Kay's point at the time is that this was what you could do if you were willing to spend a large amount of money per user on hardware. It wasn't cost-effective yet. Someday it would be. Xerox was willing to spend the money, so when the hardware became available, they'd own the market.
This wasn't the first GUI. They'd downsized the 1968 Mother of All Demos to a machine that sat alongside a desk. When Engelbart did that, it required a room-sized mainframe, a video link from the mainframe to the demo site (a very big deal in 1968), and a sizable crew of support people to keep it all running. All to support one user.
This technology reached production with the introduction of "workstations". The first was the Terak, in 1977, which ran an interpretive environment called the UCSD P-System. This was a byte code interpreter for Pascal. Fast compile, slow execution, basic graphics. Perhaps the ancestor to Turbo Pascal for DOS. PDP-11 technology underneath. I used one briefly. 1979 brought the Three Rivers PERQ, which was a lot like an Alto and sold as a commercial product. It had another 16-bit CPU inside, from ICL in the UK, and was another interpretive Pascal system. I never saw one; it was considered something of a dud.
When the Motorola 68000 came out, it was clear to a lot of people that there was finally a big enough microprocessor to build a workstation. At last, a 32-bit address space. (Or at least 24; early machines didn't use the high byte.) The Apollo Domain, in 1980, was the first of those. That was the first real workstation. It had its own OS, its own networking (a coax token ring), its own file system (exclusive use locking across the network worked, unlike UNIX), and its own GUI. Apollo had their own MMU and their own paging system, which was really hard due to the M68000 not handling page faults properly. It was way ahead of anything else at the time. But the small organization behind it just had too much work to do building all that software and hardware. When I used one, it was clear this was the future, but it wasn't there yet.
Then came Sun, and a whole slew of UNIX workstations. UNIX was really the wrong tool for the job, but it was available and cheap. (Not yet free.)
Meanwhile, Apple was trying to get a low-end workstation working. The Apple Lisa (1983) was the result. It was sort of a cost-reduced Apollo - its own OS, its own MMU (a whole board, which ran the price up), and its own GUI. Also its own disk drive, the one time Apple tried to build a hard drive in house. That didn't go well. The Lisa was impressive and useful, but it cost too much to go mass market. The real UNIX workstations came with big screens, and the Lisa had a dinky screen like DOS-type computers. Meanwhile, IBM was eating Apple's lunch, replacing the obsolete Apple II with dumb DOS machines. No GUI, but enough compute power to run a spreadsheet and, importantly, a hard drive. Now people could do basic business work.
Apple's response, the Macintosh, was the world's greatest toy computer. It shipped in 1984 with 128K of RAM, one floppy drive (you needed two to get anything done), an operating system with no CPU dispatcher, no memory protection, and a nice GUI. But no hard drive. It was really slow and spent most of its time reading floppies and displaying an hourglass "wait" icon. It almost killed Apple. Sales were very low; Apple had planned to sell about 47,000 units a month, but only sold about 5,000. What saved Apple was the first desktop laser printer, the LaserWriter, in 1985. (PARC had a laser printer in the mid-1970s, but it was based on a big copier engine and bigger than a desk.) In 1986, Apple introduced the Macintosh Plus, which could, at last, support an external hard drive. That was the first successful product in the line, and launched the desktop publishing industry.
In parallel to all this, there was another line of technology, now forgotten - "word processing". Wang and IBM were the big players in this. This started with typewriters with some memory, around 1971, and by 1977, Wang had the Wang Office Information System. This involved many semi-dumb terminals connected to a shared unit with a CPU and file server. Those could in turn be networked, and documents sent around the system. Very cost-effective, because each terminal was cheap. This was a huge win for offices which did a lot of document preparation, and Wang was, for a while, a very successful company.
So Xerox tried to move into that space with the Xerox Star, in 1981. This brought the Alto technology into an office environment. Worked fine, cost too much. Like the Wang system, it was very closed. This was deliberate. Word processing and office system were used by secretaries and clerks. They had to Just Work. Exposing end users to the internal complexity of the system was considered a bad idea.
The cultural change which brought system administration to the masses wasn't seen coming. It was inconceivable in the early 1980s that clerical people and small business operators would have to worry about what was going on inside the box. But as the DOS-type machines got more powerful and moved into offices, for about two decades everyone had to become a sysadmin. Apple tried to hide more under the covers, but it didn't really work all that well in the early years.
Today, of course, the complexity has mostly been put back in a sealed box, and you can give a Chromebook type tablet to a 5 year old and they'll be able to work it.
(The price of RAM was a big problem in those days. All the early DOS machines and Macs were RAM-starved. The workstation people plowed through that problem with money, and RAM was a big fraction of machine cost. Now it takes a gigabyte to run Hello World, but, whatever.)
Smalltalk started working in 1972, the Alto in 1973, and Smalltalk on it a month or so after the Alto came alive. The GUI came more from other parts of ARPA than NLS, but there were a few NLS elements. Cheers.