It seems that in PIE the verb 'to be' had two different stems in the present, and e-grade as well as a zero grade. The conjugation proceeds as follows:
The ablaut grade was posited mainly to account for the reflexes in Latin and Sanskrit, which show an initial sibilant in a few forms: Lat. sum, sumus, sunt; Skt. smas 'we are', stha 'you (pl) are', santi 'they are'. According to Palmer (The Latin Language. 1954.), the development of the Latin forms are as follows:
The second and third person singular forms seem regular; *h1esi > es, *h1esti > est. By analogy with the singular forms a full grade was introduced in the second person plural (*h1ste > *ste > *este > estis). *s-enti > *s-onti > sunt, for the third person plural, and this apparently had some analogical bearing on the first person singular *somos > sumus. This new first person singular form in turn created a new first person singular, from an original *esmi (*esmi > sum).
Conclusion: the paradigm is an analogical mess in Latin, one which must be sorted out if my paper on the first laryngeal in the PIE verb 'to be' is going to be resuscitated.