UHD delivery: why streaming won, and why I still miss the dish

I deliver video over the internet for a living, so you’d expect me to be 100% on team streaming. Mostly I am. But I want to start this post with a confession: I have a real fondness for satellite delivery, and I think the industry left something on the table when it walked away from the dish.

Let me make the case for both sides (with receipts) and then explain why a £1,200 television has me thinking about all of this again.

Satellite was the UHD early adopter. It really was.

There’s a lazy narrative that satellite was the legacy tech that streaming disrupted. On UHD specifically, the timeline says otherwise.

Eutelsat put Europe’s first dedicated Ultra HD demo channel on a satellite in January 2013: 4K at 50 frames per second, transmitted at 40 Mbit/s from EUTELSAT 10A.¹ In 2013, mainstream home broadband could not dream of a sustained 40 Mbit/s video stream; Netflix wouldn’t begin 4K streaming until the following year, at less than half that bitrate.

And then there’s my favourite piece of broadcast history: NHK BS8K, the world’s first 8K television channel, launched via satellite on 1 December 2018, carrying 7680×4320 with 22.2 multichannel sound, beaming footage from Antarctica and the ISS into Japanese living rooms.² The day after launch, NHK and Eutelsat transmitted a live 8K broadcast from the Vatican to Tokyo.³ Nearly eight years later, no streaming service offers anything comparable as a linear proposition.

This is the counterfactual I wish had played out: satellite as the 8K pioneer and, more importantly, as the guarantor of wide 4K penetration. A transponder doesn’t care whether one household or ten million are tuned in. Broadcast is the ultimate multicast: the marginal cost of an extra viewer is zero. For pushing a small number of pristine, high-bitrate channels to an entire country, physics still favours the dish.

So why did the industry go the other way?

Because delivery cost per channel isn’t the whole equation, and everything else favoured the internet.

The consumption data is unambiguous. In May 2025, Nielsen’s Gauge recorded streaming at 44.8% of total US TV usage, overtaking broadcast and cable combined for the first time.⁴ By December 2025 streaming hit 47.5%, the largest share ever reported.⁵ Since Nielsen began the monthly report, streaming viewing has grown 71% while broadcast and cable declined 21% and 39% respectively.⁴

The structural reasons will be familiar to anyone in CDN land:

• On-demand beat linear. Satellite excels at one-to-many simultaneity; audiences stopped wanting simultaneity. A transponder can’t serve ten thousand different titles to ten thousand households at once. An edge cache can.
• Unicast economics improved; transponder economics didn’t. Shifting 4K over an orbital link is expensive per channel, per month, forever. Meanwhile CDN egress costs kept falling and codecs (HEVC, AV1) kept shrinking the bits.
• The upgrade path is software. Adding HDR, a new codec, or a new resolution tier to a streaming platform is a deploy. On satellite it’s new receiver hardware in millions of homes, with the same certification-and-set-top-box friction that slowed DVB UHD rollouts for a decade.
• Even the satellite operators agree. Sky, the UK’s satellite institution, has stopped installing dishes for new products, pushing Sky Stream and Sky Glass instead, with the satellite fleet ageing toward retirement.⁶ Channel counts on the Sky satellite platform keep shrinking as broadcasters go streaming-only.⁷ Tellingly, even on Sky Q the UHD content largely arrives over IP on-demand, with satellite carrying the HD linear feeds.⁷

Streaming won for good reasons. I just note, slightly wistfully, that it won the distribution argument without ever matching satellite’s early quality ambition.

What streaming does that a dish never could

There’s also a capability gap that pure delivery economics understates. Streaming is a two-way conversation: per-viewer sessions enable interactive watching: multi-angle sport, watch parties, live stats overlays, “skip intro”, start-over from the server. The features live in the platform, upgradeable with a deploy.

Satellite’s answer to even basic trick-play was elegant in a very particular way: make it the client’s problem. Pause, rewind and recording on Sky+ worked by writing the broadcast transport stream straight to a local hard disk, entirely achievable even with the modest drive write speeds of the era, because a broadcast mux is a fixed, predictable bitrate.⁸ It worked brilliantly, but notice the architecture: every “feature” satellite gained was really a feature of the set-top box, bolted onto a one-way pipe that couldn’t participate.

