The AI Trade Beyond Models: Who Builds the Memory, Power, Cooling and Compute
AI feels weightless. You type a question. A chatbot replies. But behind that answer is one of the largest physical infrastructure buildouts in modern markets.
AI is not just software. It is memory. It is foundries. It is specialised inference chips. It is GPU cloud capacity. It is optical fibre. It is on-site power. It is liquid cooling. It is safety-certified software for cars, robots, factories and machines.
That is why some of the most important AI companies in 2026 are not model developers at all. Micron feeds the GPU with high bandwidth memory. Intel is trying to rebuild advanced chip manufacturing through 18A. Etched is betting transformer inference deserves its own specialised silicon. CoreWeave rents GPU capacity to labs and hyperscalers that cannot build fast enough. Nokia connects AI data centres through optical networking and explores AI-RAN with Nvidia. Bloom Energy tackles the power bottleneck when grids cannot connect quickly enough. Vertiv distributes electricity and removes heat inside high-density AI data centres. BlackBerry QNX provides the safety-certified software layer for physical AI in vehicles, robots and industrial systems.
Together, they form the body behind the chatbot.
The opportunity is real. So is the risk.
Across the AI infrastructure stack, the same pattern keeps appearing: Huge backlog. Large customer commitments. High customer concentration. Big valuation premiums. Execution that must happen over several years.
Markets are not only pricing AI demand. They are pricing the assumption that backlog converts into revenue on schedule, customers stay committed and infrastructure gets built without major delays. That is why Decentralised News focuses on the Perfection Premium Ledger.
The question is not only whether a company has AI exposure. The question is whether it is already priced as if nothing can go wrong.
AI has a body now. The next winners will be the companies that turn memory, power, cooling, networking and compute into durable cash flow.
The Stablecoin Margin War: Circle, USDC, Coinbase, Open USD and Visa Explained
Stablecoins are usually described as digital dollars. That misses the business model.
Circle’s USDC is backed by cash and short-term Treasuries. Those reserves earn interest. That interest is the engine of the business.
In Q1 2026, Circle generated $694.1 million in total revenue and reserve income, according to the Decentralised News framework. $652.5 million of that came from reserve income. That is roughly 94% of the total. So Circle is not only a stablecoin issuer. It is a reserve-income business with a token wrapper.
The problem? Circle does not keep all the economics. Under its revenue-sharing agreement with Coinbase, Coinbase receives all the interest income on USDC held directly on Coinbase and half the interest income on USDC held elsewhere.
In 2024, Coinbase received $908 million from Circle, roughly 54% of Circle’s total revenue that year, despite directly holding only about one-fifth of USDC supply. That is why stablecoin circulation alone can be misleading.
A stablecoin can grow supply and still face margin pressure if the distribution partners capture a large share of the float. This is now the real stablecoin war. Not just USDC versus USDT. Not just regulated versus offshore. Not just blockchain speed or redemption rails. It is a fight over who owns the customer, who controls distribution and who keeps the interest.
That is why Open USD matters. More than 140 companies, including Visa, Mastercard, Stripe, BlackRock, BNY and Coinbase, launched a consortium-governed stablecoin designed to return most reserve income to distribution partners rather than one issuer.
Visa’s Stablecoin Platform then made the strategy even clearer. Visa does not need one stablecoin to win. It can profit from the rails. If USDC wins, Visa can support it. If Open USD wins, Visa can support it. If USDG wins in certain corridors, Visa can support that too.
The AI Infrastructure Bubble Debate: $7.6 Trillion of Compute, Power and Risk
The AI boom is usually framed as a chip story. It is not. It is becoming one of the largest infrastructure capital cycles in modern economic history.
Amazon, Microsoft, Alphabet and Meta are guiding to roughly $725 billion in combined 2026 capital expenditure. Goldman Sachs models a broader $7.6 trillion AI infrastructure buildout between 2026 and 2031. That is not a normal software cycle. That is compute, data centres, memory, cooling, land, transformers, grid connections and electricity.
Nvidia is the visible winner. Its data centre business has become the headline symbol of the AI boom. But the deeper bottleneck is no longer only GPUs. It is power. AI data centres are now colliding with the limits of the electricity grid. U.S. grid interconnection queues have become a major constraint.
In some regions, data centre demand is already feeding into higher capacity prices and household electricity bills. That changes the politics of AI. When the cost of the buildout starts showing up outside the data centre, regulators notice. The revenue side is just as important.
OpenAI and Anthropic are scaling quickly. But their revenue is still small relative to the hundreds of billions being spent on infrastructure to serve future AI demand.
That is the real question: Can inference demand grow fast enough to justify the capital being deployed today? Not just chatbot demand. AI agents. Enterprise workflows. Coding systems. Search. Customer support. Data analysis. Robotics. Autonomous operations.
This is why Decentralised News built the DN Query Cost Ledger. The tool focuses on the missing number in the public conversation: What does one AI query actually cost to run? And how much of what users pay is electricity versus the full capital stack behind the query? Chips. Memory. Cooling. Networking. Depreciation. Data centre financing. Cloud margin. Model operations.
The electricity cost may be small. The infrastructure behind the query is not. For crypto and DePIN investors, this matters.