Is the AI Boom Unsustainable?

More: How to slash your AI workflow costs and Learn AI prompts

Author

Renjit Philip
September 28, 2025 • Estimated Reading Time: 13 minutes

In partnership with

“Wealth consists in not having great possessions, but in having few wants.”

Epictetus

Date: 28-Sept-2025

Hey AI enthusiast,

And just when you think you’ve seen it all…there’s always One More Thing in AI.

In this edition:

Also: introducing you to our podcast edition for those of you who want to listen to it on the go. Check it out on your favorite platforms:

I hope you enjoy this edition!

Best,

Renjit

PS: If you want to check out how to implement AI in your Insurance broking business and get more revenue with the same number of employees, book time here:

The AI Boom: Unsustainable Without Revenues

A new research note from Deutsche Bank warns the current AI boom is unlikely to be sustainable.

Data centre buildout is propping the US economy according to a DB report


What’s Happening
• AI capex is massive: Goldman Sachs estimates $368B in AI capital expenditure through August 2025.
• It’s propping up the U.S. economy: Deutsche Bank says this investment is large enough to keep the U.S. out of recession. Essentially reminds us of the tremendous infrastructure build ahead of the dot com bubble in the late nineties / early 2000s, when fibre and internet infrastructure was built out.
• But GDP growth is weak: Strip out tech spending, and real GDP growth in 2024–25 is roughly 0%. The economy is hovering near recession. This is potentially due to the impact of softer spending by customers, job losses and the impact of tariff uncertainty on small and medium businesses in the US.

Where the Growth Is Coming From
• Not from AI software output, but from construction and equipment orders, building data centers and securing power supply. What happens when these data centers are built out or even shelved?
• About half of S&P 500 gains this year are driven by tech stocks, highlighting how concentrated the boom is. The real economy stocks are not growing in real terms. One can look at the stock market and falsely assume that all is well with the US economy.

The Revenue Problem
• Bain & Co. projects the sector will need $2T annually by 2030 to fund compute demand. (link to the report: »)
• Even after accounting for efficiency gains, the world is facing an $800B shortfall.

The Outlook
• Deutsche Bank says sustaining GDP momentum would require the tech cycle to accelerate quarter after quarter which is a parabolic path that looks unlikely.
• Goldman Sachs expects productivity gains from AI to lift GDP by 0.4% annually soon, adding up to 1.5% total as adoption spreads.
• This suggests a softer landing: productivity benefits will arrive, but too slowly to justify today’s breakneck capital spending. Personally, I am hoping for this relatively benign outcome, but am prepared to handle a larger shock.

Why AI Isn’t Replacing Radiologists

Almost a decade ago, Geoffrey Hinton predicted radiologists would soon be out of work, replaced by AI image recognition. That hasn’t happened. Radiology is thriving, and the field is still growing.

This isn’t just about radiology, it’s about the wave of naive predictions on how AI will reshape jobs. A year ago, someone even asked whether software engineers would still exist today. Spoiler: they do.

The case of radiology shows why it’s not so simple:

• Benchmarks don’t reflect reality: Test sets of images are far from the complexity of actual cases.
• The role is multifaceted: Radiologists do much more than interpret images. Judgement and convincing fellow doctors is still very much human.
• Deployment frictions: Regulation, insurance, liability (who will you blame?), and institutional inertia slow down adoption. This is a big factor in highly regulated industries like healthcare.
• Jevons paradox: When AI makes radiologists faster, demand for scans increases.

Radiology was never the best candidate for early disruption; it’s high-risk, highly regulated, and too complex to reduce to one task. If you’re looking for near-term job impact, watch areas where tasks are:

• Repetitive
• Independent
• Closed (not requiring deep context)
• Short in duration
• Low-stakes when errors occur
• Already digital and automatable

So what will be the impact of AI on your job?

Difficult to say, but even in the above domains, AI will likely arrive as a tool first. Jobs will refactor, shifting from manual work toward monitoring, oversight, and exception handling.

Best bet is to learn AI and see how you can make yourselves more productive and valuable for your employer and your customers.

📖 Read the full piece in Works in Progress:
Why AI Isn’t Replacing Radiologists

Learn Prompting

This one is a gem from Hasan Toor on X, especially if you want to dive deeper into your own behaviors and to learn how to address it in a positive way. 

Act like a Nobel-winning psychologist. Analyze my behavior pattern: {describe it}. What’s likely causing it? What mindset shifts, habits, or actions would help me overcome it?

I tried it on {Procrastination} and I got these helpful insights from ChatGPT5:

Analysis of Your Procrastination

As a psychologist would frame it, procrastination is rarely about “laziness.” It’s usually a coping strategy. When you delay tasks, it often reflects an underlying conflict between intention (what you know you should do) and emotion (how the task makes you feel).

The likely causes in your case could fall into these clusters:

  1. Cognitive Causes

    • You overestimate the difficulty or time required, so your brain signals avoidance.

    • Perfectionism: you delay because the task must be “just right.”

    • Choice overload: too many possible priorities leads to decision paralysis.

  2. Emotional Causes

    • Anxiety about failure or outcomes, so avoidance becomes emotional relief.

