Nvidia CEO Jensen Huang is set to take the stage in Berlin on Wednesday, as the world’s most valuable company faces heightened investor scrutiny over whether the massive artificial intelligence spending boom – the engine behind the chipmaker’s meteoric growth – can deliver long-term returns. The conference agenda focuses on sovereign AI, industrial automation and operational infrastructure – key sectors where Nvidia hopes to drive demand beyond the handful of US tech giants currently dominating the investment landscape. It follows Huang’s meeting with German Chancellor Friedrich Merz on Monday, when Nvidia and partner Siemens showcased AI deployments in production planning and circuit-board inspection. Huang’s keynote at Nvidia’s GTC conference, scheduled for 11am local time or 5pm Hong Kong time, comes as its customers battle rising borrowing costs and mounting delays in building the data centres required to run its processors. Financing risks across the AI ecosystem have come into sharp focus following a sell-off in the shares and bonds of CoreWeave, an Nvidia-backed cloud provider and major graphics processing unit (GPU) buyer. Due to its reliance on debt, CoreWeave is widely considered a bellwether for whether specialised AI cloud operators can sustainably fund their expansion. Research firm Gartner warned last week that more than US$3.5 trillion in global AI infrastructure spending could face impairment by 2030 if investment continued on its current trajectory, noting that projected revenues would be insufficient to yield profitable returns on planned capital expenditure. For Nvidia, these financial dynamics carry immediate consequences: even if demand for computing power remained strong, financial stress among key customers could stall purchase timing. Wall Street remains divided. Goldman Sachs upgraded CoreWeave to “buy” on Monday, arguing that investors might be overestimating financing costs and underestimating contract returns. The bank noted that distressed yields on unsecured bonds offered a poor guide to the asset-backed financing that underpinned most of CoreWeave’s expansion. Physical bottlenecks pose an equally urgent challenge. At least 47 major US data centre projects had been delayed, cancelled or blocked this year amid local opposition, permitting delays and power grid constraints, according to a Financial Times report on Sunday. With delivery schedules now as important to growth as chip performance, purchase commitments are only as valuable as the physical space available to plug them in. Research firm SemiAnalysis offers a more optimistic outlook. In a report published this week, it forecast roughly 50 gigawatts of additional global data centre capacity by 2027. Pointing to more than 24GW already under advanced construction, SemiAnalysis argued that modular designs, on-site power generation and international expansion outside the US would help overcome bottlenecks. At the same time, intensifying competition threatens to chip away at Nvidia’s market share. Custom accelerators like Google’s tensor processing units and Amazon Web Service’s Trainium offer major hyperscalers viable alternatives for specific workloads. While Goldman Sachs this month raised its shipment forecasts for Nvidia-powered server racks, it also lifted projections for competing AI servers due to growing demand for custom AI chips, signalling a broader market shift towards diversified hardware. Despite the headwinds, Huang has set an ambitious bar. At the company’s GTC event in California in March, he projected at least US$1 trillion in revenue for Nvidia’s Blackwell and Rubin platforms between 2025 and 2027, pitching an integrated ecosystem combining GPUs, central processing units, networking and specialised inference processors – all designed to slash the cost of AI output. Nvidia said in August that production shipments of its Vera Rubin platform were under way, projecting around 70 per cent revenue growth for its 2028 financial year – a target it noted remained bound by supply constraints.