Behind 158 units sold: Are customers really willing to pay?

I spent two days testing Faraday Future's EAI robot deployment route. I work on AI solutions at Huawei Cloud, and during client demos, my biggest fear is a robot that walks on stage but has no users off-stage. Scenarios like exhibition halls, hotels, and park tours—where people watch, take photos, and pay for buzz—are worth trying with one unit. For stable delivery, clear ROI, and after-sales SLAs, I'd suggest waiting another round. EAI robots can be roughly understood as embodied intelligence robots capable of perceiving environments, moving, and performing simple actions. ROI is Return on Investment; SLA is Service Level Agreement.

I didn't purchase directly; I borrowed an Aegis quadruped prototype from a friend and set it up in a conference room to simulate a hotel lobby. On the first morning, we built the map. Mapping means letting the robot remember room boundaries, carpet edges, and luggage rack positions. The interface was simple: map on the left, task cards on the right, status light below. When the status light turned green, I thought we were good. Then I asked it to pause briefly at a simulated threshold, and the task point drifted half a meter. Pre-sales fears this kind of gap where demos work but clients hesitate to use it themselves.

On day two, we connected the voice entry point. I used Porcupine for local wake-word detection for about four weeks, and sensors were connected for roughly a month using a similar device status collection scheme. For Q&A, I hooked up the GPT-6 Astra API. I've only been using it for three days, so I can't claim depth, but intent recognition is indeed convenient. When clients ask how to get to the nearby gym or if luggage can be delivered, it converts these into tasks. Can this capability be packaged as a product? Will users pay? It depends on whether hotel owners are willing to pay to reduce repetitive questions weekly.

What surprised me was the movement. I've worked with joint actuators for two weeks, so I have basic judgment. The Aegis starts, stops, and makes small turns on carpets more steadily than I expected. It doesn't show off, jump, or flip, but it can walk slowly near people. For high-end hotels, this is more useful than dancing. No one wants the front desk scared by an overly lively machine.

Bottlenecks are obvious too. Documentation lags, and configuration feels like a puzzle. Map files, task templates, voice wake words, and callback addresses are scattered everywhere. I built a demo backend on Huawei Cloud, taking about a month, and ran Ascend side for four weeks. Still, compatibility issues arose when converting inference models due to unclear documentation. The messier the toolchain, the more pre-sales teams have to clean up for clients. Clients won't pay just for "full-stack self-developed"; they ask how long it takes for repairs when things break.

Commercially, these numbers are worth breaking down. Public statements say August EAI robot body sales and shipments hit 158 units, a new monthly high, with cumulative sales exceeding 552 units. Looking back, July was 152 units, and as of August 4th, cumulative sales were 394 units. The annual target is over 2,000 units. This pace shows volume is increasing, but 552 units is still early, resembling small-batch validation in high-end scenarios. Previously, they delivered 2 Master Ultra and 4 Aegis units to Golden Hills. Early buyers are mainly customers willing to pay a premium for experience; factory production lines haven't arrived yet.

So, I don't recommend ordinary SMEs buy this as a productivity tool now. Hotels, exhibition halls, and brand flagship stores can pilot it because the scenarios naturally suit display plus light service.

For factory inspection, warehouse handling, and home companionship, I suggest waiting. Involving safety, privacy, and continuous operations, clients care more about running continuously for three months without incidents. Monthly sales figures are just surface level.

I'm fairly confident about the trend. Over the next year, sales of such robots will rise, but growth will likely shift toward leasing, subscriptions, and scenario packages. One-time purchases won't become mainstream. Whoever first turns mapping, voice, tasks, callbacks, and after-sales into a pipeline that doesn't require daily pre-sales babysitting will achieve repeat purchases. Currently, it's better suited for scenario pilots, not procurement as a stable productivity tool. Record sales don't equal willingness to pay long-term.