Technology

Amid a flurry of AI developments and geopolitical shifts, NVIDIA's recent quarterly earnings call has raised compelling questions about whether investors are focusing too narrowly on short-term figures and overlooking the broader technological and strategic evolution underway.

Richard Clode, Portfolio Manager at Janus Henderson Investors, shared insights following the earnings report, highlighting a key theme: AI tokens are the new oil. Tokens—units of data processed during AI model training and inference—have seen a dramatic surge, indicating not only escalating usage but also the growing intelligence and complexity of AI models.

Microsoft revealed that it generated 100 trillion tokens in Q1 2025, up fivefold year-on-year, and processed 50 trillion tokens in March alone. Google, at its I/O event last week, disclosed that it now processes over 480 trillion tokens monthly, marking a 50x increase year-on-year. NVIDIA itself reported a 10x increase in token generation over the past year, a key metric of expanding AI utility.

This token explosion underscores the rising dominance of reasoning models, which, unlike basic chatbot queries, perform complex computations, explore multiple problem-solving pathways, and validate answers—generating significantly more tokens in the process. According to NVIDIA, reasoning models can be 100x to 1000x more compute-intensive than standard inference models, unlocking new applications in agentic AI, autonomous vehicles, and humanoid robotics.

Blackwell: Designed for the Reasoning Era

In response to this evolution, NVIDIA’s new Blackwell chip was purpose-built to handle the intensive demands of reasoning models. Unlike its predecessor, Hopper, which was optimized for training large language models, Blackwell delivers 40x the inferencing performance. The NVL72 rack, weighing two tonnes and containing 1.2 million components, reflects this leap in infrastructure design, marking NVIDIA's first rack-level product architecture.

Despite initial production hurdles—including a chip re-spin at TSMC and board yield issues—NVIDIA is now scaling up supply. The company reports that hyperscalers are deploying up to 1,000 NVL72 racks weekly, equivalent to 72,000 Blackwell GPUs. Sampling has also begun for Blackwell Ultra GB300 racks, which promise a 50% performance boost and smoother rollout in the second half of 2025.

AI and Geopolitics: A Shifting Landscape

The intersection of AI and geopolitics remains pivotal. Following the US ban on H20 chips, NVIDIA’s presence in the Chinese datacenter market appears to be nearing its end. However, CEO Jensen Huang reiterated that isolating China may be counterproductive, as the country already possesses advanced AI capabilities. Huang advocated for integration instead, suggesting it would be more strategic to have China’s vast AI market rely on US-built infrastructure.

NVIDIA is reportedly exploring the possibility of developing degraded AI chips for the Chinese market, although regulatory and competitive hurdles remain. Meanwhile, the recent rescinding of the Biden-era AI Diffusion Rule has enabled high-profile AI chip deals in Saudi Arabia and the UAE, reinforcing NVIDIA’s position in the Middle East.

Additionally, NVIDIA clarified that while Singapore accounts for a large portion of reported sales, the end-users are predominantly US-based, addressing concerns about chip diversion.

In a strategic pivot toward domestic manufacturing, NVIDIA confirmed that by mid-2026, AI chips will be produced at TSMC fabs in Arizona, and AI supercomputers assembled in Texas, supporting President Trump’s agenda to rebuild US-based tech manufacturing.

Looking Beyond the Earnings

While NVIDIA's earnings remain a barometer for both the technology sector and global geopolitics, Richard Clode cautions investors against missing the larger trend. The inflection point in AI infrastructure is shifting from training large frontier models to scalable inferencing deployments. This transformation, powered by more intelligent reasoning models and rising token generation, is laying the foundation for a longer, broader, and more sustainable growth trajectory—not only for NVIDIA but also for the global economy.

Meta Tags (comma-separated):
NVIDIA, AI tokens, Blackwell chip, reasoning models, Janus Henderson, AI infrastructure, token generation, AI inferencing, Microsoft AI, Google AI, AI chip supply, geopolitics AI, UAE tech news, AI news 2025, Jensen Huang, Richard Clode, NVL72, TSMC Arizona, AI chip ban China, AI growth trends

Meta Keywords (comma-separated):
NVIDIA earnings, AI token surge, reasoning AI models, Blackwell GPU, NVL72 racks, AI infrastructure shift, Microsoft token data, Google AI processing, Janus Henderson insights, AI in UAE, AI chip manufacturing USA, AI chips China ban, AI geopolitics, President Trump AI deals, Saudi UAE AI chip deals, AI supercomputers Texas, TSMC Arizona chips, AI investment trends 2025, sustainable AI growth, agentic AI, autonomous AI systems

• Tags
• #Business
• #Magazine
• #Lifestyle