Growth Sovereignty and AI

Po­lit­i­cal un­cer­tain­ty, price and per­for­mance are dri­ving clients’ AI mod­el choic­es. On­ly af­ter the White House green­lit the re­launch of one mod­el on Ju­ly 1, 2026, did na­tion-states ap­pre­ci­ate the dan­gers of the de­pen­den­cy syn­drome. Da­ta re­ten­tion rules have al­so ham­pered da­ta re­ten­tion.

The avail­abil­i­ty of cheap, open-weight AI mod­els gives AI users a wider range of al­ter­na­tives to the large fron­tier mod­els of­fered by lead­ing US labs. Spend­ing on the largest and prici­est AI mod­els has plateaued at un­der 15 per cent of over­all out­lay on AI tools. This dis­rupts the old pat­tern of cor­po­rate users de­fault­ing to the most pow­er­ful mod­els for most of their work­loads.

If this trend con­tin­ues, it will rad­i­cal­ly al­ter the busi­ness mod­el of fron­tier labs, which un­til now have fun­nelled the bulk of their multi­bil­lion-dol­lar de­vel­op­ment spend­ing in­to train­ing larg­er mod­els.

The West In­dies faces an awk­ward choice

The West In­dies faces an awk­ward choice. We can de­lay build­ing hy­per­scale AI cam­pus­es while we build na­tion­al da­ta li­braries and da­ta em­bassies to pro­tect our da­ta.

But that could mean for­go­ing eco­nom­ic growth as screen economies flour­ish else­where. Or we can ac­cel­er­ate AI adop­tion and fall prey to the de­pen­den­cy dis­or­der. A ter­tium quid, or third path, is to build our own mod­els and ar­chi­tec­tures.

Ja­maica’s Mae­stro is un­der­go­ing in­ten­sive “red-team” se­cu­ri­ty test­ing. The ven­ture aims to build the Caribbean’s first Sov­er­eign LLM from scratch us­ing re­gion­al da­ta. GPU pow­er comes from a part­ner­ship with Nvidia. The core team con­sists of AI agents and three hu­mans. The Amer­i­cans built and de­ployed fifty-nine no­table AI mod­els in 2025. Chi­na made thir­ty-five. France and the UK each pro­duced one. The Amer­i­cans host three-quar­ters of the world’s AI com­pute. Eu­rope has just five per cent. Rus­sia’s AI land­scape cen­tres on achiev­ing “sov­er­eign AI.”

Rus­sia’s Gi­gaChat us­es a mix­ture-of-ex­perts ar­chi­tec­ture with 20

Rus­sia’s Gi­gaChat us­es a mix­ture-of-ex­perts ar­chi­tec­ture with 20 bil­lion pa­ra­me­ters to op­ti­mise en­er­gy and com­pute ef­fi­cien­cy. Yan­dexG­PT/Al­ice AI is a fam­i­ly of large lan­guage and vi­sion mod­els that pow­ers the Al­ice as­sis­tant. T-Pro and T-Lite are de­signed for en­ter­prise and con­sumer di­a­logue tasks.

Saiga is an open-weight, fine-tuned mod­el. Un­der re­cent leg­is­la­tion, mod­els with more than one bil­lion pa­ra­me­ters are strict­ly clas­si­fied as “sov­er­eign”, re­quir­ing lo­cal host­ing and da­ta con­trol. A Russ­ian start­up, Mostik, has de­vel­oped a new ap­proach to com­bin­ing AI mod­els’ ca­pa­bil­i­ties.

Lega­cy mul­ti-agent AI sys­tems are time- and com­pute-in­ten­sive, with one mod­el writ­ing out text to­kens for an­oth­er to read. Mostik avoids lan­guage en­tire­ly by pass­ing raw, high-di­men­sion­al hid­den states from one neur­al net­work to an­oth­er.

