Nye's Digital Lab is a weekly scribble on creativity in an age of rapid change.
Most of my career was spent making movies nobody was allowed to see.
Previsualization (known as 'previs' for short) is the rough draft of a film, built in 3D before anyone builds anything real. Gray blob actors. Cameras that fly through walls. Every big movie on my resume, and there are a few big ones, went through a version of itself that looked like a video game from 1998.
The measure was speed not quality.
In finals, a visual effects artist might get weeks on one shot. In previs you're doing at least shot a day, often times, many more. You learn to stop being so friggin' precious.
That pace did something to the economics that I don't think anyone really planned. The most expensive mistake in visual effects is finding out a shot doesn't work after you've built it. Previs let a director find that out in gray boxes, in a day, for almost nothing. So effects got cheaper per shot. A lot cheaper.
But if you notice, the movies did not get cheaper. Not once. The studios pocketed the savings and bought more stuff. More shots, bigger sequences, whole third acts made of explosions of pixels. The effects budget went up while the price per shot went down. We spent the same and got absurdly more.
I bring this up because the next version is aimed right at the next generation.
Let's talk about televisions.
In 1960 there were twenty-seven American companies making TVs. Zenith and RCA had about a fifth of the market each. By the end of 1987 there was one American TV maker left.
Japan didn't win with a better picture tube. It won with the unglamorous stuff. Scale. Owning the whole supply chain. Building so many factories they had to export to keep the lines running. Sony's first pocket radio hit the States in 1957 for $39.95, and Akio Morita turned down a giant order to sell it under an American store's brand. He wanted his own name on the box.
America tried to stop it. First came accusations that Japan was dumping cheap sets below cost. Then, in 1977, a deal that capped Japanese TV imports at 1.75 million a year. But, It didn't matter. Sony had been building televisions in San Diego since 1972. Matsushita bought Motorola's TV business and kept selling the same sets under the name Quasar. A set built in California isn't an import. The wall went up, and the factories just walked around it.
Zenith is the one you should be looking at. Consumer Reports rated its sets among the best built all through the seventies. The quality was real. And in the middle of that, Zenith laid off half its research engineers. It kept the name on the box and fired the people who knew how to make the next box.
OK, now let's do cars. Chinese Cars.
BYD sells an electric car in China for under ten thousand dollars. Chinese battery packs run about a third cheaper than ours, and Chinese factories can make roughly three times the batteries the planet buys. Europe put up tariffs to stop them. The US put up a hundred percent tariff! Volkswagen's margin dropped under one percent for a quarter. And just like Quasar did in the 70's with TVs, the factories are walking around the wall again. One of them, Hyundai, is a short drive from my house.
I've watched this movie. I know how this goes.
How about some numbers and some predictions, Nye?
Since late 2022, the price of a fixed unit of AI intelligence has been dropping about ten times a year. Not 10%, Ten TIMES! What cost twenty dollars then costs forty cents now. Batteries are supposed to be the miracle of cheap. They fell eight percent last year.
And the same players are running the same play.
By this spring, about six in ten of the AI words moving through the biggest independent model marketplace came from Chinese open models. Nearly as good, a tenth of the price. Every update from the frontier seems to get an open Chinese version, merely weeks after.
When the first cheap Chinese model dropped, Wall Street panicked and sold Nvidia. Wrong target, folks. The chip makers are fine. The people getting squeezed are the brand names in the middle, the ones who thought the name on the box was the business.
Animation is where my heart bleeds. Jeffrey Katzenberg once said it took 500 artists five years to make a world-class animated movie, and last year he said that he expected that to shrink by ninety percent. OpenAI is backing a feature made in nine months for under thirty million dollars. Netflix quietly opened a generative animation studio in March. Raw AI video runs somewhere between thirteen and two hundred dollars per finished minute, depending on how many tries it takes to get the shot right. Which, by the way, is the previs question.
How many tries do we get?
So where does it go?
I have guesses, and they're not equally likely.
The studios could pull a Zenith. Keep the brand, keep the quality rating, lay off the people who invent the look, and slowly become a nameplate on somebody else's work. Disney's cuts this spring hit Marvel's visual development team, who are the people who decide what a movie looks like. That's the Zenith move, and I don't think they're doing it on purpose, which is worse.
Or it goes like previs. Nobody spends less. Everybody makes more. The big employers shrink and the number of people who can afford to make animation explodes. A kid in Statesboro with a laptop and a story. I kinda lean this way, because I've seen it happen once.
