Five volts on VCC

VCC wants 5 V

The data sheet gives the LED a supply of 3.5 to 5.5 V, and says it works fully from 4.5 V. A 3.3 V board's 3V3 pin is under both. So VCC goes to the 5V pin: on an ESP32, an ESP32-S3 or an Uno it is the USB cable's 5 V, and on a Pico it is the pin called VBUS.

On a 3V3 pin the bar may still light, dimmer. That is the trap: it looks like it works, and nothing about it is promised.

The data is read against VCC

The LED decides whether DATA is HIGH by comparing it with its own supply. A HIGH must be at least 0.7 × VCC; a LOW at most 0.3 × VCC. Between the two, the data sheet promises nothing.

On an ESP32-S3 with a USB cable in, the 5V pin is about 4.6 V: 5 V through a Schottky diode. 0.7 of 4.6 V is 3.22 V, and a 3.3 V pin's HIGH clears it by less than a tenth of a volt. That is the reason the plain wiring usually works.

Feed VCC a full 5 V instead, from an Uno or a bench supply, and the LED wants 3.5 V. A 3.3 V pin is then 0.2 V short on paper. Many WS2812s read it anyway, which is why the direct wire is everywhere, but a different batch, a longer data wire or a warmer room is enough to lose it.

Only LED1 has to read your pin

The margin matters once. LED1 reads your board's signal; every LED after it reads a fresh copy from the LED above, at the bar's own supply. So a level problem shows as a wrong or flickering first LED with the others behaving.

When to add a shifter

On an ESP32 or ESP32-S3 with VCC from its own 5V pin: usually not. With VCC on a separate 5 V supply, or when the first LED misbehaves, put a fast level shifter in the data line. The logic level converter book walks through one channel of a TXB0108. A MOSFET converter such as the TK97 raises its output through a resistor, and for pulses a third of a microsecond long that is marginal.