For fifteen years, above 2 kW the answer was HMI. LED climbed, but it did not climb that far. Two fixtures have just changed that: the Aputure Storm XT52, announced at 5,200 W, and the Nanlux Evoke 5000B at 5,000 W.
I have not had either on a set. What follows reasons from the published figures, sourced at the bottom, and from what those figures mean when you are putting an order together. The feel of using them belongs to the people who have shot with them.
What the stated equivalence covers, and what it does not
Both manufacturers point at the same opponent: a 9 kW HMI. Nanlux presents the Evoke 5000B as the equivalent of a 9 kW HMI or a 24 kW tungsten fixture. Aputure places the Storm XT52 against traditional 9,000 W heads.
For scale, an ARRI M40 is 4 kW and an M90 is 9 kW. This is the top of the package, the head you pull out to punch a window, hold a day exterior or light a background from a distance.
That equivalence is about illuminance at a point: the lux measured at beam centre, at a stated distance and angle. It is a real, checkable measurement. It is not output.
The total luminous flux of these LEDs stays well below an M90. At a comparable beam angle, the average across the whole lit area is not equivalent: the centre matches, the edges do not. An efficient optic concentrates the available flux into a narrower cone and produces a strong centre figure without making any more light.
Centre-beam lux is not output
A fixture can show the same centre-beam illuminance as another while making noticeably less light. Lux measures what lands on a point. Lumens measure what leaves the source, across the whole beam.
Concentrate the flux into a tighter cone and the centre reading climbs, without a single extra lumen being produced. That is what a good optic does, and it is entirely legitimate. The trouble starts when that number gets read as a blanket equivalence.
Even at an identical stated beam angle, the distribution inside the beam differs. A high-output LED tends to be peakier: it catches the HMI at the centre and falls away at the edges. The HMI fills its cone better. On the graph the two curves meet at the centre, but the area under the curve, which is the output, is not close.
Half the current, and time back
The clearest gain is electrical. Going from ten kilowatts to five on one position means one less line to run, or a generator you can drop a size. On a location where the supply is counted, that turns straight into budget.
The second gain is time. An LED switches on and gives you its light. No strike, no warm-up, no refusing to restrike hot because somebody killed it at the wrong moment. Anyone who has waited for a 9 kW to come back between takes knows the feeling.
The third is dimming. An HMI dims badly and shifts colour as it goes. An LED goes to zero cleanly, which changes how you balance a set without touching positions.
Then there is variable colour temperature: 2,500 to 10,000 K stated on the Storm XT52, 2,700 to 6,500 K on the Evoke. An HMI sits at 6,000 K and stays there, gel included.
29 kilos, a dedicated line, 15,000 dollars
Weight first. The Storm XT52 head is quoted at 29 kg. This is not a fixture you put on a medium stand hoping it will do. It means calculated rigging, a heavy stand, weights, and two people to get it up.
Then price. The Evoke 5000B is announced at around 14,800 dollars, 15,500 with the case. On a rental that is absorbable, but it puts the fixture in the category where you justify its presence on the list.
And the line, all the same. Five kilowatts on single phase 230 V is roughly 22 amps: a dedicated 32 A run. On 120 volts you are close to 42 amps, which explains the 60 A connectors on the Evoke. Halving the draw does not mean you plug into a wall socket.
Where it replaces nothing
On a location fed by 16 amps, these fixtures do not exist. The replacement question does not even come up: you cannot feed them. A well placed 2.5 kW will do the work, or you bring a generator, and at that point it is a different prep.
The moment you diffuse widely, the equivalence collapses. A big frame, a ceiling bounce, a large softbox: those setups work on output, not on beam centre. The M90 stays ahead, and not by a little.
Same for wide coverage. Open both up to cover a surface and the output gap returns: the centre holds, the edges drop away faster.
Where the LED wins is as a point source over a known distance: punching a window, laying in a hard backlight, holding a background at a fixed throw. The tight beam and the centre measurement describe exactly that use.
Then there is matching. Adding a fixture from another generation and another brand to an existing package is the thing that gets paid for in the grade. One head that does not match the eight others on the set is not a bargain, whatever its output.
What still has to be seen on set
The class is new and the electrical argument is solid: half the current for the same illuminance on a point changes a whole prep.
But the equivalence stops there. These fixtures do not make an M90's output, and it shows the moment a large diffusion goes up. Reading them as universal replacements sets you up for a bad surprise on the day the frame goes in.
What remains to be seen happens on set. Fan noise during takes first, because a fixture you kill through dialogue stops being much use. Then whether the output holds across a full day. And how it matches the rest of the package, which is what really decides whether a head comes back on the next list. No spec sheet gives you any of that.
Sources
Frequently asked questions
Does a 5,000 W LED really replace a 9 kW HMI?
On a point, yes: centre-beam illuminance at the stated distance and angle can be equivalent. In total output, no. These LEDs make noticeably less light than an M90, and at a comparable beam angle the average across the lit area is not the same: the centre matches, the edges fall away. The difference shows immediately behind a large diffusion, which integrates everything that hits it instead of measuring one point.
Why does the equivalence collapse as soon as you diffuse?
Because a frame does not measure beam centre, it integrates everything arriving on its surface. The manufacturer figure is an illuminance at a point, obtained by concentrating the available flux into a tight cone. An efficient optic raises that number without making a single extra lumen. As soon as the setup works on output, large diffusion, ceiling bounce, big softbox, the gap reappears.
What supply does a 5 kW LED fixture need?
On single phase 230 V, 5,000 W is about 22 amps, so a dedicated 32 A line. On 120 volts you approach 42 amps, which is why the Nanlux Evoke 5000B carries 60 A connectors. On a domestic 16 A supply these fixtures are simply not usable.
How much do these fixtures weigh?
The Storm XT52 head is quoted at 29 kg. That weight means calculated rigging or a heavy stand properly weighted, and two people on the build. It also counts in the truck and in the rigging time.
What is the advantage over an HMI, beyond the draw?
Instant switch-on with no strike and no hot restrike, dimming to zero without marked colour shift, and variable colour temperature: 2,500 to 10,000 K stated on the Storm XT52. An HMI stays at its temperature and gets corrected with gel.