The decreasing cost of training neural networks is making the technology more accessible, while rising demand is driving significant investments in computing power. By 2030, global spending on AI infrastructure could approach $1.5 trillion, according to ARK Invest. 

AI adoption is outpacing the internet, and infrastructure is scaling to match.

We believe this is the beginning of a massive buildout, as consumers and enterprises signal strong demand. @downingARK shares the latest on AI infrastructure in a new blog. https://t.co/tatNHMFiuM

— ARK Invest (@ARKInvest) March 25, 2026

Prices Fall, Demand Grows

Analysts report that the costs of training neural networks are decreasing by 75% annually. Inference for models achieving over 50% in benchmarks is dropping even faster—by an average of 95%.

Source: ARK Invest. 

While cheaper technology typically reduces costs, the situation with artificial intelligence is different: as training and operating models become more affordable, the range of tasks where their application is economically viable expands.

The mass adoption of AI is occurring twice as fast as it did with the internet. In just three years, the technology's penetration reached 20%, while the web took over six years to achieve the same.

Corporate demand is also surging. The volume of token requests via OpenRouter has increased 28-fold since December 2024. Anthropic's annual revenue skyrocketed from $100 million in 2023 to $14 billion by February 2026. OpenAI reached 1 million business clients by November 2025.

Source: ARK Invest. 

Infrastructure Boom

Since the launch of ChatGPT, demand for accelerated computing has soared. Nvidia's annual revenue grew from $27 billion in 2022 to $216 billion in 2025. Analysts predict it will reach $350 billion in 2026.

Global investment growth in server systems accelerated from 5% per year (over the decade leading to 2022) to 30% in the past three years. According to ARK, GPU-based solutions and specialized chips (ASICs) have become the dominant segment, accounting for 86% of the server computing market.

Private investments in AI infrastructure exceeded $200 billion in 2025, with around $80 billion going to foundational model developers. Hyperscalers are seeking alternative funding forms: Meta's $30 billion deal with Blue Owl became the largest private capital transaction in history.

Chip Wars

The surge in demand has intensified competition among equipment manufacturers. AMD has managed to catch up with Nvidia in terms of total cost of ownership (TCO) for inference of smaller models. However, Nvidia maintains its performance lead in heavy models thanks to its Grace Blackwell architecture.

Source: ARK Invest. 

Hyperscalers are actively developing their own semiconductor solutions. Google has been designing TPUs for ten years. According to SemiAnalysis, using custom chips for internal tasks could reduce computing costs by 62% compared to Nvidia's architectures. 

Amazon is promoting Trainium, making it the preferred platform for training Anthropic. Microsoft is rolling out its second generation of Maia accelerators, optimized for inference.

Broadcom dominates backend design, partnering with Google TPU, Meta MTIA, and OpenAI's future chip. Citi forecasts the company's AI revenue will grow from $20 billion in 2025 to $100 billion in 2027.

Startups are emerging with new architectures. Cerebras, known for its Wafer Scale Engine chip, plans to go public this year. Meanwhile, Groq signed a $20 billion licensing agreement with Nvidia.

Forecast

ARK estimates that by 2030, annual investments in AI infrastructure will reach $1.5 trillion—a threefold increase over five years. The share of specialized ASICs in computing power will grow to one-third of the market.

Source: ARK Invest. 

“The infrastructure being built today is not a bubble ready to burst, but the foundation of a platform shift that occurs once in a generation. Useful AI agents are just beginning to be deployed; they are 'token-hungry' but far more capable than what users are accustomed to. Scaling these agents across millions of businesses will require colossal computing power, justifying the investments,” experts concluded. 

It is worth noting that Citrini Research experts predicted an economic collapse due to artificial intelligence.