Nvidia at a crossroads in AI evolution

Nvidia at a crossroads in AI evolution is profiled by BTW Media because published evidence links it to internet infrastructure, governance, operational dependencies, or market visibility.

• GPU aggregator warns Nvidia risks IBM-like decline amid centralised AI strategy
• High-end $30,000 GPUs seen as misaligned with scalable, decentralised AI future

What happened: A strategic inflection point for Nvidia’s AI future

At its annual GTC conference last week, Nvidia attempted to regain control of the AI narrative after a rocky start to the year. From the DeepSeek scandal to the underwhelming reception of the RTX 50-Series, 2024 has tested investor confidence in the semiconductor giant. But new concerns have surfaced regarding the company’s long-term direction—especially its reliance on ultra-high-end, expensive GPUs as the cornerstone of its AI infrastructure strategy.

Tory Green, CEO of GPU aggregator io.net, drew a sharp parallel in an interview with Capacity, warning that Nvidia could suffer the same fate as IBM. Once a dominant force in computing, IBM fell behind as the industry shifted towards decentralised, cost-efficient models. According to Green, Nvidia’s $30,000 GPUs are already out of step with the evolving demands of AI workloads.

Green’s argument centres on decentralisation: a shift towards distributing compute tasks across a diverse array of lower-cost GPU resources, rather than consolidating them in hyperscale data centres filled with expensive hardware. This model—fostered by edge providers, smaller data centres, and even individual contributors—can intelligently pair tasks with the right performance tier, reducing inefficiencies and costs.

Nvidia has acknowledged some need for change. The company has recently pivoted towards photonics and quantum computing, reversing its earlier dismissals. CEO Jensen Huang announced a new quantum research centre in Boston and launched “Quantum Day” at GTC. But critics like Green suggest these moves may be reactive rather than visionary.

In particular, Nvidia’s pricing remains a concern. Analysts note that its Blackwell chips are costlier to manufacture than prior generations, potentially undermining margins. If decentralised models gain traction, Nvidia’s centralised approach could become a costly liability rather than a strategic advantage.

Also read: DigitalBridge CEO on AI infrastructure and investment trends
Also read: Iliad Group invests $3B in AI infrastructure

Why it’s important

The concerns raised by Tory Green highlight a broader shift in how artificial intelligence infrastructure is evolving. At the heart of the debate lies the question of scalability: whether the future of AI belongs to centralised supercomputers or distributed GPU networks. Nvidia, currently the dominant provider of AI chips, is betting heavily on the former—with its flagship $30,000 GPUs powering most hyperscale data centres. However, if decentralised models gain traction, that investment could lose strategic relevance.

Decentralised computing offers flexibility by matching specific tasks to appropriate performance levels, something hyperscale models may struggle to achieve efficiently. With inference workloads growing in complexity and volume, the need to route them cost-effectively becomes paramount. Green argues that decentralisation exposes a broader pool of GPU resources, enabling smarter allocation and potentially unlocking substantial savings.

Importantly, Green’s analogy to IBM serves as more than historical commentary—it is a cautionary tale. IBM’s fall from mainframe dominance came not from a lack of technology, but from an inability to adapt to emerging computing paradigms. Nvidia’s future may hinge on whether it can balance the prestige and power of its flagship chips with the efficiency and accessibility demanded by next-generation AI developers and enterprises.

With photonics and quantum computing entering the strategy, Nvidia is not ignoring the tide. But it remains to be seen whether these bets will align with decentralised trends or further entrench a centralised architecture.