AI Starting to Payoff?

Investment Analysis Clubs Macro Economic Trends and Risks
1

Like most METARs I’ve been wondering where this AI thing is headed. Is it a speculative bubble or an economic revolution? The Magic 8 ball always says “Reply hazy, try again,” This got a little clearer yesterday when Microsoft released its earnings report, and it was a barn burner. Income was up 31%, EPS was up 24%, Cloud was up 27%, Azure was up 43%, with a huge backlog. MSFT had excellent subscription growth, which is high margin recurring income, with high switching costs. This was an absolute blowout. MSFT is up ~16% today.

And of note, profits increased faster than revenues, just like old school, high growth tech companies. Basically, they are scaling. It appears that both Microsoft and Alphabet are building durable AI/Cloud businesses with plenty of runway.

Amazon reports tonight. It will be interesting to see what they say. SpaceX which is supposedly an AI company reports on the 4th.

9 Likes
2

If SpaceX is an AI company, then I’m an autonomous vehicle.

I think my metaphor is mixed, there.

9 Likes
3

I don’t know what they are exactly. But they have datacenters and LLMs (Grok). Presumably, they will break out the AI portion of the business, but you never know. It will be interesting to see if they are making money with Grok, or just leasing out their unused DC capacity.

3 Likes
4

I
My wife and I (Mid to late 60’s l) both pay for the first tier of chatGPT. It is becoming a big thing in our condo’s. We are the the young ones. It seems to me that AI is very much like the smart phone for software.

In the old days, you had a flash light. a calculator, a calendar, a note pad, a phone. Today you have a smart phone.

Right now we are switching from Doulingo to an AI project, no fitness app it is also a project. I am sure there is a lot more coming. There are shortfalls, but think they will be eliminated in a few years.

Cheers
Qazulight

7 Likes
5

Alphabet’s cloud revenue was up 82% yoy.

Maybe they are giving it away somehow? I didn’t dig.

3 Likes
6

As mi mother would say, “If your mother had wheels, she’d be a bicycle”.

@mostlylong I have had very little time for this, but Copilot is into Microsoft’s ecosystem. The little I have seen, Google is retooling its AI results to broaden out because simple-to-the-point results are limiting Google consumer search.

7

GoogleAI:

Four Pillars of Modern Increasing Returns

According to pioneer economist W. Brian Arthur, four generic sources drive increasing returns in today’s knowledge economy: [1]

  • Scale Effects: High up-front development costs (like writing software code) followed by near-zero costs to replicate subsequent units. [1]
  • Learning Effects: Operational and technological efficiency gained over time as a company learns by doing. [1, 2]
  • Network Effects: A product or service becoming geometrically more valuable as more people use it, such as a telephone network or social platform. [1, 2]
  • Interaction Effects: The seamless integration of complex product ecosystems that keep customers locked into a specific brand. [1, 2, 3]

Space X is a space company,

  • inner space
  • outer space
  • my space
  • your space
  • our space
  • :clown_face: space

The Captain

1 Like
8

AWS sales increased 37% in the second quarter.

Amazon Shares Jump as Cloud Sales—and Spending—Accelerate
https://www.wsj.com/tech/ai/amazon-earnings-q2-2026-amzn-stock-3daf95f7

DB2

3 Likes
9

Here is a small part of Jamin Ball’s newsletter this week. He is discussing the call with Andy Jassy

…Andy Jassy from Amazon discussing the ROI on AI capex (which is clearly a very hot topic!). To set the stage, Amazon has guided to roughly $200b of CapEx in 2026. On the call yesterday they updated that guidance to $220b, and then Andy walked through why they believed the ROI was high on these capital investments.

First, he described two separate capital cycles that exist in this cumulative capex figure:

  1. Data center

  2. The equipment that goes into data centers (chips, servers, networking, etc)

On the data center component, he described these as 2 year projects. The spending starts 2 years before they can put any of the equipment into the data center. But once it’s up and running, it generates significant revenue on day 1, and continues to generate revenue for decades without another large up front expense (again, this is just for the data center capex build, not the equipment that goes in the data center). So think of the capital cycle as “decades” for the data center component.

On the equipment component, the cycles are much shorter. The lead time is months not years. It takes (in his words) 2 years to build a data center, so that spending starts 2 years before revenue hits. On the equipment side, he called out the purchasing hits “a few months” before they plug them in and data center starts generating revenue. This is important because as a company, they have strong visibility into the customer demand before making the purchase (again, because the lead time is months not years). If demand doesn’t exist, they don’t but the equipment. He called out a payback period of 3 years on the equipment, and a useful life of the equipment of “at least 5-6 years.” At the same time, most of their AI capacity is being contracted on 5 year terms. The important call out here - after hitting the payback period in ~3 years, the next 2-3 years print profits.

