Nvidia raised its quarterly revenue forecast on Thursday, telling investors that orders for Blackwell Ultra accelerators exceed $180 billion and extend delivery schedules into 2027. The disclosure sent shares up 6.4 percent in after-hours trading and reassured markets that artificial intelligence infrastructure spending has not yet peaked.

Chief financial officer Colette Kress said hyperscalers — Microsoft, Amazon, Google, and Meta — accounted for 62 percent of data-center revenue in the quarter ended May. Sovereign AI programs in Saudi Arabia, Japan, and France contributed 11 percent, up from 4 percent a year earlier. Enterprise buyers made up the remainder.

Numbers That Mattered

Data-center revenue reached $38.6 billion, beating consensus estimates by $2.1 billion. Gross margins held at 75.1 percent despite ramp costs on new packaging lines. Nvidia reported supply of high-bandwidth memory remains the primary constraint, not wafer output.

Free cash flow totaled $14.8 billion. The company repurchased $9 billion in stock while increasing research spending on networking and CPU products that compete with custom silicon from cloud providers.

Why Backlog Persists

Cloud executives on concurrent earnings calls described AI workloads shifting from training to inference — a phase analysts once feared would reduce GPU intensity. Nvidia argued that inference at scale, especially for agentic applications with long context windows, consumes more compute per dollar of revenue than earlier chatbot deployments.

Chief executive Jensen Huang cited agent frameworks from major software vendors as a new demand layer. Each autonomous workflow may invoke models dozens of times per user task, multiplying token volumes.

Risks Investors Flagged

Bearish analysts questioned whether $180 billion in backlog includes double-ordering as customers hedge supply uncertainty. Kress said contracts carry cancellation penalties after deposit stages, though specifics vary by customer.

Export controls to China continue to cap revenue from that market. Nvidia ships downgraded variants compliant with U.S. rules; Chinese domestic competitors capture share in advanced nodes Nvidia cannot serve.

Competitive Pressure

AMD and custom ASIC programs at Google and Amazon win inference workloads priced below Nvidia's premium tiers. Nvidia responded by launching reference designs for rack-scale liquid cooling that reduce per-token cost 18 percent in internal benchmarks.

Intel's Gaudi 3 shipments remain modest. Broadcom and Marvell benefit from networking attach rates on Nvidia clusters rather than displacing GPUs directly.

Market Implications

The S&P 500 information technology sector rose 1.9 percent Friday in sympathy. Power and cooling suppliers — Vertiv, Eaton — gained on extended data-center build timelines. Bond markets showed little reaction; Nvidia finances expansion from cash rather than new debt.

For portfolio managers, the print reinforced concentration risk: Nvidia alone contributed a disproportionate share of index earnings growth in 2026. Diversification into power infrastructure and software monetization plays remains a common hedge strategy.