What this driver is
AI compute demand is the end-market pull on AI chips and the infrastructure that runs them. It is distinct from capital spending. Capital spending measures what companies are committing to build. Compute demand measures what is being used and ordered right now.
When this driver is active, the engine is detecting that the demand side of the AI compute equation is strengthening. Semiconductor companies are reporting strong utilisation, chip pricing is holding or rising, and the forward order book is full.
What activates it
The engine uses the SMH semiconductor ETF as its primary sensor. A strong five-day return in SMH, combined with NVDA price confirmation, signals that compute demand is being priced in by the market. The engine does not rely on a single data point. Both instruments need to be moving in the same direction.
What it connects to
Strong compute demand flows downstream into:
- Memory — high-bandwidth memory for AI accelerators is in tight supply during demand spikes
- Networking — InfiniBand and Ethernet switching inside data centres carries the compute traffic
- Power — each chip requires electricity; demand growth means power consumption growth
- Software and cloud — the companies selling access to compute capacity benefit when demand is high
What creates risk
If hyperscaler earnings calls reveal that AI model training is shifting toward more efficient architectures, the raw compute demand story weakens. Demand for the same model output would drop even if the output grows, because each unit of output requires fewer chips.
Supply catching up to demand is the other risk. If capacity additions outrun end-market growth, pricing weakens.
How Decifer tracks it
SMH and NVDA are in Decifer's live sensor set. Their five-day returns are computed each cycle. The engine applies a threshold rule: both must clear their floors before the driver activates. This prevents a single outlier move in one instrument from triggering a false signal.