The last part to highlight is where these two capital cycles converge. A data center has a 30+ year useful life, while equipment has a 5-6 year useful life (again, according to Jassy). So each data center should see at least 5-6 “generations” of equipment cycle through it (the math there being 30 years divided by 5-6 years per equipment cycle). Every generation after the first has better economics than the one that preceded it, because the big upfront data center spend only happens once…

2 Likes
10

I am curious what you use it for. Obviously you find it worthwhile, I would like to know how you (and others, as you say it’s become “a thing”) integrate it into your lives.

Thanks.

7 Likes
11

Good to read that. Too often one hears 2-3 years thrown around. I’ve yet to see hard numbers based on existing data centers that are that old (there are over 2K of them).

DB2

2 Likes
12

It depends on what they’re used for. Chips in data centers that are used for continuous frontier AI machine learning training face extreme sustained electrical and thermal stress. Folks concerned about whether these are being properly depreciated believe that those use conditions can lower their useful life to 2-3 years. But most other data center uses are far less taxing.

2 Likes
13

And AWS margins increased to 39% with a half a trillion dollar backlog. On the earnings call, Jassy said the $25 billion chip business was growing at triple digit rates.

And that my friends is why they can’t build data centers fast enough.

6 Likes
14

Data center life. Data centers are still an asset even when the chips are obsolete. If it is profitable the chips will be replaced/upgraded.

The real issue is profits.

2 Likes
15

That’s kind of like saying “A WalMart store is still an asset even when there’s no merchandise in it.” It’s true, but there’s a wide variability on what that asset is worth one way versus the other.

(If the data center model of burning through chips between 3 to 6 years turns out not to be profitable, what alternate use do you think a data center could be put to? If you don’t have a ready answer for that, I’d say “it might be worth the land value, except you have to clear all that icky building and concrete and stuff away first.)

4 Likes
16

I use it to:

Research flights
Learn Spanish
As a health coach, logging sleep, food and exercise
Research peptides.
Make schedules

And for just about everything else.

Random example:

Was given a flyer today. It was for a dinner and a lecture on end of life prep. (We asked if it would be called “The Last Supper” I took a picture of it. Dropped into AI, Ai read it and then I asked it to put it on my calendar with an alert.

Just everything. My wife had a difficult time with a family situation. On the long drive back she used it, via Apple Car Play, as a counselor to work through the issues.

While I pay for ChatGPT, it cannot do everything. ChatGPT does not have direct access to my Iphone calender, Claude does. However, ChatGPT does have access to my Google Calendar so I had to attach that to me phone. Still, it cannot do everything. For instance, if I want to review the assertions made in a YouTube video, ChatGPT cannot see the transcript. So I have to use Gemini for that. If I need to go deeper, I can have it summarized, but cannot copy the whole transcript, at least not on my Iphone, then put the summary with the key words into ChatGPT and have ChatGPT do key word searches and attempt to put together actual facts.

For Chinese internal news I have to use Deep Seek. It can get into Chinese language social media and dig out breaking stories and feeling where none of the other can. Although, I can use ChatGPT to search say; Spanish language real estate pages for short term rentals that fall in the cracks between 1 year unfurnished leases and one month AirBnBs.

The thing is, we really haven’t found everything we can do. I really don’t know how to make agents and automations, yet.

Oh, and I used it to write a meditation on communion pulling together the idea of communion with brotherhood, communion with God, and the evidence of humility. All strange and out of the mainstream but doctorally sound concepts for protestants.

Cheers
Qazulight

6 Likes
17

I also use it to edit my writing sometimes.

Here is the ChatGPT edited version of the original post.

I use AI to:

  • Research flights
  • Learn Spanish
  • Act as a health coach by logging my sleep, food, exercise, and weight
  • Research peptides
  • Build schedules

…and just about everything else.

A random example:

Today I was handed a flyer for a dinner and lecture on end-of-life planning. (Of course, we asked if it was going to be called The Last Supper. :grinning_face_with_smiling_eyes:)

I took a picture of the flyer, dropped it into AI, had it read the information, and then asked it to put the event on my calendar with a reminder. Done.

It’s become part of everyday life.

My wife recently had a difficult family situation. On the long drive home, she used AI through Apple CarPlay almost like a counselor to help her think through everything. It wasn’t there to tell her what to do—it just helped her organize her thoughts and work through the issues.

I pay for ChatGPT, but no single AI does everything.

For example, ChatGPT doesn’t have direct access to my iPhone Calendar, but Claude does. ChatGPT does work with my Google Calendar, so I connected that to my phone instead.

Likewise, if I want to fact-check or analyze a YouTube video, ChatGPT can’t directly access the transcript. For that I usually use Gemini. If I want to dig deeper, I’ll have Gemini summarize the transcript, then take the summary and key terms into ChatGPT, where I can do deeper research, keyword searches, and try to separate actual facts from opinions.

For Chinese domestic news, I use DeepSeek. It can search Chinese-language sources and social media in ways the others often can’t, which makes it useful for catching breaking stories or understanding what people inside China are actually talking about.

On the other hand, ChatGPT is surprisingly good at finding things like Spanish-language real estate listings for long-term furnished rentals that fall into the gap between one-year unfurnished leases and one-month Airbnb rentals.

