This article was originally published on Fool.com. All figures quoted in US dollars unless otherwise stated.
Key Points
- Nvidia controls the irreplaceable GPU infrastructure powering every major artificial intelligence (AI) model, with data center revenue hitting $41.1 billion last quarter.
- Meta Platforms' free LLaMA models are a Trojan horse for the world's most profitable advertising machine, generating $46.6 billion in quarterly ad revenue.
- Amazon's AWS dominates AI hosting with 31% market share, turning every company's AI adoption into recurring revenue streams.
The artificial intelligence (AI) revolution needs three things to function: the chips to run it, the models to power it, and the cloud to host it.
These three companies own those critical layers, making them permanent fixtures in the AI economy, regardless of which chatbot wins or which start-up goes viral.
The silicon monopoly
Nvidia (NASDAQ: NVDA) has turned AI compute into a tollbooth business. Every major AI breakthrough -- from ChatGPT to Midjourney to autonomous vehicles -- runs on Nvidia silicon.
Q2 fiscal 2026 total revenue hit $46.7 billion, up 56% year over year, with data center revenue alone reaching $41.1 billion. The company's Blackwell architecture is already sold out through 2026, with customers including Microsoft, Alphabet, and Meta Platforms (NASDAQ: META) placing orders measured in tens of thousands of units.
The moat goes deeper than just fast chips. CUDA, Nvidia's programming platform, has 5 million developers trained on its architecture -- a switching cost measured in years of retraining. Trading at 39 times forward earnings might seem rich, but consider this: 73% gross margins, $54 billion in Q3 guidance, and a new $60 billion buyback authorization. When Jensen Huang says "the next industrial revolution has begun," he's not being hyperbolic -- he's describing a market Nvidia already owns.
The attention merchant's AI play
Meta Platforms discovered something competitors missed: You don't need to charge for AI when you own the world's attention. The company's LLaMA models, given away free, have become the Linux of AI -- powering everything from start-up prototypes to enterprise deployments. Meanwhile, Meta's real AI runs behind the scenes, optimizing 3 billion daily users' feeds and delivering ads with surgical precision.
Q2 2025 revenue jumped 22% to $47.5 billion, with advertising alone generating $46.6 billion. Reality Labs burns over $4 billion quarterly chasing the metaverse, but that's a rounding error against $20 billion in quarterly operating income. The company sits on $47 billion in cash and marketable securities while trading at just 28 times forward earnings -- cheaper than most software companies by a wide margin. Meta's platforms generate billions of new posts, photos, and videos daily -- a proprietary training dataset no competitor can buy at any price.
The "everything store" for AI
Amazon (NASDAQ: AMZN) plays the ultimate arms dealer strategy -- it doesn't matter whose AI wins, as long as they're all paying AWS to run it. Q2 2025 saw AWS revenue hit $30.9 billion, up 17.5%, capturing 31% of global cloud infrastructure spending. The unit's 32.9% operating margins might be down from its prior peak, but that's because Amazon is aggressively building AI capacity. Its custom Trainium and Inferentia chips offer customers alternatives to Nvidia at 40% lower cost, while Bedrock makes deploying AI models as simple as clicking a button.
The bull case is straightforward: Amazon generated $121 billion in operating cash flow over the trailing 12 months, giving it unlimited firepower to build AI infrastructure. AWS's $195 billion backlog (up 25% year over year) locks in future revenue with multiyear contracts and massive switching costs. Every AI start-up, enterprise deployment, and government project strengthens the moat.
At 34.6 times forward earnings, Amazon trades like the toll road to the AI economy -- and that's precisely what it is.
The three pillars of permanent value
These aren't speculative AI bets chasing viral adoption. Nvidia dominates the compute layer, controlling more than 90% of the AI training GPU market. Meta owns the consumer layer, with 3 billion users generating unmatched training data for its algorithms. Amazon controls the enterprise layer, with cloud infrastructure and multiyear contracts that no start-up can replicate.
Hundreds of AI companies will rise and fall over the next decade, but these three will collect rent from all of them -- making them true buy-and-hold candidates for the long term.
This article was originally published on Fool.com. All figures quoted in US dollars unless otherwise stated.
