Here are collected tidbits that don't fit in tabular form.

Q: Why were there so many companies based on MIT lisp machines:
A: My understanding is that there was an irreconcilable dispute between Greenblatt and essentially the rest of the original Lispm crowd about the Right Way To Go.  Most of the rest of the principals signed up as founders of Symbolics, which had a high finance business plan (20 years before the dot com era) to collect and spend vast quantities of venture capital, while building up a big company as rapidly as possible; and eventually to go public.  Greenblatt was inalterably opposed and founded LMI to take the low finance road,; planning to bootstrap the business using its own revenues.   It all happened pretty much as planned for Symbolics, which spent investor's money like water, and sold machines as fast as they could make them.  LMI struggled along and also sold machines as fast as they could make them, but due to limited resources that wasn't very fast.  Eventually LMI sold rights to their technology to Texas Instruments, TI which produced their own lispms based on the same design ideas and software.  TI was a genuine big fish - a multi billion dollar chip maker, so it remains a puzzle why they had so little impact on the market.

Q: What about Xerox, why did they make lisp machine? (and didn't they make good copiers too?)
A: The legendary Xerox Palo Alto Research Center, PARC was the nexus for Interlisp, the "brand X" of the lisp community.  Given the amazing, ground breaking technology being developed elsewhere at PARC (the alto, the mouse driven interface, smalltalk, etc) it's not surprising that there was a Xerox lisp machine too.  And given Xerox legendary failure to make a successful business out of any of their other ground breaking computer technology, the eventual fate of Xerox lisp machines should be no surprise.

Q: What about home and business personal computers?
A: Lisp machines had big bit mapped screens with multiple windows, mouse driven GUIs,  megabytes of  main memory, hundreds of megabytes of disk,  virtual memory, and networked by ether net..  They were also very expensive. At the same that time home personal computers were using kludged TV adapters for display, cassette tapes for mass storage, had 16K main memory, and networked by "sneakernet", but the price was right.  These machines were initially  much too puny to be of interest to businesses, but Moore's law was at work, and the amount of hardware you could buy for $1000 doubled every year.  Pretty soon little home machines were getting pretty big, and businesses were using them for serious tasks.  Today, every computer you are likely to see is descended from these puny home machines, not from the mighty Lisp machine workstations.

Q: Why did all these lisp machine companies fail?
A: You'll get a lot of opinions on this one, but here is my interpretation. In retrospect, Lisp machines were part of a transition from the giant mainframe hardware of the 60's and 70's to the workstation hardware of the 90's.  Lisp machine hardware was always very expensive, and was sold to businesses for use by well paid programmers (or sold to universities for use by really smart slave labor - same thing). Economically, it makes sense to buy the best available tools and put them where they are most useful.   While Lisp machines were trying to supply the same features cheaper,  personal computers provided a constantly expanding set of features for a constant price.  Eventually the two lines had to cross, (with lisp machines getting cheaper and pc's getting more capable), and at that point the game was over.  Lisp machines were not alone in the "top down" game, and IBM PC's were not alone in the "bottom up" game; but the "bottom up" crowd clearly won.  It's interesting to speculate what might have happened if Moore's law had broken down about 1990.

Q: Ok, so lisp hardware was overtaken by Moore's law.  What about Lisp software?
A: A lot of features of the Lisp machine environment have become completely standard (windows, networks. virtual memory) and a lot of features of Lisp as a programming language have become standard features of modern programming languages (objects, garbage collection). - but Lisp itself has reverted to a niche used for research.  Why?  Lisp never developed a good way to deliver small, modular programs.  Lisp supports and encourages building monolithic, tightly coupled environments, containing lots of closely coupled modules.  The standard unit of deliverable lisp program is called a  world, and that is not an accident!  When PCs won the hardware wars, they saw no need to adopt the monolithic environment or spaghetti coding style of Lisp machines, but they have been busy reinventing it.

Q: How many Lisp machines were made (and sold)?
Not very many, probably less than 5,000 units combined, including all generations of machines made by all manufacturers.
 

MIT CONS	1	prototype
MIT CADR	25	MIT internal production run
Symbolics LM-2	100	$70K each.  Originally only a few were to be made, but Symbolics couldn't produce the first 3600 soon enough and had to keep making them.
Symbolics 3600 - class	1000?	$50-$100K depending on date and configuration.
Symbolics 3650 - class	1000?
Symbolics XL - class	2000?
LMI Cadr
LMI Lambda	~200
Xerox Dolphin
Xerox Dandelion
Xerox Dorado

Q: Who used them?

• A lot of the total production run, especially in the early years, were used internally at the various lisp machine companies.  Some were used by their programmers and hardware designers, others by sales and manufacturing in support of more sales and more manufacturing.  Of course, these machines were part of expenses, not sources of income.
• Another big chunk went to ARPA sponsored projects, sometimes they stayed in publicly visible places such as ISI and SRI; sometimes they were "laundered" through public affiliates but used elsewhere.  I remember one batch of 50 or so 3600's that were delivered to ISI but were immediately re-shipped to ARPA-specified destinations.  Government research projects were probably the biggest single paying customer.
• The Symbolics Graphics division sold quite a few machines to run "S-products", the suite for generating animation using lisp machines. The peak year in my mind was 1986, when about 1/3 of the Siggraph film show was produced on Symbolics machines.  These machines were sold in small groups to many different customers.  As far as I know no one ever built a giant "render farm" of Symbolics machines except Symbolics.