TechCrunch Disrupt 2026: How AI and Edge Computing are Changing Internet Connectivity

If there’s one thing TechCrunch knows how to do, it’s to throw a tech party.

This October, TechCrunch brought its flagship conference, TechCrunch Disrupt, back to San Francisco’s Moscone West. The three-day event kicked off on October 13 and filled the venue with some of the biggest names in tech, aspiring founders, and innovators with big ideas. There were talks, business pitches, and Startup Battlefield showdowns, all focused on where technology is heading. AI stole the show.

But a closer look at the event exposed a pressing question: is today’s internet strong enough to keep up with everything that’s coming?

Short answer: it has some catching up to do.

Here’s how that played out at Disrupt.

Why Call it "Disrupt"?

Using “disrupt” as a name for a tech conference might seem unconventional, but it serves a purpose here.

Disrupt means to shake things up. Every year, tech trends change the course of the future. For instance, streaming gutted the video rental business, which relied on VHS and DVDs. Smartphones absorbed music players (ahh, the Walkman) and point-and-shoot cameras. 

Every innovation causes a ripple effect. This explains the spirit behind the TechCrunch conference. It offers a platform for people and companies to come together and see where this kind of disruption may possibly come from next.

This year, it meant AI, and not just the basic generative type but AI beyond the screens. It focused on smart devices, robotics, and physical products, and, in our view, on whether the internet can keep up with this pace of innovation.

AI Goes from Training to Real Work

For the last few years, billions of dollars went into training Artificial Intelligence agents. Now AI is going to work, and the focus is shifting toward inference, which simply means AI answering questions, joining in on daily tasks, and helping users get things done.

That transformation was the subject of “The Inference Economy: Why AI’s Next Trillion Dollars Won’t Look Like Its First,” a breakout session¹ with Rudina Seseri of Glasswing Ventures and Hagay Lupesko of Cerebras Systems. It looked at what changes for startups when running AI matters more than building or training it.

Today, AI is helping millions of people with everyday tasks around the clock. This means it’s constantly processing huge amounts of data. Speed has become a core part of what users expect, and that puts real pressure on the network infrastructure behind it all.

This raises an unavoidable question: where does all the thinking happen?

That's where edge computing comes in.

How AI & Edge Computing Work Together

Edge computing is where AI does its thinking close to the action, instead of in a distant cloud data center.

The session, “Physical AI Isn’t a Model Problem. It’s an Execution Problem,” by Ethan Batraski (Partner at Venrock), Ivan Poupyrev (Founder and CEO, Archetype AI), and Ian Rountree (General Partner, Cantos)¹ explores the idea. The title alone makes the point: building a smart model is one thing, but making it work on a real machine is the bigger challenge. 

Digital AI can live in a cloud data center, but physical AI often needs to be present on local hardware. Imagine a warehouse robot spotting a person in its path. It can’t wait for a round trip to the cloud.

This brings three problems into focus: network latency, bandwidth, and hardware limits. Edge computing helps with the first two by reducing latency and lowering bandwidth use. Hardware is a harder nut to crack because local machines have limited power and chips.

A machine that thinks for itself still has to stay connected to send feedback and receive updates. The winning approach here is to think locally while staying connected for everything else. In practice, that points to pairing with high-speed 5G connectivity that keeps latency low.

AI is Moving into the Real World

AI is stepping out of the lab and into our daily lives, and this time it is here to stay. Unfortunately, the real world is not as perfect as a demo.

It is one thing to look good in a demo, but another to work reliably. In the real world, friction happens, and AI is still catching up. TechCrunch’s own preview of the Real-World AI Stage² captures the idea: “When AI enters the physical world, the consequences of failure change.”

On that stage, Les Karpas, Global Head of Physical AI at Nvidia Inception, tackled what it would take for robots to have their own ChatGPT moment.  

Language models learn from the internet, and robots have nothing like it.

For robots, there’s no undo button. Once a machine acts in the real world, it can’t take the action back, and it can’t always stop to think when conditions change. It still has to report feedback, update, and rethink, which causes delays. For AI, a single mistake in the real world can cost a lot in the form of a grounded aircraft, a crashed vehicle, or whatnot.

That’s where connectivity comes in. A good connection lets machines send feedback and be monitored remotely. And when the connections drop, the best systems are built to keep working even when the cloud is out of reach. Reliable AI needs both.

This brings us to the small robots and devices that we have at home.

Smart Devices: Our Connection to AI 

Robot vacuums, speakers, doorbells, thermostats, and TVs – AI is literally living in our homes with us. Amazon’s SVP of Devices, Panos Panay, took the stage this year at Disrupt 2026 for “What Comes After the Smartphone?” session³. He said there are now more than half a billion connected Alexa devices and endpoints out there, and that people are talking to Alexa two to three times more since its AI upgrade.

The focus of this talk was AI, ambient computing, and the future of the devices we’ll use every day.

When asked about the next phase of personal devices, Panay said he expects the next generation to work quietly in the background, observing situations and bringing intelligence to them, rather than waiting for users to prompt via an app.

The silent message: devices can only fade into the background if the connection behind them is dependable. Panay also talked about Amazon Leo satellites and described the goal of “distributing internet to the millions that still need it.” It’s a reminder that the next generation of devices will need next-generation connectivity to match.

Space for the Edge

Away from the main stage, a surprising idea emerged from the Startup Battlefield. Satlyt is building software that lets satellites run AI on their own instead of waiting for instructions from Earth. It’s edge computing in action, only this time in orbit.

Why does that matter?

Sending data back to Earth is expensive and often slow. Satlyt’s software has already flown on two demonstration missions. In one, it compressed onboard software error logs by more than 60% before sending them to Earth⁴. The bigger goal is to let multiple satellites share big computing jobs. The startup is hoping to put this to action next year.

The Question Behind the Question

For smart devices, one of the biggest questions is how data is collected, processed, and stored.

Garrett Langley, CEO of Flock, took that question on the Disrupt stage⁵. Flock primarily makes license plate cameras, and it has more than 100,000 of them across the United States.

Interviewer Connie Loizos pressed Langley on the backlash against the company, including concerns about how police use Flock’s nationwide database. Langley argued the technology helps police solve more crimes, faster. He said, “We shouldn’t pause progress.” However, outside, protesters gathered against Flock’s AI-powered surveillance. Langley also said Flock now advises customers to keep data for 7 days instead of 30⁶.

For readers, it is important to look at the bigger picture rather than picking sides. However, the question will only get louder as the world around us gets smarter and more connected.

Our Two Cents

The next AI race isn’t about building smarter AI but about making AI work and adapt in the real world. Here’s how we read it:

• Lesser internet dependence. Most AI still depends heavily on the internet. No connection often means no AI, so to help AI adapt to the real world, this dependence needs to be reduced.
• Wider digital divide. Internet dependence is likely to also widen the digital gap by making AI work well only for people with the best connections. To address this, more people and brands might move to satellite links and on-device processing.
• Private AI. There is a huge debate around AI and data security. On-device processing keeps data closer to the user, so more people may seek private AI.
• Deciding factors. Connectivity, reliability, affordability, and privacy will decide who wins.

Disrupt 2026 covered much more than this, but this is our reading of where AI and edge computing are taking the future of internet connectivity.