This AI (Bittensor) Could Be the Next Amazon of the 2030s

By: Andy

Everyone’s obsessing over “revenue” while completely missing what actually matters. Let me show you why the next Amazon of AI won’t be who you think.

CoreWeave just got called “the next Amazon of the 2030s” by Motley Fool. Their revenue grew 134% YoY to $1.4 billion, with $55.6 billion in backlog contracts. Sounds incredible, right?

Here’s what they didn’t emphasize: CoreWeave spent $10 billion on capex against $4.3 billion in revenue over the past 12 months. They’re spending $2.33 for every $1 they make, before paying employees, before electricity costs, before anything else. They’re hemorrhaging cash building data centers that might be obsolete in a year, because GPU lifecycles in AI training can be as short as 12 months. Yet everyone talks about their “revenue growth” like it means something.

To understand what actually matters, look at history.

Amazon 1994–2002 tells the story. After going public in 1997, revenue grew explosively — from $148M in 1997 to $2.76B in 2000 — while losses deepened every single year. Six straight years of losses while revenue grew. Skeptics everywhere: Amazon will never be profitable, it’s just burning cash, retail has no margins.

Then AWS launched in 2006. Quiet at first. Not the main business. But it changed everything. By 2015, AWS generated $7.9B in revenue at 24% operating margins, while Amazon’s retail business ran at 1–3% margins. Today, AWS is only 18% of Amazon’s revenue but over 60% of its operating profit. The marketplace everyone shops on barely makes money. The infrastructure marketplace makes all the money.

This is the pattern everyone misses. Amazon didn’t become Amazon by selling more books. It became Amazon by building infrastructure that other businesses needed, then creating a marketplace where value could flow through them. E-commerce was the hook. AWS was the business model.

Now look at AI infrastructure. Everyone is building the same thing: centralized cloud compute rented by the hour. CoreWeave, Lambda Labs, Together AI, Crusoe. Same model. Rent GPUs. Race for capacity. Burn billions building data centers.

Bittensor is doing something completely different: a marketplace where AI services compose and value flows through protocol.

Amazon became dominant not just by having warehouses, but by creating a marketplace. AWS became dominant not just by having servers, but by offering composable services. Bittensor is applying this model to AI.

Score (SN44) delivers computer vision models, reaching 76.9% accuracy (2.1% from human expert level) in four weeks, already with paying customers and revenue flowing. Templar (SN3) handles pre-training large language models through decentralized mining, publishing checkpoints anyone can use. Gradients (SN56) focuses on post-training and alignment, taking Templar checkpoints and making them conversational.

This is composable AI infrastructure. Score doesn’t need to train language models. Templar doesn’t need post-training. Gradients doesn’t need computer vision. Each specializes. Each composes. Value flows through the protocol. That’s the AWS model, not the data-center rental model.

The math that matters makes this clear. CoreWeave reports $4.3B in revenue alongside over $6B in operating losses, driven by $10B in capex. Meanwhile, Bittensor subnets already have profitable revenue flowing. Score has paying customers, SOTA performance, and was built in weeks, not years.

History repeats this lesson. In 2000, everyone obsessed over “eyeballs” and revenue growth, metrics with no path to profitability. Then Google arrived with a different model — not just search, but a marketplace where advertisers bid for keywords. The infrastructure was the hook. The marketplace was the business. By 2010, Google generated far more profit than Amazon, despite lower revenue years earlier.

The same split is happening now. AI infrastructure companies are burning billions, racing for capacity, competing on price, and watching margins compress. CoreWeave, Lambda, and Together AI all sell commoditized GPU hours. Bittensor, by contrast, builds specialized subnets that compose. No single company has to build everything. Value flows instead of competing.

The next Amazon of AI won’t be another cloud provider. Amazon didn’t win by building warehouses. AWS didn’t win by having the most servers. They won by creating marketplaces. CoreWeave’s $55.6B backlog is impressive, but it’s still selling the same compute everyone else sells. Bittensor has 120+ specialized subnets offering differentiated services across vision, language, post-training, alignment, and more.

What the market doesn’t see yet is that revenue is easy to measure, but margins, profitability, and composability are what matter. Some Bittensor subnets are already profitable at small scale. The infrastructure compounds instead of racing. Subnets build on each other.

This pattern has played out before — in the 2000s with data centers, and again with AWS. It’s playing out now in AI.

The marketplace always beats the warehouse. The platform always beats the vendor. The protocol always beats the company.

That’s what history shows. That’s what the data proves. And that’s what everyone misses while they obsess over revenue growth instead of business models.