Why Enterprises Are Shifting AI Workloads from the Cloud to Local Devices

Walking into a data centre debate these days feels like stepping into a boxing ring where the bell never quite rings. The punches just keep flying — surging demand for generative and agentic AI, runaway cloud bills, and a constituency of African CIOs quietly asking whether the central cloud model still makes sense for them. According to ITWeb's recent sit-down with Yesh Surjoodeen, HP's managing director for Southern and Central Africa, the answer is increasingly no — at least not for everything.

The pull back to the device

The conversation lands on AI PCs, the latest generation of laptops and desktops equipped with neural processing units (NPUs) designed to handle AI workloads locally. Surjoodeen frames it as a rebalancing rather than a revolution: inference happens on the device, the cloud becomes a fallback rather than the default. Ketan Patel, president of HP's personal systems business, had earlier argued that rising cloud costs tied to agentic AI are nudging companies toward that same local-first posture.

The reasoning is practical, not ideological. Three friction points keep surfacing: the cost and reliability of connectivity to a remote AI data centre, the latency penalty for anything time-sensitive, and — perhaps most acutely for African enterprises — the security and data sovereignty question. As Surjoodeen puts the last point, processing information locally in a secure environment gives organisations more genuine control over their IT infrastructure rather than renting it from someone else's jurisdiction.

The numbers quietly shifting underfoot

The market backdrop makes the case in dollars. Fortune Business Insights projects the global edge AI market will climb to $46.96 billion in 2026, up from $35.60 billion the prior year, and balloon to $445.75 billion by 2034. Inside HP itself, AI PCs already made up 44% of the company's shipment mix in Q2 FY 2026 — a sharp jump from 35% just one quarter earlier. That kind of step-change in mix share is the kind of internal signal investors usually only see when a category stops being experimental.

Surjoodeen is careful not to oversell it. AI PCs are not for everyone, all the time, he acknowledges; they are a journey. But the directional arrow is clear, and as demand for non-AI PCs softens, the economics of supply will tilt further toward the neural-equipped end of the portfolio.

The African affordability tightrope

Where the conversation gets genuinely interesting — and where Surjoodeen's local lens earns its keep — is the affordability question. Entry-level devices remain the realistic procurement option across much of the African market, and a premium AI PC stack risks pricing out the very customers who might benefit most from local inference. HP's framing is that this is not a market to retreat from but a constraint to engineer around: diversifying component sourcing across multiple supply chains so the company can hold price points while still shipping the newer silicon.

It is a quiet form of competitive moat. If HP can land an NPU-equipped machine at entry-level pricing while rivals are still charging a neural tax, the supply curve does the persuading for them — and the data centre buildout debate shifts another inch toward the edge.

What to watch next: whether the next HP quarterly report widens that 44% mix figure further, and whether competitors respond with their own Africa-friendly AI PC tiers. The edge is not arriving all at once, but the shipment sheets suggest it has already started moving in.