Listen to the audio version of this article (generated by AI).
Last Wednesday, Walmart (WMT) began preparing for a different kind of shopper… the non-human kind. One that doesn’t need to browse the aisles or even look at a screen.
The retailer announced tools that will let customers authorize AI agents to shop on their behalf at Walmart and Sam’s Club. Shoppers will be able to set spending limits, require purchase approvals, and track their agents’ orders separately. A new gateway, already in beta, gives those agents a dedicated connection to Walmart’s shopping systems.
Think about what that means.
A routine household chore is becoming something you can delegate to software. And Walmart is building the systems to make that delegation part of everyday commerce.
On the same day, Anthropic introduced a research preview of Claude Duet, an assistant designed to work across connected apps, continue tasks after users step away, and check with them before acting on their behalf.
These are early deployments that customers have yet to fully trust. But for investors wondering where the next wave of AI demand comes from, this is a concrete answer: software doing more of the work people currently do themselves.
Each additional task requires computing power. If agents become useful enough to handle recurring responsibilities for millions of people and businesses, their usage could extend well beyond today’s question-and-answer interactions.
That brings us to Monday’s Anthropic IPO filing.
The company behind Claude disclosed $11.6 billion in second-quarter revenue and its first GAAP operating profit: $974 million. Its rapidly expanding business is beginning to demonstrate the economics that could support this next phase of adoption.
Folks, that’s why I think this filing matters across the AI trade.
I’ve been bullish on Anthropic for a long time. Back in April, I said I thought it could become the most valuable company in the world within five years. That remains my long-term thesis.
Now we have a clearer view of the business behind that conviction — and, alongside fresh earnings from its infrastructure suppliers, stronger evidence of where the investment opportunities may be developing.
Let’s connect the dots.
AI’s Economics Are Improving
Early last year, Anthropic faced the problem that has haunted the entire AI trade: Its technology was impressive, but delivering it cost far more than customers were paying.
For every dollar the company spent on computing and infrastructure, it brought in about 40 cents in revenue. That left investors funding the difference, before accounting for the other costs of running the business.
The bears had a reasonable question: What happens when investors get tired of covering the bill?
Over the next five quarters, Anthropic began changing that equation.
More businesses put Claude to work. As usage expanded, revenue grew faster than computing expenses. By the second quarter of this year, every dollar of compute spending supported roughly $2.08 in revenue.
That’s a meaningful transition. Anthropic was generating enough revenue to cover its computing bill and contribute substantially toward the rest of the business. That quarter, it recorded its first GAAP operating profit: nearly $1 billion.
You wouldn’t necessarily recognize that progress from the headline-grabbing $80.7 billion first-half net loss. Most of that figure reflects noncash accounting adjustments tied to convertible financing. The distinction matters when assessing how the underlying business is developing.
There’s another revealing detail here: Anthropic is using Claude to help build Claude. The company says its AI generated more than 90% of September’s code contributions. Human oversight remains essential, but the product Anthropic sells is also helping its own employees get more work done.
Folks, that’s the economic progression we’ve been watching for: wider adoption, more revenue from each computing dollar, and tools that improve productivity inside the companies deploying them.
The next wave of capacity could pressure margins, and Anthropic’s results don’t establish that every AI company (or every customer) will earn attractive returns.
But they strengthen the case that useful AI can support a profitable business. And that gives the infrastructure build-out a firmer foundation.
Why I Think Anthropic Could Become the World’s Most Valuable Company
My thesis starts with AI becoming embedded in the work businesses perform every day: Writing software, analyzing information, supporting customers, etc.
A trusted provider of that intelligence could participate in an enormous amount of economic activity.
And Anthropic’s customer data makes that opportunity tangible.
Customers spending at least $100,000 annually nearly doubled in the third quarter to more than 11,900. Net revenue retention reached roughly 900%, signaling extraordinary spending expansion among existing customers.
More than half the Fortune 500 already spend at least $100,000 annually with Anthropic. Yet average spending among those customers remains below 0.1% of total expenses.
The next leg depends on agents becoming useful enough to handle more work and reliable enough that businesses keep expanding deployment.
That’s the foundation of my long-term thesis.
Follow the Computing Commitments
Anthropic still describes itself as compute-constrained. And our tally of fixed and minimum supplier commitments approaches $850 billion, with a broader potential total closer to $1 trillion. Treat those as estimates, as amended contracts may overlap, and the filing leaves the aggregate headline figure blank.
Even with that qualification, the scale is consequential.
Agreements span major chipmakers and infrastructure providers, while Anthropic is moving toward directly leased chips, dedicated data centers, and on-site power.
Google (GOOG, GOOGL) and Broadcom’s (AVGO) TPU ecosystem stands out. The expansion also supports demand for Advanced Micro Devices (AMD), high-bandwidth memory, Taiwan Semiconductor Manufacturing (TSM), and Nvidia (NVDA).
Our inference is that dedicated facilities should extend the opportunity into power, cooling, networking, and optics.
That’s one reason I continue to like Astera Labs (ALAB). Its connectivity technology helps AI systems work together, and its role in Amazon’s (AMZN) Trainium ecosystem provides exposure to Anthropic’s expansion.
You can participate in this build-out before ANT starts trading.
The Supply Chain Confirms the Demand
The latest earnings reports reinforce that view.
Cadence Design Systems (CDNS) posted a record $8.2 billion backlog. Rambus (RMBS) guided to accelerating product growth. Amkor Technology (AMKR) collected roughly $3 billion in customer prepayments. Corning’s (GLW) optical business grew 44%.
Celestica (CLS) guided to $40 billion in 2027 revenue, roughly 90% above its updated 2026 outlook, with customer demand forecasts pointing higher again in 2028.
These businesses span chip design, memory interfaces, packaging, connectivity, and finished systems.
That breadth strengthens the case that we’re still in the middle of a substantial build-out.
We remain bullish on Cadence, Celestica, and Corning. ALAB remains among our favored dip-buy opportunities.
Financing Will Separate the Winners
The bears have legitimate concerns.
Anthropic’s estimated supplier commitments far exceed its contracted future customer revenue. Supplier financing creates complicated relationships between buyers and sellers. And a massive IPO could draw capital away from other AI stocks.
But the strongest buyers continue to attract funding — and they’re putting that money toward additional computing capacity.
That leaves investors with an important distinction to make. Some infrastructure operators could struggle with financing even as demand for the equipment, power, and technology behind the build-out keeps expanding.
We expect company selection to matter more as that gap widens.
Anthropic’s filing helps explain why. Its economics are improving, customers are expanding usage, and the company still needs more computing power. Success creates another round of demand for infrastructure that already takes years to build.
That’s where my research has been focused, including an ambitious project connected to Elon Musk that I believe could help address some of AI’s most consequential constraints.
The implications extend beyond the next generation of data centers. They reach into America’s ability to secure the technology its economy increasingly depends on.
In a special presentation, I explain the shift, outline three steps investors can take to position themselves, and reveal the name and ticker of one investment I believe is positioned to benefit, free of charge.
Watch my presentation to see where I believe the next wave of AI investment could flow.
