AI Infrastructure Bottlenecks: Why GPU Supply Won't Meet Enterprise Demand Until 2028
Former VP of Cloud Infrastructure at a hyperscaler discusses GPU allocation strategies, the real constraints on data centre build-outs, and why enterprise customers are being deprioritised.
This 52-minute expert call features a current VP of Cloud Infrastructure at one of the top three global hyperscalers discussing the company's GPU procurement pipeline and allocation strategy.
Key topics include why enterprise customers are being systematically deprioritised in GPU allocation versus internal AI workloads, the real timeline for next-generation data centre capacity, competitive dynamics between NVIDIA and AMD from a buyer's perspective, how custom silicon programs are reshaping negotiating leverage, and the internal economics of inference versus training workload allocation.