An influential investor recently highlighted three interconnected trends shaping the future of AI hardware, suggesting a new landscape is emerging beyond the current market leaders.
The first claim is that Amazon's custom AI chip, Trainium, is significantly undervalued. This isn't just speculation; it's backed by Amazon's colossal investment plans. The company has earmarked around $200 billion for capital expenditures in 2026, with a large portion dedicated to AWS infrastructure, including its own chips. Furthermore, AWS has launched "AI Factories," a program designed to make it easier for customers to use Trainium, effectively creating a dedicated sales channel to compete with NVIDIA's GPUs where total cost and latency allow.
The second, more futuristic claim, is the viability of space data centers within two years. This idea stems from a very real problem on the ground: a shortage of power. The largest U.S. electrical grid, PJM, has seen electricity prices spike, largely due to the immense energy demands of new AI data centers. This power crunch is forcing tech giants to look for alternatives. We're already seeing this happen. Blue Origin has filed plans with the FCC for an orbital computing constellation, and Google is reportedly in talks with SpaceX for a similar project. These are no longer sci-fi concepts but concrete business explorations.
The final piece of the puzzle is TSMC, the world's leading chip manufacturer. The argument is that TSMC's measured approach to increasing production capacity is preventing an AI hardware bubble. Instead of flooding the market, TSMC is carefully managing the expansion of its most in-demand technology, an advanced packaging process called CoWoS. This creates a deliberate bottleneck, which keeps the supply of high-end AI chips tight. This discipline helps maintain high prices and prevents the kind of oversupply that could lead to a market crash.
In essence, these three narratives are linked. The insatiable demand for AI is straining Earth's power grid, pushing companies toward space. Meanwhile, Amazon is building its own powerful alternative (Trainium) to the dominant GPUs. And holding it all in balance is TSMC, whose cautious strategy ensures the entire ecosystem grows sustainably, not explosively.
- Glossary -
- Trainium: An AI accelerator chip designed by Amazon Web Services (AWS) for training and running artificial intelligence models.
- CoWoS (Chip on Wafer on Substrate): An advanced packaging technology from TSMC that stacks multiple chips together to create a single, more powerful processor, essential for modern AI GPUs.
- Capex (Capital Expenditure): Funds used by a company to acquire, upgrade, and maintain physical assets such as property, buildings, and equipment.