Incorrect Semantics

Semantic elements and attributes provide a higher level of communication. While a figcaption can be used as a valid replacement for alt to supply a context dependant caption of an image constrained to being inside of a figure, it does not provide an explicit, and strong semantic for an image long description. Terse and verbose descriptors differ. As Gregory J. Rosmaita aptly explained,

alt text is the brief "at a glance" or "cognative thumbnail"... alt text needs to be terse for a number of reasons, including usability, extremely limited viewports, small amount of video "real estate" (iPad and smaller devices) etc.

The prevailing rationale for keeping figure/figcaption change proposal stated, that figcaption was not a long description but rather a short description:

The figcaption element exposes a short description of the figcaption content, which may be read out to help a user determine if they wish to look at the content of figcaption now or later.

Lacks Support

With the one exception of Firefox on Windows figcaption is not accessibility supported semantics wise. HTML5 accessibility states that it,

provides the same amount of semantic information to AT as a div element.

Forces a Visual Encumbrance

The figcaption technique would force a visual encumbrance on sighted users unless additional code (in the form of CSS or hidden) were added in order to hide the visual indicator. Please refer to Cascading Style Sheets and hidden for details of why using these techniques is a bad idea.

Lacks Ability to Provide Description in a Separate Document

The figcaption technique lacks the key functional requirement of providing a long description in a separate document.

Reinvents the Wheel

Attempting to use figcaption to provide a long description reinvents the wheel by duplicating a basic method that has already previously been created.

No use cases of figcaption for a long description has been presented that are not already satisfied by longdesc.

Lacks Educational Material

No tutorials, books, or documentation exists that explain to authors how to make an accessible long text alternative with figcaption.

No Examples in the Wild

No examples in the wild of accessible long text alternatives with figcaption have been presented for any of the use cases. No evidence exists that authors have or will use it for accessible long text alternatives.

Lacks Backwards Compatibility

The figcaption technique lacks all backwards compatibility. It is not:

For further rationale please consult the backwards compatibility document.

Retrofitting Problem

Any proposed solution should be easy for authors who are already publishing content with longdesc to retrofit their existing pages. figcaption is new and possesses a radically different authoring pattern than longdesc. Because of this as well as its on-page limitations, retrofitting with figcaption would be difficult, labor intensive, and error prone.

No retrofitting is required for longdesc as it is an existing HTML feature not a new one.

No Evidence of Improvement

No evidence has been presented that figcaption produces more accessible content for long descriptions than longdesc or that more authors would use it correctly.