Newsletter
AI

AI Shopping Trust, Anthropic's IPO, and Growing Infrastructure Costs

Photo: Brecht Corbeel on Unsplash
In this article 13 sections
  1. Anthropic is approaching its Wall Street moment
  2. The infrastructure bill keeps growing
  3. AI agents are becoming a security problem, too
  4. My AI Diary: The Week AI Stopped Being Just a Chatbot
  5. November 15, 2026 | AI | 7 min read
  6. By Thomas | financial enthusiast
  7. In 30 seconds
  8. Another Sunday, another AI reality check
  9. Google is turning up the pressure
  10. OpenAI wants agents to handle your money
  11. What I’m watching next
  12. The bottom line
  13. Sources & further reading

What if the agent selects the wrong product? What if a website provides misleading information? What happens when a purchase is technically authorized but based on a misunderstanding?

Payment approval is an important safeguard, but it doesn’t solve every problem.

An agent also needs to explain what it is buying, why it selected that option, and what alternatives it rejected.

Otherwise, we’re simply replacing manual shopping with automated mistakes.

My view is that the winning AI shopping experience won’t necessarily be the one that completes purchases fastest.

It will be the one people trust enough to use repeatedly.

And trust is much harder to build than a checkout button.

Anthropic is approaching its Wall Street moment

While Google and OpenAI are competing for users and developers, Anthropic is facing a different kind of test.

The company has begun marketing its planned initial public offering, with a reported indicative price range of $70 to $78 per share.

For me, this is one of the most interesting business developments in AI right now.

An IPO Initial public offering. A company selling its shares on a stock exchange for the first time. forces investors to look beyond impressive model demonstrations.

Suddenly, the questions become much more traditional.

How much revenue can the company generate? How expensive is it to operate its models? How quickly are customers adopting its products? And, most importantly, can revenue eventually grow faster than the cost of providing the service?

These questions aren’t particularly exciting.

But they matter.

Building advanced AI systems requires enormous computing resources, specialized hardware, electricity, networking, and engineering talent.

Those costs don’t disappear because a model becomes popular.

Anthropic’s IPO process could offer investors another opportunity to evaluate the economics of a major AI model developer.

That doesn’t automatically make the company an attractive investment.

A strong product and an attractive stock valuation are two very different things.

I’ve learned that lesson from watching technology markets over the years.

Sometimes the technology really does change the world.

That still doesn’t mean every company building it will generate attractive returns for shareholders.

The infrastructure bill keeps growing

There is another part of this story that I think deserves more attention.

Electricity.

I’ve written about this before, but the more I follow the AI industry, the more convinced I become that power infrastructure is one of its defining constraints.

We spend a lot of time discussing Nvidia NVDA ▼2.0%, GPU Graphics processing unit. A chip with thousands of small cores; the workhorse for training and running AI models., and model performance.

Fair enough. Those are important.

But an AI accelerator sitting in a data center is useless without reliable electricity, cooling, networking, and the equipment needed to support it.

And the industry’s ambitions are expanding.

Training large models is expensive. Running millions of AI interactions every day also consumes resources. Autonomous agents that perform several steps, call tools, and coordinate with other agents can add even more computational work.

That creates a fascinating economic problem.

AI companies want to make their services cheaper and more accessible.

At the same time, they need to finance increasingly complex infrastructure.

Investors should pay attention to that tension.

I still watch the major chipmakers, but I also keep an eye on data center operators, power equipment suppliers, memory manufacturers, and companies involved in cooling infrastructure.

Not because every company connected to AI will be a winner.

Quite the opposite.

When an industry attracts this much capital, some investments will inevitably disappoint.

The difficult part is separating sustainable demand from overly optimistic expectations.

AI agents are becoming a security problem, too

Another theme running through recent developments is the growing importance of agent security.

The more independence we give AI systems, the more seriously we need to think about permissions.

A chatbot producing an incorrect answer is annoying.

My AI Diary: The Week AI Stopped Being Just a Chatbot

November 15, 2026 | AI | 7 min read

Google is pushing the model race forward, OpenAI wants AI agent An AI system that plans and carries out multi-step tasks on its own, using tools such as a browser or code. to handle payments, and Anthropic is preparing for Wall Street. After another busy week in artificial intelligence, I’m starting to think the biggest change isn’t about smarter models anymore. It’s about who controls what AI can actually do.

By Thomas | financial enthusiast


In 30 seconds

AI had another eventful week. Google expanded its Gemini 4 lineup, OpenAI introduced a wallet for AI-powered purchases, and Anthropic moved forward with its IPO preparations. Meanwhile, autonomous agents are becoming more capable, and the infrastructure needed to run them keeps getting more expensive.

My takeaway? The industry is moving from answering questions to performing real tasks. That creates opportunities, but also some uncomfortable questions about trust, costs, and control.