A large, pow­er­ful mod­el con­sid­ers a prompt

A large, pow­er­ful mod­el con­sid­ers a prompt and per­forms the con­cep­tu­al “think­ing,” but it nev­er gen­er­ates a fi­nal text re­sponse. A small, trained bridge mod­ule trans­fers those in­ter­nal math­e­mat­i­cal val­ues di­rect­ly to a small­er, light­weight mod­el. Nei­ther con­nect­ed AI mod­el needs re­train­ing or struc­tur­al changes; both re­main com­plete­ly frozen while the bridge trans­lates be­tween their math­e­mat­i­cal spaces.

Latam-GPT is a project co­or­di­nat­ed by Chile’s Na­tion­al Cen­tre for Ar­ti­fi­cial In­tel­li­gence (CE­NIA). The aim is to de­vel­op an open-ject in­volves a con­sor­tium of 65 in­sti­tu­tions across 15 coun­tries

Glob­al­ly, cor­po­rate users now pre­fer low-cost, open-weight AI mod­els. Old dri­vers are fad­ing. In a few short months, AI will screen im­ports at bor­ders, de­cide on tax re­turns for au­dit, nav­i­gate au­tonomous ships and trains, and clear bank ex­pen­di­tures.

For­eign af­fairs, trade and in­dus­try, and tech­no­log­i­cal in­no­va­tion

For­eign af­fairs, trade and in­dus­try, and tech­no­log­i­cal in­no­va­tion are no longer dis­crete de­part­ments of state. Screen economies will be fu­elled by an in­ter-min­is­te­r­i­al spine: a tri­ad of these gov­ern­ment de­part­ments. It is an un­com­mon mo­ment.

Fun­da­men­tals are be­ing un­earthed. Economies are at vary­ing stages of net­work readi­ness, and the fu­ture skills work­force is at vary­ing stages of pre­pared­ness. Every­thing we thought we knew about the world econ­o­my has been de­railed, in­clud­ing the be­lief in the un­fail­ing su­pe­ri­or­i­ty of open mar­kets, trade lib­er­al­i­sa­tion, and the in­eluctable rise of world­wide free mar­ket cap­i­tal­ism, which took on the in­ex­orable glow of in­evitabil­i­ty and in­vin­ci­bil­i­ty.

We now un­der­stand that eco­nom­ic net­works cre­ate im­bal­ances and pres­sure points be­cause states have vary­ing re­et over­con­cen­tra­tion in crit­i­cal tech­nol­o­gy net­works can cre­ate choke points

In a speech ti­tled A new age

In a speech ti­tled “A new age of cap­i­tal: growth, sov­er­eign­ty and AI”, de­liv­ered in Vi­en­na at the Hof­burg im Di­a­logue event on Sep­tem­ber 14th, 2026, the Eu­ro­pean Cen­tral Bank Pres­i­dent, Chris­tine La­garde, urged Eu­rope to de­sign and de­vel­op its own AI if it hopes to have any foothold what­so­ev­er in the in­fos­phere rev­o­lu­tion and in the glob­al Screen Econ­o­my.

Pres­i­dent La­garde went on to out­line three ba­sic steps to re­main com­pet­i­tive and close the gap. Step one is to aug­ment Eu­ro­pean com­put­ing ca­pac­i­ty. The sec­ond step is to build AI mod­els that run on Eu­ro­pean AI ar­chi­tec­ture. The third step is to se­cure ac­cess to fron­tier mod­els to re­duce the risk of be­ing cut off.

Dr Fazal Ali com­plet­ed his Mas­ter’s in Phi­los­o­phy at the Uni­ver­si­ty of the West In­dies. He was a Com­mon­wealth

Schol­ar who at­tend­ed the Uni­ver­si­ty of Cam­bridge, Hugh­es Hall; the provost of the Uni­ver­si­ty of Trinidad and To­ba­go; the act­ing pres­i­dent of UTT; and the chair­man of the Teach­ing Ser­vice Com­mis­sion. He is the pres­i­dent of NI­HERST and an ex­ter­nal ser­vices con­sul­tant with the IDB. He can be reached at faza­lal­itsc@gmail.com