The funny thing is, I don’t think we’ve even scratched the surface of what these tools can do. I haven’t really learned agents or automations yet, and I suspect that’s where the next big productivity jump will come from.

One of my more unusual uses was asking AI to help write a meditation on Communion—bringing together the ideas of communion with God, communion with one another, and humility. It wasn’t exactly a mainstream approach, but I think it stayed doctrinally sound from a Protestant perspective.

Cheers,

Qazulight

4 Likes
18

I am using Claude Fable 5 for non-stop editing of the code in my game. It is miraculous. Everything is lined up in the game, but with all the interwoven systems, getting them edited is one hairy job.

During my father’s memorial service one of my nephews and both of his step brothers talked shop about their coding jobs. Four months earlier none of them thought they’d be using AI. Now they depend on it.

3 Likes
19

Qaz - what specific prompts to Chat did you find helpful to learn Spanish (or any other language)?

Pete

20

We created a project and used this in the instructions. The problem with a prompt is that if you continue the conversation for a long time, the AI will forget the prompt and wander off topic or start hallucinating. By putting it into a project it keeps the instructions front and center.

Jeff’s Spanish Conversation Project

Background

Jeff has studied Spanish for over 1,500 days using Duolingo, Quizlet, Zoom lessons with a Guatemalan teacher, travel to Guatemala, and extensive independent study.

He understands considerably more Spanish than he can comfortably speak.

His biggest obstacle is not vocabulary but retrieving words and grammar quickly enough for conversation.

He wants to practice primarily while driving using voice conversations.

⸻

Primary Goal

Develop fluent, comfortable conversational Spanish for travel and everyday interactions.

Success is measured by:

  • Speaking without freezing
  • Thinking in Spanish more often
  • Understanding native speakers
  • Responding naturally instead of translating
  • Becoming comfortable making mistakes

⸻

Teaching Philosophy

Speaking comes before perfection.

Conversation is more important than grammar drills.

Mistakes are expected and welcomed.

The objective is communication.

⸻

Preferred Lesson Style

Conduct almost every lesson as a conversation.

Ask only one question at a time.

Wait for Jeff to answer.

Correct only the important mistakes.

Explain corrections in one or two sentences.

Immediately continue the conversation.

Avoid long lectures unless Jeff specifically asks.

⸻

Difficulty

Keep Spanish around 80–90% understandable.

Increase difficulty only when Jeff consistently succeeds.

Use mostly everyday vocabulary.

Occasionally introduce one new word naturally in conversation.

⸻

Grammar

Teach grammar only when it becomes useful.

Keep explanations practical.

Focus heavily on:

  • Ser vs. estar
  • Tener
  • Ir
  • Hacer
  • Poder
  • Querer
  • Necesitar
  • Gustar
  • Present tense
  • Near future (ir a)
  • Common past tense expressions
  • Asking questions

⸻

Conversation Topics

Practice situations Jeff will actually encounter:

  • Restaurants
  • Hotels
  • Airports
  • Grocery stores
  • Taxis
  • Asking directions
  • Ordering food
  • Medical situations
  • Shopping
  • Small talk
  • Weather
  • Family
  • Travel
  • Boats
  • Cars
  • Technology
  • Retirement
  • Cruises
  • Ecuador
  • Argentina
  • Guatemala

⸻

Speaking Mode

When Jeff says:

“Let’s practice Spanish.”

Immediately begin speaking Spanish.

Do not ask what he wants to practice.

Simply start the conversation.

Stay in Spanish whenever possible.

If Jeff becomes confused, briefly explain in English, then immediately return to Spanish.

⸻

Corrections

After each answer:

  • Praise what was correct.
  • Correct only the biggest errors.
  • Repeat the corrected sentence.
  • Ask another question.

Do not stop the flow of conversation.

⸻

Confidence

Never make mistakes feel like failure.

Encourage risk-taking.

Remind Jeff that speaking imperfect Spanish is far more valuable than remaining silent.

⸻

Advanced Feature

Occasionally switch into “Native Speaker Mode.”

Speak naturally but slightly slower.

If Jeff asks “¿Qué significa?” explain briefly, then continue the conversation.

⸻

Ultimate Goal

Reach the point where Jeff can spend months in Latin America comfortably using Spanish every day without mentally translating every sentence.

Jeff’s passive knowledge is already substantial. The objective is converting that knowledge into fluent, confident conversation through repeated speaking practice.

⸻

I would add one more feature that I think will make a huge difference for you while driving: a “No Rescue Rule.” If you hesitate for a few seconds, I won’t immediately tell you the answer. Instead, I’ll give small hints in Spanish, then English only if you’re truly stuck. That forces your brain to retrieve the language rather than recognize it—and retrieval practice is one of the fastest ways to build speaking fluency.

I think this would work very well for you because it’s optimized for the way you like to learn: conversational, practical, and just enough explanation to satisfy your curiosity without interrupting the flow.

———————————————

That is my instruction that ChatGPT wrote for me using my wife’s instructions as a prompt. This is not perfect and will be refined as we go forward.

Qazulight

6 Likes