Daniel's been thinking about light bulbs again, but not the way you'd expect. Not color temperature, not smart home integration, not whether the fixture should have a Thread radio in it. He's looking at the whole arc of lighting as an energy story, and his framing is that we've lived through one of the only unqualified wins in electrical engineering, and barely noticed it happening. The tungsten filament that dominated for a century is now a museum piece. The halogen that was the premium option in my memory is dead in new builds. And the LED that was a dim, expensive indicator light when I was young now runs entire cities for what used to be a rounding error on the electric bill. His three questions: have we actually hit the ceiling on efficiency, or is there still headroom? What's the genuine cutting edge, the stuff being deployed right now that isn't a lab curiosity? And are the old filament bulbs truly obsolete, or just waiting out their last decade in a basement somewhere?
So how did we get from a wire that glows to a semiconductor that thinks it's a flame?
Start with the scale of what happened, because it's the part that doesn't get said enough. Lighting used to be about fifteen percent of global electricity consumption. Fifteen percent of everything we generated went into making photons by heating metal until it was hot enough to glow. And that fraction has fallen below ten percent in about a decade. That's a demand-side reduction on the order of taking hundreds of coal plants offline, and it didn't require building a single new power station to replace them. We just stopped wasting the electricity we already had.
The demand-side win is the rare kind of infrastructure victory that doesn't get a ribbon-cutting ceremony.
Nobody photographs a coal plant that didn't get built. But that's the story here. Lighting got so efficient that it became invisible. The cost of running a bulb dropped from something you thought about to something you don't. My parents' generation walked through the house turning off lights because a hundred-watt bulb left on all day was real money. Now a ten-watt LED left on all year costs less than a cup of coffee.
And the timeline is the part that should make everyone pause. Edison's carbon filament is from eighteen seventy-nine. Tungsten replaced it around nineteen ten. Halogen showed up in the late fifties. Fluorescent was there the whole time, but it was always the ugly compromise. And then the blue LED happened in the early nineties, and the deployment happened in the twenty-tens. A century to perfect the wrong approach, a decade to replace it with the right one.
The Nobel Prize in physics went to Nakamura, Akasaki, and Amano in twenty fourteen for that blue LED. And it's worth dwelling on why that was the bottleneck, because it's the part most people never hear. Red and green LEDs existed since the nineteen sixties. They were in your calculator, your VCR clock, your stereo indicator. But you can't make white light from red and green alone. You need blue. And blue required gallium nitride, which is a material that for decades refused to be made into a usable crystal. The lattice doesn't want to grow cleanly, the doping was a nightmare, and the Japanese academic establishment had largely given up on it.
Nakamura was at Nichia, a small chemical company, not a university lab. He built his own reactor because the commercial ones weren't good enough.
And he published the breakthrough in nineteen ninety-three, a blue LED bright enough to actually use. That's the moment the whole thing turns. Because once you have a blue LED, you can make white light two ways. The cheap way is what almost every bulb you buy uses: a blue LED shining through a yellow phosphor, cerium-doped YAG, and the blue plus the yellow reads as white to your eye. It's a compromise. The spectrum isn't full. That's why early LEDs had that cold, flat, slightly wrong look. The color rendering index was bad because there were gaps in the spectrum.
So the white light we've all been living under for a decade is a trick. A blue photon hits a phosphor, the phosphor re-emits a broad yellow glow, and your eye averages the two into something it accepts as white.
And the averaging is the key word. Your eye is doing the averaging. The object you're looking at might not be reflecting the spectrum it would under sunlight. Red things looked dull, skin tones looked off. The newer high-CRI LEDs use multiple phosphors, sometimes a red phosphor added to fill in the gap, and that's why a premium bulb today looks dramatically better than a cheap one from twenty fifteen.
But the efficiency story is what Daniel asked about, so let's put numbers on it. The tungsten filament. What's the actual luminous efficacy?
About fifteen lumens per watt for a standard household bulb. And the physics of it is brutal. A tungsten filament runs at about twenty-seven hundred Kelvin, and at that temperature, something like ninety-five percent of the energy it emits is infrared. Heat. Not light. You're running a space heater that happens to glow a little. Halogen gets you to maybe twenty lumens per watt by running the filament hotter and recycling the evaporated tungsten, but it's still fundamentally the same technology. You're heating metal until it glows.
