By the middle of 1986 it was clear that no existing operating system was capable of hosting the toolkit, at least not on a personal computer level. Instead of making and selling a toolkit, the business plan changed to making and selling complete machines running it on top of a Unix-like Mach-based OS. The later would be created by a team led by Avie Tevanian, who had since joined the company. The name of the company was changed to NeXT Computer Inc..

By 1987 NeXT finished construction of a completely automated factory for their first product, the NeXTcube. Stories about Jobs demands for the factory and the cube are now legend, including re-painting the factory several times in order to get just the right shade of grey, and instituting a series of time consuming changes to the line so that the cube's expensive magnesium case would have perfect right-angle edges.

Another example of what appears to be hubris can be seen in the selection of a drive mechanism. At the time most machines shipped with hard drives of 20 or 40Mb, onto which software (including the OS) was loaded using floppy disks. Even in the late 1980's this was starting to be a real problem, as the user needed to swap huge stacks of floppies to load the ever-growing applications onto the hard drive. This was even more of a problem for the OS itself, which often spanned a dozen disks or more.

The solution to this problem today is to use a CD-ROM (or even DVD), or to download it over Ethernet. Even at this early point in time, the CD-ROM would have been an acceptable solution for most. But NeXT would try to do one better, replacing both the hard drive and floppy with a single removable medium. This was in the form of a 256MB Magneto-optical device, which was just coming on the market.

The cubewas based on a 25MHz Motorola 68030 which had recently come to market, making it compete with the workstation vendors like Sun in terms of performance. There had been some discussion of using the Motorola 88000 RISC chip, but it was considered to risky as they weren't available in quantity at the time. The 68030 was supported by the 68882 FPU for faster math, the 56661 DSP for multi-media work, and a unique DMA channel-controller which allowed much of the I/O to be offloaded from the CPU and boost the speed of common tasks considerably. In head to head comparisons the cube consistantly trounced other Unix workstations of the same class, both in terms of performance, and by costing roughly 1/2 as much.

Comparing the cube with more common Intel based machines was actually more difficult. The machine shipped with a huge 8M of RAM, the 256Mb MO drive, Ethernet, and a large greyscale "megapixel" (1120 x 832 pixel) grescale display. Meanwhile the typical PC still used the 8086 CPU, had either a 320 x 240 4-color or 640x480 black and white display, no networking, and may or may not have a hard drive.

Prototype cubes were shown to standing ovations in late 1988, and a slew of magazines reviewed the system - all concentrating on the hardware. By 1989 the machines were in beta form, and they started selling limited numbers to universities with a 0.9 version of the OS installed.

The machines weren't ready for "real" sales until 1990, when they went on the market for $9999. At the time Jobs was concerned that the market was quickly stratifying and the window of opportunity to introduce any new platform was rapidly closing. Just after their release he noted that "this will either be the last machine to make it, or the first to fail".

When it was discovered that the MO drive led to very serious performance problems in real-world use (as well as costing about $100 per disk), NeXT as a whole gained a reputation for failure that would never rub off.

This problem was rectified by 1991, when a new series of machines with floppy disks and hard drives shipped. A new line then introduced the newer and much faster 68040. The same parts were later put in a new "pizza box" case, creating the NeXTstation, which sold at a lower price point and became fairly popular.

With all of the attention focused on the hardware, the true gem of the system, NeXTSTEP, was lost in the hype. Nevertheless you would often see articles in major programming magazines like Dr. Dobbs from NeXT staff, showing how a recent articles 3+ pages of code was implemented under NeXTSTEP in perhaps 10 lines.

A number of programs started shipping for the system, including the amazing Lotus Improv spreadsheet, and the first web browser. The system also shipped with a number of "smaller" applications built in that would actually improve the enviornment considerably without being obvious, things like the Miriam-Webster Collegiate Dictionary, Oxford Quotations, the complete works of William Shakespeare, and Digital Librarian search engine to access them all.

In all, some 50,000 NeXT machines were sold. This was a tiny segment of the market, and proved Steve's own words prophetic. It's worth pointing out that there hasn't been a single successful desktop platform since. Nevertheless it's an open question as to whether the systems would have been more successful had they avoided the performance and price problems by including a hard drive in the first machines.

NeXT Software

By 1992 work had already started on a port of the NeXTSTEP operating system to the Intel platform. At the same time work began on replacing the 68000 series CPU's with the new PowerPC which was starting up as a joint program between Apple, IBM and Motorola.

By late 1993 the Intel port was complete, and was released in the form of NeXTSTEP 3.1 (also referred to as NEXTSTEP 486). Work on the PowerPC machines stopped along with all hardware production, and the company renamed once again, this time to NeXT Software Inc.

NeXTSTEP 3.x was later ported to PA-RISK and SPARC based platforms, for a total of four versions:

NEXTSTEP/NeXT (for NeXT's 68k "black boxes")
NEXTSTEP/Intel
NEXTSTEP/PA-RISC
NEXTSTEP/SPARC

None of the non-NeXT versions appear to have seen much use. At the time the performance of the Intel platforms was quite limited (although not for long) and running it on the other two systems meant replacing their "native" OS's outright.

At this point NeXT's attention turned away from supplying a complete OS. NeXT and Sun started an effort that would lead to OpenStep. This was basically NeXTSTEP without the Mach-unix below it, the OO toolkit running on top of some other OS.

The company had now come full circle. Originally intending to sell a toolkit running on top of other OS's, they had ventured into hardware, failed, and returned to selling a toolkit running on top of other OS's. But no matter how successful the toolkit was, NeXT never managed to avoid the smell of failure - they were always the company who exited the computer market, never the company that conquered a programming market.

New products continued to ship on the platform, but Steve had clearly lost interest and was spending all of his time at Pixar.

End of NeXT

In 1996 Apple purchased the assets of NeXT Software in order to use OpenStep to replace the now outdated MacOS, but it would be a long four years before it would be released as Mac OS X.