CES 2026 and the Shape of the Stack

What Matters When the Demos Fade

The Consumer Electronics Show (CES) has always been a strange ritual. It lands immediately after the holidays, just as everyone is trying to remember how calendars and Zoom work again. Thousands of people descend on Las Vegas to sprint through ballrooms, squint at demos, and declare the future before their coffee has kicked in. I’ve attended enough times to know that, for me, it’s rarely the best way to understand where things are actually heading.

This year, I skipped it. Instead, I did what I increasingly prefer to do with CES. I watched how others processed it. I listened. I read. I let the noise settle.

Tom Merritt and the Daily Tech News Show team were on the ground and, as usual, did an exceptional job covering the breadth of what was announced. The Verge, Reuters, AP, and several others filled in the rest. Between them, there was more than enough signal to work with once you filtered out the gadgets designed mainly to survive a demo floor.

I think that filtering matters. CES still pretends to be a gadget show. It isn’t. We see a variety of announcements that impact mainly consumer electronics but we also get automotive, healthcare, media and other sectors. Likewise there has also been an increase in chip and data center announcements as the enterprise cloud systems drive so much of the underlying innovations these electrics depend on.

What emerged this year had very little to do with breakout devices. It was more about systems settling into place. AI compute continues to spread outward across categories. Discovery shifts out of apps and into operating systems. And delegation is becoming an assumption baked into design, not a feature you opt into.

I went through the coverage with a narrow lens, intentionally filtering out what doesn’t materially affect the media and technology stacks many of us work on every day. When you do that, three themes surface consistently across reporting, commentary, and even skepticism.

Zoom out far enough and those three shifts explain most of what mattered at CES this year for me. The rest is just demo noise.

Compute Is Everywhere. Power Isn’t.

CES 2026 made explicit what last year only hinted at. AI is spreading outward into cars, PCs, TVs, appliances, and edge devices that now assume some form of local inference as a baseline.

NVIDIA’s latest platform announcements reinforced the same architecture they’ve been quietly standardizing for years. Massive, centralized training upstream. Increasingly capable inference everywhere else. AMD and Intel followed the same logic, positioning AI PCs and heterogeneous compute not as premium experiments, but as the new default.

AI is no longer treated as a differentiating feature. It’s getting treated as infrastructure. Vendors aren’t debating whether AI belongs in the device. They’re pressure testing what workloads belong where. We are far from finalizing this - there will be a lot of experimentation and evolution that occurs in this space.

The takeaway message shouldn’t be that edge AI is “winning.” It’s that compute is fragmenting outward while leverage continues to consolidate inward. The hardware may be everywhere, but control over models, training, and defaults remains tightly held.

Training pipelines, model control, and platform defaults remain concentrated among a small group of players. Everything else, no matter how distributed it looks on the surface, is ultimately a distribution strategy layered on top.

Discovery Is Moving Up the Stack

CES didn’t introduce better recommendation engines, but it shows us where discovery and the battle for our attention is relocating.

Announcements from Amazon around Fire TV, Google’s Gemini integrations into Google TV, and Samsung’s continued push toward “AI living” all point in the same direction. Discovery is no longer something that happens inside apps. It’s becoming system behavior.

The Verge’s CES coverage, and especially the Vergecast’s skepticism about “prompting your TV,” inadvertently circled the real shift. They’re right that most people won’t sit on the couch generating content or chatting with their television. But that framing misses the point. The change isn’t about new user behavior. It’s about where decision-making authority now sits.

Discovery is moving out of apps and into operating systems and platforms. Explicit search is giving way to inferred context.

When discovery logic lives at the OS or assistant layer, platforms no longer need users to ask explicitly, whether it’s what to watch, what to share, or what action to take. The system decides what feels timely, relevant, or interruptible before the app ever loads. The longer it has access to your usage patterns, the more natural those decisions begin to feel.

That matters for media companies more than any single CES announcement. Branding and UI lose leverage when content is summarized, suggested, or skipped upstream. Metadata, trust signals, and platform alignment start doing more work than front doors ever did.

CES 2026 wasn’t selling new discovery features. It was normalizing discovery as ambient system logic, increasingly mediated by AI agents.

Agentic and Ambient AI: Delegation Without Consent

A slight tangent to the discovery story is this. CES surfaced a lot of skepticism around agentic and ambient AI, and that skepticism is understandable. Many early implementations still feel awkward. Too chatty or too fragile. And always too eager to help in ways that don’t quite map to real-world use. This doesn’t mean the underlying idea is wrong.

It means product teams are still experimenting with where agentic behavior belongs and how it changes the relationship between systems and users. It does feel like right now we are just throwing LLMs into a service and calling it a feature. Large language models are good at interpreting vague intent and bad at deterministic execution. That tension shows up quickly when conversational interfaces are applied to lights, locks, or appliances, where precision matters more than interpretation.

The result often feels like a regression, not progress. But agentic systems aren’t really meant to do chores. They’re meant to arbitrate what matters now, which system should act, and whether an action is worth interrupting you at all. That becomes clearer in domains where coordination matters more than obedience.

Automotive systems are one example. Several CES announcements focused on in-vehicle AI that adapts media, notifications, and suggested actions based on driving context, time, and inferred attention. These systems increasingly decide what not to surface as much as what to show, prioritizing timing and relevance over completeness. The value isn’t conversation. It’s selectivity.

The same pattern shows up in platforms like LG’s webOS. Rather than improving discovery inside individual streaming apps, LG is pulling recommendation and coordination logic up into the operating system itself. The OS summarizes, surfaces, and suggests across services without waiting for explicit user queries. Intelligence lives in orchestration, not interaction.

Even a much-discussed AI coffee machines fit this pattern, unintentionally. A conversational barista that struggles to make coffee isn’t evidence that ambient AI is a dead end. It’s evidence that teams are still testing how much agency to expose, where automation should live, and when systems should defer rather than act.

Ambient AI isn’t a personality. It’s an orchestration layer. CES didn’t prove agentic AI is ready. It showed that delegation is becoming an assumed design direction, even as companies continue to experiment with how to introduce it, explain it, and earn trust around it.

Closing Note

Put the pieces together and the system comes into focus. Chips are expanding capacity. Discovery is shaping attention. Agents are normalizing delegation.

Seen together, CES 2026 wasn’t really about robots, gadgets, or clever AI tricks for me. It was about defaults that harden at the system level. This is where intelligence lives and where decisions get made. And it’s the area where we have to reflect on how much agency we as users are expected to hand over without being asked explicitly.

As compute spreads outward, control doesn’t follow evenly. Discovery moves up the stack, away from individual apps and into operating systems and assistants that decide what qualifies as an option in the first place. Agentic systems then take the next step, acting on that context by prioritizing, filtering, and deferring on our behalf.

None of this is inherently malicious. But it is consequential. Once defaults set in, they’re hard to see and harder to unwind. They shape what gets surfaced, what gets skipped, and what never even enters the decision space. Over time, they become the system people design for, build against, and quietly accept.

That’s why CES matters, even when the demos don’t. The real signal isn’t in the spectacle. It’s in how the pieces get assembled into the stack we interact with, and who gets to define its behavior. That’s the layer worth watching now.