Half a Trillion Dollars for Artificial Intelligence

Three of the world's most capital-rich technology companies are on track to burn through a historic sum on artificial intelligence infrastructure within a single year. According to Nasdaq Markets, Alphabet has raised its capital expenditure forecast for 2026 to up to $205 billion. Amazon follows closely with around $200 billion, while Meta is planning investments of up to $145 billion.

Combined, that exceeds $550 billion — from just three companies.

Over half a trillion dollars. From three companies. In one year.
Tech Giants Blow Past $500B: Nvidia Reaps the Rewards - Bilde 1

Nvidia: The Undisputed Toll Collector

The immediate winner is Nvidia. The company controlled an estimated 93 percent of server GPU revenue in 2024 and around 90 percent of the data center chip market, according to industry estimates cited by Nasdaq Markets. When Alphabet, Amazon, and Meta place hardware orders of this magnitude, a disproportionately large share flows into Nvidia's coffers.

This is happening even as Nvidia's latest Blackwell chips are reportedly pre-ordered roughly twelve months out. Demand far exceeds available capacity.

$550B+
Combined AI capex from Alphabet, Amazon, and Meta in 2026
93%
Nvidia's share of server GPU revenue in 2024
Tech Giants Blow Past $500B: Nvidia Reaps the Rewards - Bilde 2

Hyperscalers Are Also Betting on Their Own Chips

Even as they purchase Nvidia hardware at enormous scale, all three companies are simultaneously working to reduce that same dependence over the longer term.

Google and TPUs

Google was the first major cloud player to develop its own AI chips. The company's eighth-generation TPUs — TPU 8t for training and TPU 8i for inference — power its Gemini models, among others. According to Nasdaq Markets research, Google has used its AI system AlphaChip to design the chip configuration itself, reportedly cutting design time from months to hours.

AWS Trainium and Inferentia

Amazon Web Services offers Trainium chips that, according to the company, deliver up to 50 percent better cost-to-performance for model training compared to Nvidia's A100 and H100. The latest Trainium3 can scale up to 144 chips in a single system. AWS is also reportedly exploring selling Trainium chips to third parties — a move that could directly challenge Nvidia's core business.

Meta's MTIA

Meta is developing its own Meta Training and Inference Accelerator (MTIA), designed to run and train the company's own AI models more efficiently and reduce operating costs.

The Counterweight: Decentralized GPU Capacity

Alongside the hyperscalers' centralization, an alternative market is emerging. Decentralized AI compute networks — such as io.net, Aethir, Akash Network, and Render Network — offer GPU capacity at a fraction of hyperscaler prices, according to industry reports.

It should be noted, however, that these figures largely originate from the players themselves and should be read with a critical eye. The market for decentralized AI compute was estimated at $1.8 billion in 2025 — a small sliver compared to the centralized giants.

For Norwegian and European AI startups that lack access to Nvidia hardware through the standard queue, such alternatives may nonetheless offer genuine practical value.

What This Means Going Forward

The enormous capital injection from Alphabet, Amazon, and Meta in 2026 confirms that the race for AI infrastructure is not slowing down — it is accelerating. Nvidia reaps the rewards in the short term, but the hyperscalers' parallel investment in proprietary silicon suggests they have no intention of remaining dependent on a single supplier over the long term.

For investors and market participants, it is worth noting that these investment levels are without historical precedent — and that the benefits will take time to materialize on companies' bottom lines.