Nvidia revenue growth targets 70 percent as Blackwell GPU demand surges

The projected 70% growth indicates that the capital expenditure cycle for AI infrastructure is not yet cooling, despite the emergence of custom silicon from Amazon and Google. For the streaming and media ecosystem, this suggests that the cost of running large-scale generative AI models will remain tied to Nvidia's premium hardware margins for the foreseeable future. As AI-native startups continue to spend heavily on compute, the industry must prepare for a market where foundational hardware access dictates the pace of software innovation. Watch for whether competitors like Cerebras or Etched can secure meaningful production volume to challenge this pricing power in late 2026.

Nvidia's Blackwell architecture has become the default compute layer for the largest AI labs and hyperscalers, creating a supply-constrained environment that extends well beyond traditional data center cycles. In August 2026, Microsoft disclosed that its Azure AI infrastructure buildout had consumed over $80 billion in capital expenditure for fiscal year 2026, with the majority of GPU allocations going to Nvidia's GB200 NVL72 rack-scale systems. Amazon similarly confirmed in its Q2 2026 earnings call that AWS had deployed more than 500,000 Blackwell-class GPUs across its US regions, primarily to serve Anthropic's Claude model training workloads. These deployments illustrate why Jensen Huang's 70% growth projection rests on contracted demand rather than speculative orders.

The competitive landscape around Nvidia's dominance is intensifying, though no challenger has yet achieved production scale sufficient to pressure pricing. Cerebras Systems announced in July 2026 that its WSE-3 wafer-scale engine had been selected by three sovereign AI programs in the Middle East and Southeast Asia, representing the company's first multi-datacenter production contracts outside the United States. Etched, which raised $120 million in a Series B led by Primary Venture Partners in late 2025, reported in June 2026 that its Sohu ASIC had achieved 2.5x throughput per watt over Blackwell on transformer inference benchmarks, though the company has not yet disclosed any hyperscaler production deployment. Goldman Sachs analysts estimated in a September 2026 research note that custom silicon from Amazon's Trainium and Google's TPU programs collectively captured only 8% of total AI accelerator spend, leaving Nvidia's share above 85% for the third consecutive year.

For the streaming and media sector, Nvidia's pricing power directly affects the economics of AI-driven content production, recommendation engines, and real-time video processing at scale. The GB200 NVL72 configuration, which pairs 72 Blackwell GPUs with 36 Grace CPUs in a single rack, delivers up to 1.4 exaflops of FP4 inference performance per rack according to Nvidia's published specifications, a density that enables hyperscalers to consolidate video understanding and generative workloads that previously required distributed clusters. At the 2026 Hot Chips conference, Nvidia's chief scientist Bill Dally presented data showing that Blackwell's NVLink 5.0 interconnect reduced multi-GPU training communication overhead by 40% compared to Hopper, a gain that directly benefits large-scale video model training pipelines used by companies building AI-native content tools.