If you are following the A.I. story, peeking through the hype is the realization A.I. is not paying off – is not close to paying off – 95% of the projects are a loss.

MIT says 95%:

https://www.axios.com/2025/08/21/ai-wall-street-big-tech

Cost constraints are choking A.I. development.

Some wags say build more data centers and eventually costs will fall.

Those would be the guys from the data center construction lobby.

Here’s an interesting idea – what if you eliminate the costs of a new data center – or scores of them, and if you eliminate the costs of thousands of megawatts of power – and hey, if you eliminate the litigation from angry citizens withpolluted drinking water?

Maybe without those costs, the A.I. thing might just pay off 20 years, or a half-century earlier?

That was exactly the result Government intel officials – who were thinking about a nuclear war possibility – needing massive compute – without a data center – realized.

The insight to eliminating the need for a data center was taking the path less traveled – low I/O optimization or better said – reduce I/O and if you do, everything down stream changes.

You can read lots of our stuff at TheSustainableComputingInitiative.com about how customers, and soon the U.S. war fighting community are proving every day there is no need for a new data center for any A.I. or other application.

Those of you who challenge us – like the guys from the telco who visited Texas last month – watched us run scores of applications that engulfed an $80 million Oracle Cluster – taking days to run – and we ran all of them on rack plugged into a wall outlet – in triplicate – in minutes – using less power than a Tesla.

So those of you who dig this kind of stuff are seeing it and we have lots of fun introducing you to industry disruption.

In this post, let’s not spend time on the Fractal Equation – we will simply restate it:

I/O optimization – pioneered by the U.S. Government intel agencies – delivers the following equation:

1. 95% of what a computer – and data center – does is I/O
2. If you reduce or optimize I/O, every application can run 1,000 to a million times faster.
3. Thus, you get QUANTUM SPEED – TODAY – ON CURRENT HARDWARE
4. Thus, every application needs 1/1,000th the hardware
5. Thus, that application needs 1/1,000th the power
6. Thus, a current data center can do the work of 1,000 new ones
7. Thus, YOU DO NOT NEED MORE DATA CENTERS!

If you are new to the team, we have all kinds of articles on how low or optimized I/O brings these benefits today. That’s your homework.

Now let’s speculate on a different possibility – delivering the jackpot of A.I. everywhere – almost for free!

We know, it’s not free, because you have to pay some guys to write code, and all the other stuff, but let’s say without a data center and without additional energy – it’s sorta free – at least comparatively.

It’s free of the major published constraints – let’s go with that.

Now that we’ve covered constraints, let’s examine the possibilities of A.I. everywhere – because the overwhelming constraints – data centers and power – and miffed Virginians litigating – are off the ledger.

Does cost limit the use of any tool?

Of course it does, you learned that in Econ 201 on Price Theory.

But, what happens when the price constraint is disrupted?

We did not say lowered, we said disrupted.

Remember when a cell phone call had a cost line item attached to it?

If you are under 40 you may have never seen such a bill.

Today you get a number – like $250 a month – and make all the calls you want. That’s why so many people miss the green traffic light changing – they are on their phone, texting.

Anyway, when the cost constraint was lifted, everything went to the phone – and if you are under 30, if you lose your phone you may perish from starvation – it has your money, photos, contacts, Zoom log on, email – it’s your life.

When a disruptive technology – a mobile phone – has few cost constraints – it becomes omnipresent – everywhere, all the time.

Everything goes there – as marginal costs approach zero.

Back to A.I.

Low I/O eliminates – not reduces – eliminates the need for more data centers and more power – period.

Challenge us.

The last two who did that now bid Fractal deals to the U.S. Government.

On our team – because we live in a world where data centers and power are not a cost – or a constraint – we see A.I. everywhere.

Of course we do – low I/O and I/O optimization allow marginal costs for A.I. applications to approach ZERO.

One of our first surprises was how “NO DADT CENTER” impacts battlefields – real ones with guns and bullets – battlefield applications.

It seems making 100,000 drones operate as if they were a flock of birds has some real world applications – as we are now deep into demonstrating.

While we speculate on how A.I. will develop as more people investigate the low I/O cost constraint disappearance – we are finding a hidden little gem making A.I. even more lucrative.

Development times can be measured in days and weeks – maximum of 90 days – and the impact of being able to deliver A.I. driven applications can be weeks not years.

Let’s take a short side trip.

To get to low I/O and I/O optimization, the engineers had to make applications small.

These applications are tiny compared with conventional applications – and to achieve that, much of the “wisdom” of how certain entities – a drone, electric meter or a phone bill – operate is kept outside the code.

The engineers explain this on the corporate site: Fractal-Computing.com

What that means to the programmer is large, million line applications with conventional technology are often replaced with a Fractal application with 2 – 3 pages of code.

Converting a complex billing system – which should take 18 months to 3 years, cost $25 million, inflicting massive internal and customer disruption – can be done in 90 days, no code reviews, no risk because it’s done in a Digital Twin parallel system – and there is no need for a data center. And, nothing close to $25 million bucks.

The corporate billing system may run for clock-days, on a mainframe – the Fractal equivalent runs for 10 minutes on a desktop, or 3 of them if you want to do triplicate.

The reason we bring this to your attention is because every call we get is about eliminating data centers and energy costs.

Those are great objectives – but we want to introduce you to the third constraint – data centers, energy, and DEVELOPMENT time.

We think the third constraint – cutting Dev/Ops to 90 days or days and weeks is the real hidden payoff – and can help deliver A.I. that hits the jackpot – for everyone.