    • Boredom: if the reward feels distant or the task is uninteresting, motivation collapses.

    • Self-criticism: negative inner talk makes starting painful.

  3. Behavioral Causes

    • Reinforcement loop: every time you procrastinate and feel temporary relief, the avoidance is rewarded.

    • Environment: digital distractions or lack of structured cues make it easy to drift.

Mindset Shifts That Help

  • Shift from outcome to process: Instead of “I need to finish this report,” reframe as “I’ll spend 10 minutes starting this report.” Small wins create momentum.

  • Adopt self-compassion over self-criticism: Research shows self-kindness predicts persistence far better than guilt.

  • Redefine success: Perfection isn’t the bar—progress is. Treat each draft or attempt as data, not judgment.

  • See time as a series of choices: Ask, “What’s the next best use of the next 15 minutes?” This shrinks overwhelming tasks into actionable slots.

Habits and Actions to Counter Procrastination

  1. Behavioral Design

    • Break tasks into “micro-steps” that take <5 minutes.

    • Use “temptation bundling”: pair the unpleasant task with something enjoyable (e.g., doing email only with your favorite music).

    • Set visible cues: keep only the work you intend to do in sight.

  2. Time Structuring

    • Try the “just start for 5 minutes” rule. Most people continue well beyond once momentum begins.

    • Use time-boxing (calendar blocks) instead of to-do lists. Externalizes commitment.

    • Apply the implementation intention: instead of “I’ll work on slides,” decide “At 10:30, in the study, I’ll open the deck and add 3 slides.”

  3. Accountability & Feedback

    • External accountability (a colleague, friend, or even a public deadline) creates healthy pressure.

    • Immediate feedback loops—small rewards for progress, not just completion.

  4. Emotional Regulation

    • Practice mindfulness or breathing resets before starting. Often the hardest moment is the first 90 seconds.

    • Rehearse self-talk: swap “I don’t feel like it” with “I’ll just begin for a bit.”

👉 The core insight: procrastination is a mismatch between what matters long-term and what feels tolerable short-term. The fix is less about willpower, more about designing conditions where starting feels easy, rewarding, and low-risk.

Learn AI: How to slash your AI workflow costs

This is a banger of a thread by @aivanlogic on X. I have extracted the key elements from it. Essentially useful for anyone using AI to improve business process workflows (something that we do in our consulting firm at www.futureu.co).

N8n, the open-source automation powerhouse launched in 2019, has been a lifesaver for businesses- potentially saving millions in operational efficiencies, according to industry estimates. But as Ivan discovered, unoptimized workflows can turn it into a budget black hole. His recent X thread reveals how he reverse-engineered his setup, identifying seven fixable mistakes that slashed costs dramatically.

Key Takeaways: The 7 Optimization Strategies

1. Go Modular: Ditch the all-in-one AI agent. Split tasks into specialized micro-agents using cost-effective models.

Example shared by Ivan: Processing 1,000 emails dropped from $150 to $60: which is a 60% cut!

2. Preprocess Tokens: Clean up data before it hits the AI. Ivan showed how to strip junk, summarize, and route smartly to reduce tokens from 3,500 to 1,200 per call, saving up to 65% per recent analyses. Example is a word document that usually has a lot of header code and formatting. So before sending the full document, take a while to extract only the text and then send that to the LLM for inference.


3. Batch It Up: Group tasks to amortize prompt overhead. Turn 2,000 tokens for 10 items into just 200, which is a simple math with big payoffs, as highlighted in emerging workflow papers. Basically, avoid sending multiple prompts for the same type of work.

4. Dynamic Routing: Rate task complexity with a cheap model, then assign the right AI tier of model for the job. For example, ChatGPT4o mini is cost effective fast and capable for many tasks and you don’t need GPT5 for every task. Shift 70% of work to budget options without losing quality.

5. Enforce JSON: Swap verbose text for structured outputs (in JSON fomat). Ivan showed an example of how he cut a 150-token ramble to a 25-token JSON object, streamlining everything downstream.

6. Track Tokens Relentlessly: Build a system to monitor usage and costs in real-time. Transparency builds trust and prevents bill shocks. When you notice the costs, going up, go back to the workflow steps and see how you can optimize the AI model calls.

7. Engineer Prompts for Cheaper Models: Prototype with premium AIs, then optimize prompts for affordable ones with lower cost models. Move 85% of tasks to 1/10th the cost.

These aren't just hacks- they're a shift toward smarter, scalable AI, especially in the context of workflows built for mid-size companies.

Quick Links & Resources
- Ivan's X Post: View Here
- n8n Official Site: n8n.io
- Dive Deeper: Journal of Artificial Intelligence (2024) on modular AI [Search for it]

AI Images: Create Figures made of Polygon Shapes

Check out the post on X for the prompt and the amazing work done by Ai artists like Amira Zairi.

PROMPT: A low-poly 3D render of a [subject], built from clean triangular facets with flat [color1] and [color2] surfaces. The environment is a stylized digital desert with minimal geometry and ambient occlusion.

Want to reach a quality audience?

Reply

or to participate.