But the version I'd bet on is the one Sony understood and Zenith didn't. Whoever keeps the people who know how to build the next thing gets to build the next thing. Price has nothing to do with it. The next thing is never cheap, because nobody has learned it yet.
Which is a long way of telling my students that the shot is not the job. When the shot costs nothing, the value moves to whoever people trust to pick the shot. The taste. The judgment. The name on the box, only if you've earned it.
And to think. I used to think a shot a day was fast.
Thanks for reading.
Make it Happen.
This essay is the synthesis of a collection of verbal improvisations that were recorded and then refined with generative models. Additional revisions were done with human writing. Thanks to Roger Warburton for the Japanese-TV comparison.
Nye Warburton is an artist and educator from Savannah, GA.
For more information: nyewarburton.com
My Heart Bleeds for Hollywood — The same squeeze, one industry earlier. They'll always tell stories. They just might not do it from Los Angeles anymore.
The Open Source Strategy — Written when DeepSeek first blew the barn door off. The thinking play, before it had a number attached.
The Search for Authentic Personal Value — What's left when technical execution is irrelevant. The personal side.
Verified 2026-09-24.
Previs
The Third Floor: secrets of previs — Creative Bloq — the "shot a day" quote, Duncan Burbidge
The Third Floor origin story — Post Magazine, March 2013 — founded 2005 after Revenge of the Sith, 10x growth
Iron Man 2 previs — Autodesk, 2010 — ~700 shots, 3 to 4 iterations each
Zenith
The Day the U.S. TV Industry Died — IEEE Spectrum, Tekla S. Perry, Aug 1988 — 27 makers in 1960, Zenith and RCA >20% each, only Zenith left by end of 1987, 1977 R&D halving (~200 engineers), Consumer Reports ratings through the 1970s, Matsushita buys Motorola/Quasar 1974, LG acquisition 1995/1999
Sony corporate history, Chapter 3 — TR-63 at $39.95 and Morita refusing the private-label order
Presidential Proclamation 4511 (Carter, 1977) — Orderly Marketing Agreement, 1.75M units a year
Sony Technology Center, Rancho Bernardo — Wikipedia — opened Aug 3, 1972, first foreign-owned TV plant in the US
Japan's Rapid-Growth Policy on Trial: The Television Case — Yamamura & Vandenberg — dumping history, import share curves
BYD Seagull cut to 69,800 yuan (~$9,700) — Reuters, March 2024
Lithium-ion pack prices fall to $108/kWh — BloombergNEF, Dec 2025 — China $84/kWh; North America and Europe 44% and 56% higher
Global EV Outlook 2025, batteries chapter — IEA — cell capacity >3 TWh, ~3x demand, China ~85%
Definitive countervailing duties on Chinese BEVs — European Commission — BYD 17%, Geely 18.8%, SAIC 35.3%
Volkswagen crisis, VW brand Q2 margin 0.9% — Reuters, Sept 2024
Tokens
LLMflation — Andreessen Horowitz — cost of equivalent LLM performance falls ~10x per year
Inference unit economics — Introl — GPT-4-class at $0.40/M tokens vs $20 in late 2022
Silicon Data LLM Token Expenditure Index at $0.97/M tokens, Sept 1, 2026 — via KuCoin
China's Open-Weight Takeover — Wing VC — ~61% of OpenRouter tokens from Chinese open models by May 2026; DeepSeek ~12x under GPT-5.5; Nvidia panic "backwards"
Chinese open models and enterprise AI cost — ecorpit, citing CNBC, July 7, 2026 — US firms routed up to 46% of OpenRouter tokens to Chinese open models
What to Know About Chinese AI Models — CSIS — Chinese models 41% of Hugging Face downloads
Hollywood animation and VFX unions fight AI — Context / Thomson Reuters Foundation, 2024 — Katzenberg "500 artists five years" quote
OpenAI backs $30M AI animated film Critterz — WebProNews — under $30M in nine months (stated budget, not audited)
Netflix INKubator, launched March 2026 — Metaintro, citing The Verge, May 14, 2026
AI video generation cost — LTX blog — $13 to $220 per finished minute, iteration rate as main driver (vendor source)
Marvel's director of visual development Andy Park laid off — Variety, April 2026
Disney ~1,000 layoffs, April 14, 2026 memo — Hollywood Reporter
Animation industry statistics 2026 — Vidico — LA employment declines, Animation Guild layoffs, "AI not yet the cause"
AI Won't Replace VFX Artists. It's Replacing Who Can Afford Them. — HackerNoon — LA County lost 6,700 motion picture jobs to May 2026; buyer pool larger