 12-18% of capacity. An estimated 10 million servers sit completely idle. $30 billion in equipment waiting to be used. And idle doesn't mean free. A server doing nothing still draws 20% of its maximum power. Cooling systems that keep these underutilized machines at operating temperature eat 30-40% of total facility power.

This isn't an argument against datacenters. Datacenters are essential infrastructure, and AI workloads are driving massive necessary investment, with spending surging 53% in a single year to $134 billion. McKinsey projects $6.7 trillion by 2030. That growth makes efficient utilization of existing capacity more urgent, not less. The semiconductor industry is building increasingly energy-efficient hardware that delivers dramatically more compute per watt than previous generations. The question is whether the infrastructure using that hardware is being utilized to match.

Your home internet tells the same story. A single 4K stream uses about 25 megabits per second on a 1,000-megabit connection. That's 2.5% of what you're paying for. At 3 AM your connection is 100% idle. Multiply by over 100 million American broadband homes.

$6.7 trillion in projected spending. Existing infrastructure running at 12-18%. Over 100 million idle home connections every night. The capacity to transform streaming delivery already exists. It's waiting to be orchestrated.

Who owns the pipes

There are two classes of streaming company. I know this because I wrote the Netflix Open Connect case study for the FreeBSD Foundation. I documented the system that actually delivers your Netflix stream. I know what it costs and where it breaks.

When you hit play on Netflix, the video comes from a Netflix-owned computer sitting inside or near your internet provider's network. Thousands of custom machines deployed worldwide, optimized for one thing: getting video to your screen fast.

Amazon didn't need to build this. They already had the largest cloud infrastructure on earth. Prime Video rides on hardware Amazon was already running.

Everyone else, Disney+, Max, Apple TV+, Paramount+, is renting. Paying companies like Akamai and Cloudflare to deliver their video. Paying tolls on someone else's highway every time a subscriber hits play.

That's the real divide in streaming profitability. Not content libraries. Not subscriber counts. Who owns the delivery system.

Even the winners face a question. Those Netflix machines are brilliant engineering. But physical hardware ages. And as AI workloads scale rapidly, datacenter capacity becomes more valuable and more contested. Smart allocation of that capacity between AI and content delivery, rather than building separate infrastructure for each, is the efficiency opportunity. Amazon already faces this internally: optimizing the balance between Prime Video delivery and AI services across shared infrastructure.

We solved this twenty years ago

TiVo was, without knowing it, exactly the kind of system streaming needs now.

It sat in your living room. It learned what you liked. It recorded shows before you asked. When you hit play, the content was already there. No waiting. No network traffic. It stored content close to the viewer, predicted what you'd want, and delivered it before you asked.

ReplayTV went further. It could skip commercials automatically. The broadcast industry sued it to death because it threatened the advertising model.

Every principle in the model I'm about to describe existed in a box in your living room in 2002. The industry killed it. Not because the technology failed. Because the companies in power preferred control over efficiency.

Twenty years later, centralized delivery economics are collapsing. And the box in your living room turns out to have been the right idea all along.

A delivery system that already works

A company called Qwilt, backed by Cisco, has been building the new model for years. The concept: instead of streaming from a distant datacenter, store video inside your internet provider's network, close to your home. The provider supplies the hardware. Qwilt provides the software. The provider gets paid for every gigabyte delivered.

This is real, deployed infrastructure. Comcast rolled out Qwilt's system at hundreds of locations in late 2024. By year-end, it covered 55% of US broadband households. Over 175 partners worldwide, reaching over one billion subscribers. Five of the top six US media companies deliver content through it.

Cisco, Qwilt, and Digital Alpha structured it so providers can participate with zero upfront cost. Everyone makes money.

Netflix could reduce its dependence on custom hardware by roughly 80 percent through this model. Run Netflix's software on the provider's equipment. Netflix saves money. The provider earns revenue. The viewer gets a better stream.

Why your internet provider would play fair

Why would Comcast, which owns Peacock, help deliver Netflix? Same reason phone carriers let competitors use their towers. No carrier has towers everywhere. When your T-Mobile phone connects to AT&T's tower, AT&T delivers the call and T-Mobile pays a fee. You never notice. Both participate because dropped calls lose customers.

Same logic. If providers operate as neutral partners with negotiated fees, every service gets delivered. Playing favorites becomes irrational. You earn more being neutral. Qwilt's model already proves it. Comcast earns money delivering Netflix even though Comcast owns Peacock. The economics override the competitive instinct.

Using AI to deliver content before you ask

AI and streaming have opposite schedules. AI peaks during business hours. Streaming peaks evenings. A smart system exploits that gap. Pre-load popular content to local delivery points during dead hours. 2 AM to 6 AM, when AI demand drops and capacity is free. By evening, shows are already nearby. No distant datacenter responds in real time.