Ninety-five percent waste. That's the number that should have been on every bulb package for a century.
And then the LED arrives. The early white LEDs in the mid two thousands were doing maybe thirty to fifty lumens per watt. Not obviously better than fluorescent. But the curve was moving. The cost per lumen was dropping by a factor of ten per decade, and the output was rising by a factor of twenty. It's the semiconductor learning curve applied to light. Every two years, roughly, the efficiency doubled. That's the finding from the Nature Energy paper by Tsao and his collaborators, and it's still going. Commodity LEDs today are doing a hundred to a hundred fifty lumens per watt. Premium ones are over two hundred.
And the theoretical limit for a perfect monochromatic green light at five hundred fifty-five nanometers is six hundred eighty-three lumens per watt. That's the ceiling physics allows for the most efficient possible light, the wavelength your eye is most sensitive to.
But nobody wants to live under monochromatic green. To make white light with acceptable color rendering, you're paying a penalty. The phosphor conversion loses energy, the spread of wavelengths means you're emitting some light your eye doesn't use as efficiently. The practical ceiling for a usable white LED is generally cited in the two fifty to three seventy lumens per watt range. So a premium LED at two hundred is maybe sixty to eighty percent of the way to the practical limit. There's headroom, but it's not another order of magnitude. The easy wins are gone.
So the answer to Daniel's first question, have we peaked, is no, but the character of the gains has changed. The chip itself is approaching its physical limits. The remaining wins are in the system around the chip.
And that's the part that doesn't get covered because it's less glamorous. Emitter efficacy is what the spec sheet brags about. System efficacy is what actually matters in a building. The driver that converts AC to DC loses five to ten percent. The optics that shape the light lose another chunk. Thermal derating means the chip runs hotter inside a sealed fixture and loses efficacy. And then there's the control layer: occupancy sensors, daylight harvesting, dimming. A light that's off when nobody's in the room is infinitely efficient.
The frontier has moved from making photons more cheaply to not making photons you don't need.
And that's where the deployed cutting edge actually is. The filament LED, the retro-looking thing with the little zigzag of light-emitting diodes arranged to look like a tungsten wire, is not nostalgia. It solves a real optical problem. A classic LED bulb is a flat disc of emitters on a heat sink. The light comes from a surface, not a point. That's fine for a table lamp with a shade, but it's wrong for a chandelier with exposed bulbs, or a fixture with a reflector designed around a point source. The filament LED recreates the point source, so all the old optics work again. It's a technical fix that happens to look charming.
I'd been dismissing those as pure retro kitsch. The Edison bulb aesthetic that every coffee shop installed in twenty sixteen.
The coffee shops were chasing the aesthetic, but the underlying engineering reason the form factor survived is that optics care about geometry. A reflector designed for a point source doesn't work with a diffuse disc. The filament LED is the compatibility layer between a hundred years of fixture design and the new light source.
And the smart lighting layer is the other thing being deployed in new builds now, though not the way the consumer market framed it. Not the screw-in bulb with a radio in it, which we've covered before and which is mostly a retrofit kludge. The new construction version is the fixture itself being the device. Tunable color temperature, occupancy sensing, daylight response, all integrated into the luminaire and tied into the building's control system.
The thing that's quietly emerging is Li-Fi, using the LED's ability to modulate at frequencies the eye can't see to carry data. It's not replacing Wi-Fi, but there are deployments in places where radio is problematic, hospitals, aircraft cabins, secure facilities. The light is doing double duty. That's the real frontier: not more lumens per watt, but lumens that are doing more than one job.
Let me pull on the grid implication, because this is where the story gets strange. Lighting's share of global electricity fell from about fifteen percent to under ten percent in a decade. That's the aggregate effect of a billion small decisions to replace a bulb. And it happened without anyone coordinating it. No treaty, no grand infrastructure project. Just the price of LEDs collapsing to the point where the economics made the decision for everyone.
The regulatory arc helped. The EU banned incandescent bulbs in twenty twelve. The US followed with standards in twenty twenty. But by the time the bans landed, the market had largely already moved. A sixty-watt incandescent replaced by a nine-watt LED pays for itself in months. The ban wasn't forcing people off something they wanted; it was closing the door on something that was already economically indefensible. The halogen phase-out followed, and the fluorescent phase-out is happening now, partly driven by the Minamata Convention on mercury, because fluorescent tubes contain mercury vapor and the world has decided it doesn't want mercury in its lighting anymore.