And the pressure keeps mounting on the streaming side too, to be fair. Broadcast chains deliver live pictures 3–7 seconds behind reality; standard HLS/DASH streaming has historically sat 20–60 seconds back, and closing that gap with LL-HLS and LL-DASH into the 2–5 second range is hard-won engineering, not a free lunch.⁹ Stack HFR on top (100/120fps sport doubles the frames you must encode, package and ship per second) and modern streaming is being asked to match broadcast’s latency and exceed its fidelity while still being unicast. And the bandwidth problem doesn’t stop at the edge: in the living room, 4K at 120fps needs HDMI 2.1 end to end, including a cable rated for it, because HDMI 2.0 tops out at 4K60. Nobody said winning was cheap.

…but satellite learned some remarkable tricks

Before anyone writes broadcast off as a dumb pipe, it’s worth recording how clever the one-way world got.

Targeted ad insertion, over broadcast. Sky AdSmart quietly delivered a library of adverts over satellite to the Sky+HD box’s hard disk, profiled the household locally, and spliced the best-matched ad into the live break in place of the national spot.¹⁰ Per-household ad targeting, the thing usually held up as streaming’s killer commercial feature, done with a one-way satellite link and some local storage. That’s genuinely great engineering.

Out-of-hours content push. Sky Anytime, launched back in March 2007, pushed the week’s big programmes to a reserved 140GB slice of the Sky+ HD box’s disk overnight, giving the effect of video-on-demand without a single unicast byte.¹¹ Pre-positioning popular content at the true edge (the viewer’s own hardware) during off-peak hours is exactly what we now call cache pre-warming, except it also kept the load off ISP networks entirely. Every CDN engineer who has capacity-planned for a season finale should tip their hat.

That’s the pattern with satellite: it kept finding ways to fake interactivity with broadcast plus local storage. Impressive, but always an approximation of the thing streaming simply is.

The bridge: Media over QUIC

Here’s the part that makes me optimistic the gap between the two worlds actually closes. And yes, declared interest, my employer is invested in this space.

Streaming’s remaining weaknesses against broadcast all trace back to one architectural decision: we deliver live video as files. HLS and DASH chop the stream into segments, players poll manifests and pull the next chunk, and every one of those steps adds latency and request volume. That’s why broadcast still beats an unoptimised stream to the goal celebration by half a minute. Media over QUIC (MoQ) attacks the problem at the transport layer instead: media becomes a continuous stream of frames, published once and subscribed to, with QUIC carrying video, audio and metadata as independent streams that don’t block each other and, crucially, the freedom to skip less-important media during congestion rather than stall.¹²

Look at what that pub/sub model actually is, though. A publisher sends the stream once; a relay tree fans it out to everyone watching; nothing is encoded or shipped until a subscriber asks.¹³ Latency at the live edge, one-to-many fan-out, viewers joining a broadcast rather than opening a private file-transfer session: that is transponder semantics, rebuilt on the open internet, with the two-way interactivity satellite could never have. The stale-object dropping even gives you broadcast’s “you see now, not thirty seconds ago” behaviour, something ABR-over-TCP fundamentally can’t promise. And it’s not lab-ware: the IETF working group is standardising moq-transport, open-source Rust and TypeScript implementations run in browsers today via WebTransport and WebCodecs,¹³ and Fastly has said it expects to play an active role as the ecosystem evolves.¹²

If MoQ delivers, the trade-off this whole article is built on (satellite’s effortless one-to-many versus streaming’s interactivity) stops being a trade-off. Streaming wouldn’t just have beaten broadcast; it would have absorbed it. Those 98-inch panels demanding 4K HFR live sport at broadcast latency are exactly the workload MoQ is being built for.

The uncomfortable bit: 4K still isn’t commonplace

Here’s what grates. The dish is being decommissioned in favour of a delivery model that, years in, still treats 4K as a premium add-on rather than table stakes.

Most major streaming catalogues remain majority-1080p, and the services are coy about exactly how much 4K they carry.¹⁴ Where 4K exists, it’s frequently paywalled: Netflix reserves it for the Premium tier, HBO Max for its top plan.¹⁴ If you’re paying for those tiers, play them through something that does them justice: the current Apple TV 4K handles Dolby Vision, HDR10+ and Dolby Atmos.¹⁵ Broadcast television, meanwhile, is still overwhelmingly 1080i.¹⁶ Live sport in 4K remains an event-by-event novelty rather than a default.¹⁶

Contrast that with the hardware: roughly 72% of TVs sold in the US are now UHD-capable, and the large majority of households own at least one 4K set.¹⁷ The displays raced ahead; the content pipeline never caught up. Satellite could have carried a guaranteed, always-on 4K linear tier a decade ago. The 2013 Eutelsat demo proved the pipe was there. Instead we got a patchwork of per-service, per-tier, per-title upgrades.

