Why is "build" spelt with a "u"?

I was just looking at build on Wiktionary and I noticed that in Middle English the word was bilden. Where did the u come from? I can understand why words such as guide have a u; it's to make the g "hard" (/ɡaɪd/) instead of "soft" like a "j": *gide (*/dʒaɪd/).

In Modern English, changing the spelling to bild to follow the earlier spelling wouldn't change the pronunciation (/bɪld/), so how did build gain a u between Middle English and Modern English?

I think part of the premise of this question is incorrect: build did not gain a "u" between Middle English and Modern English. Modern English spelling is standardized, so we can talk about "the" spelling of most specific words. But in Middle English, there are usually multiple attested spellings for the same word; this is partly due to overall lack of standardization (some writers might even spell the same word several different ways on a single page), and partly due to the existence of conflicting standards (Middle English is a cover term that encompasses many different regional varieties over a pretty wide range of time, traditionally from the Norman Conquest of 1066 up until the Great Vowel Shift around the 1500s).

Dictionaries, such as Wiktionary, need to choose one written form to use as the headword (in this case, "bilden"), but this doesn't mean that no other forms existed during this time period. According to the Oxford English Dictionary (OED), forms of this word spelled with "ui" or "uy" actually date back to Middle English, but their origin is unclear:

Etymology: Middle English bulden(ü), bylden, bilden < Old English *byldan to build (recorded only in past participle gebyld), < bold a dwelling. Hence the two fundamental senses are ‘to construct a dwelling’ and ‘to take up one's abode, dwell’. The normal modern spelling of the word would be bild (as it is actually pronounced); the origin of the spelling bui- (buy- in Caxton), and its retention to modern times, are difficult of explanation.

My guess is that it has something to do with phonetic rounding, as the letter "u" is generally associated with rounded vowels. This could be due to the preceding rounded consonant /b/, the historical rounding of the root vowel (Old English y was the rounded counterpart of i) or some combination of both. I'm not at all sure of this, because as the OED says, the regular development of Old English y is complete loss of rounding and merger with i, as in words like kiss v. (from Old English cyssan v.) or lice n. (from Old English lȳs n.).

However, modern standard English vocabulary does also show some influence, in both pronunciation and spelling, from dialects where y developed differently. (I'll describe this in more detail later by quoting the etymologies of specific words.) According to the link that Ricky found (Notes on Etymology by Walter W. Skeat 1901), the digraph "ui/uy" was used in Southern Middle English to represent the sound that developed from older /yː/. Skeat says this is the reason for the spelling of build (which apparently had a long vowel at this point due to homorganic lengthening before the consonant cluster -ld). To support this, he lists the words bruise, from Old English brȳsan, and buy, from a form of Old English bycgan.

I was not able to find any other examples of "ui/uy" being used this way in modern spelling. However, the OED does list muys, muyse as Middle-English spellings of the word mice (from Old English mȳs). For comparison, ice (from Old English īs, without historical rounding) does not have any spellings listed with "ui" or "uy."

Buy seems to be particularly relevant because, like build, it is pronounced with an unrounded vowel in modern English. The OED gives the following information about the etymology:

Old English bycg(e)an, bohte, geboht, corresponding to Old Saxon buggjan, *bohta, giboht, Gothic bugjan, bauhta, bauhts; of unknown origin, not found outside Germanic, and not to be connected, so far as can be seen, with the stem bug- bow n.1 The inflection was imperative byge, bycgað; indicative present bycge, bygest, bygeþ, plural bycgað; subjunctive present bycge, bycgen; whence Middle English s.w. buye, buggeþ; bugge, buyest, buyeþ, buggeþ; bugge, -en; levelled before 1500 to buy- all through, whence the modern spelling. The forms in begge, bey- were Kentish; bigge, bie, by, midland and northern; in the latter the levelling to bie, by, took place as early as 1300.

The forms with the long consonant “cg” seem to have been eliminated through leveling. Another verb that seems to have developed in a similar way is licgan "to lie (down)”: both of these words have the diphthong /aɪ/ in modern pronunciation, which generally develops from Middle English /iː/. But the different spellings suggest that buy may have been pronounced with another vowel at some point, or in some dialects. It's not clear to me from this entry if southwest buy- is the ancestor of the modern pronunciation as well as the spelling (I'm not sure how that development would work in terms of sound changes) or if the modern pronunciation simply comes from midland and northern bie, by.