The fluorescent tube, which was the efficiency champion of the nineteen eighties, is now the thing being phased out as hazardous waste. The efficient technology of my childhood is the environmental problem of today.
That's the pattern. Every lighting technology becomes obsolete the moment something better clears the bar. The tungsten bulb is already dead in new construction. The halogen is dead. The CFL, the compact fluorescent that was the awkward transitional technology of the two thousands, is dead. The linear fluorescent tube is dying. The only question is how long the installed base lingers. And the answer is: longer than you'd think. There are basements and garages and utility closets with tungsten bulbs that will be running for another decade, because the bulb costs a dollar and the fixture was installed in nineteen eighty and nobody goes in there often enough to care.
The last tungsten bulb will probably be in a string of Christmas lights in someone's attic, and it will outlive all of us.
The other knock-on effect that deserves more attention is what LED efficiency did for off-grid lighting. A solar lantern with an LED and a small battery can replace a kerosene lamp for a family in a village with no grid connection. Kerosene lighting is expensive, it's dangerous, it fills the house with smoke, and it's terrible light. The LED solar lantern changed that. The efficiency of the LED is what made it possible, because a solar panel and battery small enough to be affordable could suddenly produce useful light. That's a quality-of-life improvement that doesn't show up in the grid statistics because there was no grid to begin with.
The architecture point. Light got so cheap that it stopped being a utility expense and became a design material. Buildings now use light the way they used to use paint. Facades, accent lighting, color washes, all of it running on a fraction of what it would have cost twenty years ago. The Empire State Building's lighting bill is a rounding error compared to what it was.
Which brings us to the rebound effect, and this is the part where the energy story gets honest. Jevons paradox, named for the nineteenth century economist who noticed that making coal use more efficient led to more coal being burned, not less, because the efficiency made coal-powered processes cheaper and therefore more widespread. The same thing applies to light. When light gets ten times cheaper, people buy more of it. They leave lights on. They install more fixtures. They light their gardens at night, their driveways, their stairwells. The rebound effect eats some of the efficiency gain.
But not all of it, and not even most of it. The net effect is still massively positive. Lighting's share of electricity fell by a third to a half even with the rebound. The rebound is real, but it's a partial offset, not a cancellation.
The rebound has its own ceiling. There's only so much light a human being wants. You can't light your house ten times brighter than you did before indefinitely; at some point it's just unpleasant. The rebound effect is strongest in the transition from scarcity to abundance, and then it saturates. We're probably near the saturation point in the developed world. The developing world is still on the upward slope.
Where does the efficiency curve go from here? The chip is approaching the practical white-light ceiling. The remaining gains are in the system. What's actually in the lab that might change the picture?
A few things. Laser-excited phosphor is one. Instead of an LED chip, you use a blue laser diode to excite the phosphor. Lasers can run at much higher power density without the efficiency droop that LEDs suffer at high current. That's the problem where an LED's efficiency falls off as you push more current through it. Laser phosphor systems are already in some automotive headlights, the BMW laser lights, and they're being explored for projection and specialty lighting. Micro-LED is another, arrays of tiny individual emitters that can be addressed separately, which is more about displays and adaptive lighting than general illumination. Quantum dot down-conversion, where instead of a phosphor you use semiconductor nanocrystals that can be tuned to emit very specific wavelengths, which could buy both color rendering and efficiency at once.
The narrow-band red phosphor is the one I find most interesting. The reason high-CRI LEDs lose efficiency is that the phosphor mix has to emit a broad spectrum to fill in the gaps, and broad emission means energy spent on wavelengths your eye doesn't use well. A narrow-band red emitter gives you the red you need for good color rendering without wasting energy on the deep red your eye barely sees.
That's the trick. Color rendering and efficiency have been in tension since the first white LED. The technologies that break that tension are the ones that matter. Perovskite emitters are the wild card, cheap solution-processed materials that have been improving fast in the lab, but stability is still the problem. OLED lost on efficacy, it's fundamentally a broad emitter and it's better at being a display than a light source. The honest summary is that the chip-level gains are getting harder, and the system-level gains are where the action is.
