I made this call once already. In 2007 I said messaging was infrastructure, not a feature. I took 95 percent of Mexico and sold it to Twilio. This is the same call, one layer up.
I did not raise money to do it. I do not plan to need permission this time either.
95 percent of Mexico. 70 percent of Latin America. Google's number one vendor across 21 countries. Eighteen people, no outside capital, 70 million USD ARR at exit.
Enterprise software delivered from Mexico to US agencies, FEMA included, after the hurricanes. One of the first in the country to run the augmentation model that later became standard.
Pre-seed and seed across Latin America. I write checks to technical founders with global ambition. I do not sit on boards I cannot help.
Banks that matter will not run AI on infrastructure they cannot audit. Neither will governments.
So somebody builds the layer they run it on instead. Somebody owns where every regulated AI workload executes, which model answers, and who can prove it afterward.
It will be Saptiva AI.
The control layer under regulated enterprise. Any model. Any environment. Cloud, on-premise, air-gapped.
Banks, government, and universities run production workloads on it.
saptiva.com →I publish what I think happens next, with a date on it. When I am right it stays up. When I am wrong it stays up too.
Messaging becomes the infrastructure layer of Latin America. Not a feature.
95 percent of Mexico, 70 percent of the region, Google's top vendor across 21 countries. Acquired by Twilio in 2020.
Regulated institutions will refuse to run AI under foreign jurisdiction.
KAL live with the Mexican government. IBERO live on-premise. Banking workloads in production under CNBV supervision.
Mexico regulates AI before it regulates cloud. Data locality becomes the binding constraint by 2028.
Ask me in 2028.
By 2030 every Latin American country above 20 million people operates a sovereign model.
Most of them will not build it themselves.
Four calls on the record · One settled · Nothing removed
Spanish first, English second. Every piece at its own address, because an argument nobody can link to is not an argument.
Jurisdiction follows the operator, not the datacenter. What that means for any institution that answers to a supervisor, and why almost none of them have checked.
Su historia en video: de vender sitios web en León a construir la capa de IA de la región.
Milenio TV · May 2026Entrevista en Multiverso: regalar los datos a corporaciones extranjeras es el nuevo despojo digital.
El Financiero · Nov 2025Saptiva AI construyó el LLM nacional. Marcelo Ebrard lo anunció.
Mexico Business News · Nov 2025Quoted at KAL's unveiling: AI under foreign jurisdiction is data leaving the country.
Prensa IBERO · Abr 2026Infraestructura on-premise construida por Saptiva AI. IA bajo control institucional.
Reforma · 2024Cómo Quiubas se volvió el agregador más grande de la región.
La Frontera · 2025On building decentralized AI infrastructure for the region, and who funds it.
Líderes Mexicanos · 2024El mismo instinto que ahora define la apuesta en IA.
Techla Media · Jun 2025Su apuesta recibe el distintivo «Hecho en México» de la Secretaría de Economía.
America Digital · 2025Qué significa operar un modelo nacional dentro del país.
Ventura Labs · Apr 2025An hour in English on his thesis: data, models, and who owns the stack.
Ángeles de LatAm · Nov 2024Veinte años de lecciones: Quiubas, la venta a Twilio, y cómo invierte ahora.
Startupeable · Ene 2024Quiubas por dentro: 90% de los SMS de México con 20 personas, y la oferta de nueve cifras que rechazó.
True Growth · Jul 2023Con Jesús: la venta de Quiubas, contada por los dos que la construyeron.
Funds and operators looking at the control layer for regulated enterprise. Come with a thesis.
[email protected] →AI infrastructure, data locality, and building global companies out of Mexico. Spanish or English.
[email protected] →