The compute is coming home. The jobs that run it have not arrived yet.

The national AI supercomputer G42 and Cerebras are building in India with MBZUAI and C-DAC. American chips. Emirati capital. Hosted in India, governed by Indian law, data kept inside national jurisdiction.

Eight exaflops is the headline, but the sentence that matters sits underneath it. The national AI supercomputer that G42 and Cerebras are building with the Mohamed Bin Zayed University of Artificial Intelligence and India's Centre for Development of Advanced Computing will be hosted in India, under India-defined governance, with data kept inside national jurisdiction. The chips are American. The capital is Emirati. The control - the part that decides whose law the data answers to - is Indian. That is the new shape of the deal, and it is not the shape the back-office decade was built on.

For twenty years the Global South rented its compute. The work happened here, but the servers, the cloud regions and the governing law sat in Virginia, Dublin or Singapore. The G42-Cerebras system inverts that. Cerebras, fresh from a one billion dollar raise at a 23 billion dollar valuation and a ten billion dollar OpenAI deal, is not selling India a service - it is helping India own an asset. The system underpins the India AI Mission and follows the fifth India-UAE Strategic Dialogue of December 2025. This is statecraft wearing a data-centre badge.

The mechanism is a corridor, not a single deal. Saudi Arabia's Public Investment Fund has earmarked around 40 billion dollars for AI, semiconductors and data infrastructure. Stargate UAE is planned as a five-gigawatt Abu Dhabi campus. India is adding 20,000 GPUs to a national stack that already held 38,000, with Vaishnaw chasing more than 200 billion dollars in two years. The capital and the silicon move between the Gulf and India while the governance stays local. Africa is the third node - Cassava's AI factories and the new AISCA Foundation in Kigali make the same sovereignty argument with a fraction of the power.

The implication for the workforce is the whole point, and almost nobody is saying it out loud. Sovereign compute does not run itself. Someone has to operate it, secure it, audit it and answer for what it produces to a regulator who now sits in the same country as the machine. Those are jobs, and for the first time in the AI era they have a structural reason to sit in Bengaluru, Riyadh or Lagos rather than the US West Coast. The chart shows where the compute is being built. The job market has not yet noticed that the operators have to be built too.

From renting foreign compute. -> To hosting it at home.

The old bargain was rent. You sent your data to a foreign cloud, you paid by the hour, and the question of whose law governed that data was someone else's problem until it suddenly was not. Data-localisation rules, sector regulators and a hard geopolitical lesson about who controls the off switch turned that arrangement from convenient to risky. The shift is from renting foreign compute to hosting sovereign compute - and it has moved faster in four months than the layoff story moved in a year.

Watch what has happened since February. Krutrim, India's first GenAI unicorn, stopped trying to build a frontier model and a chip, pivoted to a domestically built AI cloud, and turned its first profit - around 300 crore rupees in revenue, three times the prior year, at a margin above 10 percent. More than 25 enterprise customers in telecom, finance and healthcare now run on it, and most of its GPU capacity is already committed to external workloads. The demand it is meeting has a single common requirement: keep the data in India.

The same pivot is visible across the corridor. Cassava launched Africa's first NVIDIA-powered AI factory in South Africa and is extending GPU-as-a-Service to Nigeria, Kenya, Egypt and Morocco, with the explicit promise that data never leaves the continent. Morocco landed a 1.2 billion dollar Casablanca facility scaling to 500MW. Nigeria's Lekki corridor brought a 100MW hyperscale site online. None of these is large by hyperscaler standards. All of them are sovereign by design.

The worker who once sold availability on someone else's cloud now has a different thing to sell - the ability to run, govern and stand behind compute that lives at home. That is a higher-value seat, and it is being created in exactly the geographies the back-office decade trained. The discount that built the hub is gone. What replaces it is not a cheaper seat. It is a seat with a regulator's phone number attached.

If the corridor is real, the scarce seat is not the model-builder. It is the human who runs sovereign compute and answers for it.

If the corridor is real, the scarce role is not the model-builder. The world has enough of those, and Gartner thinks more than 40 percent of agentic AI projects will be cancelled by the end of 2027 anyway. The scarce role is the one that keeps sovereign compute running and lawful: the operator who manages the cluster, the governance owner who answers for its outputs, and the residency-and-compliance lead who can prove to a regulator that the data never left the jurisdiction. Build toward that seat, not toward the crowded one.

The business that emerges from the same fact is sovereignty-as-a-service. Krutrim has already proved the unit economics - a domestically built cloud, profitable, with external demand it cannot fully meet. The African delegations who came to the India summit asking how to build their own sovereign models are the addressable market. Packaging Gulf capital, Indian engineering and local data residency into a single offer is a business that did not exist two years ago and has a runway measured in the rest of the decade.

For the board, the question is no longer whether to adopt AI. It is whether you own the compute your most sensitive workloads run on, or rent it from a jurisdiction you do not control. That is a capital-allocation decision and an accountability decision in the same breath, and it belongs on the agenda before the next regulator asks where your data sleeps at night.

6 rising role categories, each with a sourced hiring signal.

The layoff headlines and the buildout headlines are the same story read from opposite ends. Profitable firms are cutting experienced staff while pouring capital into compute, and the corridor is where a chunk of that capital lands. The roles rising fastest are the ones that sit between the silicon and the regulator - people who can run a sovereign cluster, govern its outputs and document its compliance. These are not entry-level seats, and that is the catch the skilling programmes have not solved.

The named demand is concrete. India needs operators for the G42-Cerebras system and for Krutrim-style clouds. The Gulf needs them for Stargate UAE and the Saudi 1.5GW build, against a hiring market where UAE AI demand has already jumped from 32 to 48 percent. Africa needs them for Cassava's factories, and AISCA in Kigali has put a number on the ambition - one million young people into AI-value-chain work, with the data staying on the continent.

The honest watchout is the one Gartner and nasscom both point at. Agent hype is running ahead of delivered value, and system-level requirements - reliability, integration, security, accountability - are getting harder, not easier, as agents scale. The job that survives the cancellation wave is the one that owns those system requirements. Chase the operator and governance seats, not the prompt-engineering seat the market is about to discount.