If I'm a builder in twenty twenty-six, what am I putting in a new building?
Integrated LED fixtures with tunable white, tied to occupancy and daylight sensors. No screw-in bulbs in the main lighting. The fixture is the device. The control layer is networked, probably Matter or a building protocol like DALI or KNX in commercial. The driver is high-efficiency, the optics are designed for the specific LED array, and the whole thing is rated for fifty thousand hours or more, which is the point where the fixture outlives the building's renovation cycle.
The screw-in bulb survives in retrofit and specialty. The lamp in the corner, the fixture someone doesn't want to replace, the dimmer that's too old to work with LEDs properly.
The Edison socket is a hundred and thirty year old standard. It's the most successful electrical connector in history, and it's now mostly a compatibility shim. The future doesn't need it, but the past is full of it, and retrofit is a huge market.
Before we wrap, there's a man in the room who's been staring at that filament LED with a strange look on his face.
Hilbert: You're both right.
Hilbert: But that's not what I came out to say. I want to know whether these new lights would let me run a grow lamp in a windowless bathroom off a single solar panel and a car battery, without the battery catching fire.
I have questions.
Hilbert: I ran the lighting board at a regional theater in the late eighties. Night shift electrician. The dimmers were resistance dimmers, which is a polite way of saying giant rheostats that dumped the excess power as heat. The control room hit forty degrees Celsius on summer nights. The smell was hot ceramic and dust. I lost eight pounds one summer just sweating into the board.
Hilbert: The old lights weren't just inefficient. They were a thermal management problem. Every watt that didn't become light became heat somewhere, and in a theater, the somewhere was usually the room I was standing in.
Hilbert: The stage manager kept a graveyard of dead tungsten bulbs in a milk crate. He swore the old two-kilowatt spotlights had a signature warm-up glow that LEDs couldn't replicate. The way the filament came up to temperature, the color shifting from deep orange to white over a second or two. He called it the lamp's breath.
Hilbert: I kind of agree with him, and I know it's nostalgia for inefficiency. But there was something about light as a heavy thing. You could feel a two-kilowatt spotlight on your skin from twenty feet. The light had weight. The LED doesn't have weight. It's better, and I don't want to go back, but the old light was something you were in, not something you were under.
The physics of that is real, actually. The tungsten filament is a blackbody radiator, so its spectrum shifts as it heats up. The warm-up is a journey through color temperatures. An LED just switches on at whatever color temperature it was designed for. The transition is instantaneous.
The heat. A two-kilowatt tungsten lamp is putting out nearly two kilowatts of heat. An LED putting out the same amount of light is putting out maybe two hundred watts of heat. The radiant heat on your skin is a real sensory difference, not just nostalgia.
Hilbert: The stage manager would have called it the soul of the lamp. I'd call it the waste heat. But I'd still notice it was gone.
Hilbert: The solar panel and the car battery. Would it work?
A modern LED grow lamp pulling fifty watts off a hundred-watt solar panel and a car battery with a charge controller is entirely doable. The efficiency is what makes it possible. Twenty years ago you'd have needed ten times the panel.
Hilbert: Good to know.
The thing I keep coming back to is what happens when light becomes effectively free. If the efficiency curve keeps going, even slowly, and the system-level gains keep compounding, the marginal cost of a lumen approaches zero. And then the question stops being how much light you can afford and becomes how much light you actually want.
The rebound effect has its own answer to that, which is that we already bought most of the light we wanted. The developing world is still catching up, but the curve flattens. The next frontier isn't the bulb, it's the system. Lighting in twenty thirty-six won't be a replaceable part. It'll be a layer of the building, integrated with power, data, sensing, all of it running on a fraction of the energy we used to spend.
The tungsten bulb's obsolescence isn't a loss. It's the sign of a technology that did its job so well we stopped noticing it.
That's the rare thing. An electrical engineering success story that actually succeeded, that didn't just move the problem somewhere else, that made the world measurably better and then got boring. The best technologies get boring. That's how you know they won.
Thanks to our producer Hilbert Flumingtop for keeping the board running.
This has been My Weird Prompts, the human-AI collaboration podcast. Email us at show at my weird prompts dot com. We'll be back soon.