Another Sunday, another AI reality check

It’s Sunday, November 15, and I’ve spent part of my morning catching up on the latest AI developments.

At this point, keeping up with artificial intelligence almost feels like a second job.

A new model appears. Another company announces an agent. Someone publishes a benchmark showing that their system is faster, cheaper, or supposedly smarter than everything else.

And then, a few days later, the next announcement arrives.

But this week caught my attention for a different reason.

For once, I wasn’t particularly interested in which model scored a few points higher on a benchmark.

I was looking at something much more practical.

What happens when AI stops giving us suggestions and starts making decisions, spending money, and completing tasks?

Because that’s where this industry seems to be heading.

And I think we’re still underestimating what that means.

Google is turning up the pressure

Google continued its AI push with Gemini 4 Hydrogen, adding another option to an increasingly competitive model lineup.

The interesting part isn’t just its reported reasoning and coding performance. It’s the combination of speed, capability, and pricing.

For developers, that combination matters more than another impressive demonstration.

A model that completes a task faster can reduce waiting time. A model that requires fewer retries can reduce the actual cost of automation. And a model that handles longer workflows reliably can become useful for jobs that previously needed constant human supervision.

I work in IT, so this is the part I pay attention to.

Not whether a model can write a beautiful poem.

I want to know whether it can understand an existing codebase, identify a problem, make a reasonable change, and avoid breaking three other things in the process.

That’s a much harder test.

Google is also investing in systems that allow multiple AI agents to work together. Its /teamwork research points toward a future where specialized agents cooperate on complicated problems rather than relying on one model to do everything.

That sounds promising.

But there is a difference between a controlled demonstration and something I would trust with a production environment.

For me, reliability remains the real benchmark.

OpenAI wants agents to handle your money

One of the more interesting developments came from OpenAI’s announcement of ChatGPT Wallet.

The idea is straightforward: an AI agent can help complete a purchase, while the user remains responsible for approving the payment.

This might sound like a small feature.

I don’t think it is.

Until recently, most consumer AI interactions followed a familiar pattern.

You asked a question. The AI answered. You decided what to do next.

Agent-based systems change that relationship.

Imagine asking an assistant to find replacement parts for your computer, compare prices, select a suitable option, and prepare the purchase.

Instead of opening five browser tabs and manually comparing everything, you could delegate much of the process.

Convenient? Absolutely.

But I immediately started thinking about the risks.

An autonomous agent making an incorrect change to a company database can become a much bigger problem.

The same applies to financial transactions, customer information, internal documents, and software deployment.

This is why I believe security, monitoring, and human approval mechanisms will become essential parts of the AI ecosystem.

Companies need to know what an agent did, what information it accessed, and why it took a particular action.

And they need a reliable way to stop it.

As someone who works with technology, I find this part of the AI transition particularly interesting.

We’re moving from systems that generate information to systems that interact with real environments.

That changes the risk profile considerably.

What I’m watching next

Looking ahead, I have three questions.

First, can Google, OpenAI, and Anthropic make their agents reliable enough for everyday work?

Not just impressive in demonstrations, but genuinely dependable when handling complicated tasks.

Second, will the economics improve?

Better models are useful, but companies eventually need to demonstrate that the value they create justifies the cost of operating them.

And third, how much control will users retain?

I like automation. I build trading bots myself, so I’m certainly not against letting software handle repetitive tasks.

But automation works best when its boundaries are clear.

I want systems that can execute well-defined instructions, explain their decisions, and stop when something unexpected happens.

I don’t want an assistant that confuses independence with permission.

That distinction will become increasingly important.

The bottom line

If I had to summarize this week in one sentence, it would be this:

AI is moving from intelligence as a product to intelligence as an action.

Google is competing on performance and efficiency. OpenAI is exploring how agents can participate in real transactions. Anthropic is preparing to face public-market investors. And the infrastructure supporting all of this continues to expand.

These aren’t isolated developments.

They’re different pieces of the same transition.

For users, that could mean less time spent on repetitive digital work.

For businesses, it could create new opportunities to automate processes and improve productivity.

For investors, it raises a much more complicated question: which companies will capture the economic value, and which will simply carry the costs?

I don’t think we have a clear answer yet.

But I do think the conversation is changing.

A year ago, I was mostly asking how intelligent these models could become.

Today, I’m increasingly asking what we’re going to allow them to do.

And that might be the more important question.

Thomas | financial enthusiast


Sources & further reading

Disclaimer: This article reflects my personal views and interpretation of publicly reported developments as of November 15, 2026. It is provided for informational purposes only and does not constitute investment advice. AI products, availability, and financial information may change.

This is my personal opinion, not investment advice. I am not a licensed financial adviser. Do your own research. Disclosure: I may hold positions in assets mentioned here. Disclaimer

Get Thomas' briefing by email

One short email on days with a new article. Nothing on the other days.

More on AI