AI also brings prediction at scale. The system knows what's popular tomorrow night. New releases. Trending series. Regional preferences. Push content to the right locations before anyone hits play. This is TiVo's idea scaled from one living room to a continent. TiVo predicted what one household wanted and stored it on a single hard drive. AI predicts what millions want and pre-loads it across thousands of delivery points. Same principle. Transformative scale.

Here's where the consumer contract feeds the infrastructure directly. The pledge model from Part 1 generates demand data months before premiere. If 80,000 people pledged for a show, you know where they live. You know their internet providers. You pre-load to those markets before the first episode drops. The audience-validation system creates the data layer that makes predictive delivery work.

Pre-loading at 3 AM also uses electricity that plants are already generating overnight. Content distributed on energy that would have been wasted anyway.

Your router as part of the network

The most radical tier. Your home router, set-top box, or smart hub caches popular shows for your immediate neighborhood. Encrypted at all times. You'd never know your device is helping deliver video nearby.

The incentive: your home serves as a delivery point, you get a credit. $5/month off premium. $3 off broadband. The provider saves on traffic. The platform saves on delivery. You get cheaper service for sharing a resource you weren't using.

This is what peer-to-peer file sharing proved technically twenty years ago but failed commercially. No content protection. No provider cooperation. No consumer incentive. This puts the business model around it. Opt-in. Incentivized. Encrypted. Platform-controlled.

The dependency: you're asking consumers to let their home equipment participate in a delivery network. That requires trust. A consumer who doesn't understand what they're paying for, who feels burned by cancellations, who suspects every price change is a trick, that consumer isn't opting in. They'll assume it's a scam.

The residential model requires earned trust. Trust comes from transparent pricing and a greenlight process that includes the audience. Fix the contract first. Then ask for participation. That's why Part 1 had to come before Part 2. The consumer contract isn't a separate issue from the delivery infrastructure. It's the foundation the infrastructure stands on.

TiVo cached content on a single hard drive for one household. The residential model caches on a gateway for an entire neighborhood. TiVo worked alone. This works as part of a continental network. But the core idea is the same one that box had in 2002: store content close to the viewer, and the distance disappears.

The economic reversal

Three tiers together. More viewers means more home devices delivering content, less load on the provider's equipment, less load on the datacenter. Demand reduces cost instead of increasing it. Every other delivery system in history gets more expensive as demand grows. This one gets cheaper.

The efficiency dividend

The industry is making real progress on datacenter sustainability. Hardware is getting dramatically more energy-efficient per computation. Major operators are achieving 100% renewable energy for facilities under their control. Closed-loop liquid cooling is reducing water consumption.

But the facilities themselves still carry a heavy footprint. A medium-sized datacenter consumes up to 110 million gallons of water annually for cooling. Northern Virginia's cluster consumed close to 2 billion gallons in 2023, up 63% from 2019. In water-scarce regions, cooling demand is becoming a construction permitting constraint.

The edge model extends the efficiency-per-watt logic from the chip level to the network level. Distribute delivery closer to the viewer, reduce the load on centralized cooling, and capture residential infrastructure that's already powered and already idle. Home electricity increasingly comes from rooftop solar and local renewables. Pre-loading content at 3 AM uses baseload generation that's already running.

The most efficient use of datacenter capacity is capturing what already exists before building more. 85% of existing infrastructure is underutilized, already powered, already cooled, already paid for. The edge model doesn't replace datacenters. It complements them by handling the content delivery workloads that don't require centralized compute, freeing datacenter capacity for the AI and high-performance workloads that do.

The new contract, the new engine

Part 1 rewrote the contract. Honest pricing. Audience-validated greenlight decisions. Originals driven by demand signal. The Endless Shrimp replaced with a deal both sides understand.

Part 2 built the engine. Delivery that scales with demand. Three layers: provider-based caching proven by Qwilt, AI-driven predictive loading, and home devices turning 100 million idle connections into the largest delivery network ever assembled. Infrastructure that gets cheaper as it grows. An efficiency model that extends hardware-level gains to the network level, freeing centralized capacity for the workloads that need it most.

But the engine doesn't start without the contract.

Trust is why a consumer lets their router participate. Trust is why a subscriber pledges $50 for an unproduced pilot. Trust is why someone pays $27/month and doesn't feel robbed. Trust is why a viewer votes on animation style for a show that doesn't exist yet.

The infrastructure is the engine. The consumer contract is the fuel.

Heat Death argued you can't reverse entropy in a franchise. You can't. But you can redesign the engine. We solved the delivery problem in a living room in 2002 with a TiVo. We proved the distributed model at continental scale with Qwilt. We know the AI layer is feasible. We know the home delivery architecture works.

What we haven't done is earn the right to ask the consumer to participate. That requires a new contract. An honest one. One where the audience isn't Lando watching the deal get rewritten by Vader.

The current contract was the Endless Shrimp. It looked generous. It was unsustainable. And everyone at the table knew it except the customer.

Time to write a new one. And this time, the audience holds the pen.