The “HDR + 1080p is enough” argument, and where it breaks

Some smart people argue the industry quietly settled on the right compromise: 1080p with HDR. And on the merits, at typical screen sizes, they have a point. HDR’s gains (contrast, colour volume, highlight detail) are visible from across the room, while extra pixels are not. At a normal 3-metre living-room distance on a 55-inch panel, most viewers can’t reliably distinguish 4K from a good 1080p picture; the resolution advantage only appears when you sit within about two metres.¹⁸ A well-graded 1080p HDR stream frequently looks better than a 4K SDR one, and plenty of the perceived “4K wow” in shop demos is actually HDR doing the work.¹⁸

So yes: as a middle ground for the 55-inch world, HDR-first was defensible. But that world is ending, and it’s ending fast.

The 98-inch plot twist

Giant TVs have collapsed in price. In the UK you can now buy a 98-inch TCL mini-LED for around £999–£1,375 (TechRadar covered a Costco deal on the 98Q9BK at £999, down from a £3,250 launch price¹⁹) and mainstream retail sits comfortably in the £1,200–£1,600 band (roughly €1,400–€1,900).²⁰ That’s not videophile money. That’s “decent 65-inch OLED” money, buying more than double the screen area.

If I were buying one now, I’d look at TCL’s 98C6K, a 98-inch QD-Mini LED panel. A step down in size, Hisense’s UR8S is a 75-inch RGB Mini LED. And if you’d rather have OLED contrast than sheer size, LG’s C6 is the one I’d pick, though at 77 inches you’re paying more for less screen.

And at 98 inches, the viewing-distance maths flips. The angular resolution argument that let 1080p off the hook at 55 inches stops working when the screen fills a third of your wall. At a normal sofa distance, a 98-inch panel puts you well inside the range where 4K detail is plainly resolvable, and where a 1080p stream starts to look soft, HDR or not. The middle ground was a function of screen size, and screen sizes just moved.

Which means the pressure lands back on delivery, my day job. Bigger panels demand real 4K at honest bitrates, which means more bits per session, which means the CDN layer matters more, not less: deeper edge caches, better codecs, smarter request collapsing for live events. The great irony being that live 4K sport at scale (millions of concurrent viewers, identical content) is precisely the workload satellite was born for and unicast streaming finds hardest. Every CDN engineer who has watched a live-event traffic graph knows we spend enormous engineering effort approximating, with caches and multicast-ish tricks, what a transponder did for free.

The verdict, and a modest proposal for the dish

Streaming wins. On choice, on economics, on capability, on the Nielsen curve that only points one way: a two-way, software-defined delivery platform beats a one-way pipe, and I say that with my CDN engineer hat firmly on.

But satellite’s fair ending shouldn’t be quietly losing channels while the dishes come down. Its natural role today is a premium-positioned product: a small number of guaranteed, contention-free, maximum-bitrate channels (flagship live sport in genuine 4K HFR, cinema channels at Blu-ray-class bitrates) sold on the one property unicast streaming can never fully promise, that picture quality doesn’t degrade when twenty million neighbours tune in. The 2013 Eutelsat demo and NHK’s BS8K proved the pipe could do it. Position it the way 4K UHD discs are heading: not the mass medium, the connoisseur’s one.

Coda: the empire strikes back?

One last irony to file away. The talk of the industry right now is data centres in space. Starcloud has already run an Nvidia H100 in orbit and raised a $170M Series A to build more,²¹ and Google’s researchers put the crossover where orbital compute becomes cost-competitive with terrestrial at launch costs of roughly $200 per kilogram (Starship’s target, down from ~$2,700/kg on Falcon 9 today).²²

If that threshold is met, there’s nothing in principle stopping VOD libraries (or even live origins) being served from orbit. Which would mean the industry that spent two decades moving video delivery off satellites ends up moving the servers onto them, this time with vastly more overhead. Pun very much intended.

Pour one out for the transponder. Then go check your ABR ladder tops out where it should.

My recommendations

TVs

• TCL 98C6K 98-inch QD-Mini LED
• Hisense UR8S 75-inch RGB Mini LED
• LG OLED evo AI C6 77-inch (OLED77C65LA)

Streaming

• Apple TV 4K (4th generation): Dolby Vision, HDR10+ and Dolby Atmos

Cables

• CAKOBLE HDMI 2.2 cable, 3m

---

Written with AI assistance, then fact-checked and edited by me. All figures are linked to their sources in the footnotes. Amazon links are affiliate links; I may earn a small commission if you buy through them. There are no intrusive ads on this site, so affiliate links are its only source of revenue.