Google-research suggests that aside from build, buy, and related words, there are no other words where "bui/buy" is pronounced like "bi." (This spelling pattern is discussed in A Survey of English Spelling, by Edward Carney, and Dictionary of the British English Spelling System, by Greg Brooks; both of these sources use the synchronic analysis that "bu" acts as a symbol for /b/.) There are some other words that are superficially similar (buoy, guild, guy, biscuit, conduit, Kuiper) but in general they have different explanations for their spelling, so they are not very useful for explaining build.

However, it might be useful to compare it to busy, for which the OED says:

The original stem vowel ĭ is shown by Old English bisig; the form bysig (when not simply an inverted spelling with y for i in areas where Old English y had been unrounded) probably shows late West Saxon rounding of the stem vowel as a result of the influence of the preceding labial consonant (compare A. Campbell Old Eng. Gram. (1959) §318); Middle English (and modern standard English) busy continues this form (although in the case of the modern standard form with the pronunciation of the unrounded variant). Pronunciation with /ɪ/ is regularly indicated for forms spelt with -u- by orthoepists from the mid 16th cent. onwards (see E. J. Dobson Eng. Pronunc. 1500–1700 (ed. 2, 1968) II. §82).

Forms such as Middle English and early modern English besy (see γ. >forms) reflect Open Syllable Lengthening of short ĭ to long close ē.

Another relevant word is bury, from Old English byrgan. According to the Online Etymology Dictionary:

Under normal circumstances [Old English -y-] transformed into Modern English -i- (as in bridge, kiss, listen, sister), but in bury and a few other words (as in merry, knell) it retained a Kentish change to "e" that took place in the late Old English period. In the West Midlands, meanwhile, the Old English -y- sound persisted, slightly modified over time, giving the standard modern pronunciation of blush, much, church.

It looks like the modern pronunciation of bury comes from dialects like Kentish, while the spelling comes from dialects like those in the West Midlands.

Build, buy, busy, and bury all have a "b" before the vowel: this is the "labial consonant" mentioned by the OED. The idea that it might have influenced the development of the following vowel in words like these is plausible, since there is strong evidence that it caused the following vowel to retain rounding in words like bull. However, it seems that in some dialects, there were words that retained rounding from Old English y even when it was not directly preceded by a labial consonant, such as bruise, blush, and church.

Purely to move things along, here is the text from Ricky's link, best as I can transcribe it (from Notes on English Etymology by Walter William Skeat, 1904):

Build. I have shown that our build is the A. S. byldan derived by vowel-change from bold, a dwelling. I have also considered the A. S. bold as borrowed from Icel. bōl, a dwelling. But I find another account of bold in an article by Sievers on the Noun-suffix -tra, printed in Paul and Braune, Beitrage, v. 529. He says bold is for bolþ-, by metathesis for boþl- == A. S. botl, a dewlling (cf. Bootle in Cumberland and Lancashire). This boþl- or botl is due to a Teutonic bo-þlo-, or bo-þro-, forms in which we recognise the Teut.base BU- [small caps in original], and the Indo-germ. suffix -tro-. This brings us to the root I have already indicated, but accounts for the suffix differently.

To appreciate Sievers' view, his other examples must be examined; we have a sure parallel in the case of needle, of which another form was neeld; for this neeld certainly contains the Aryan suffix -tr-.

Note.— The ui in build is a southern M. E. symbol for the M. E. sound arising from A. S. ȳ, due in this instance to a (temporary) lengthening of A. S. y before ld. Cf. bruise from A. S. -brȳsan (in tō-brȳsan); and buy from late A. S. bȳ, for A. S. byg- in byg-eth, pr. s. of bycgan.

There seems to be a theory that the "u" in "build" was added due to the temporary (!!) lengthening of the Anglo-Saxon vowel in the Southern version of Middle English (presumably, it shortened itself again later on).

Here's the link to Notes on English Etymology written by Walter W. Skeat that might nevertheless shed some light on the issue.