A.I. Can't Outrun Current Tech Limitations
You cannot scale A.I. if you need data centers on every corner
Running large A.I. models on the current, conventional tech stack is the technical equivalent of putting a Ferrari engine in a horse and buggy.
Or trying to.
Without the horse, of course.
The current tech infrastructure is 40-something years old.
The current tech infrastructure was built to help companies and agencies run HR systems, pay bills, run a factory accounting system.
The current tech infrastructure was not designed for A.I. - it is not optimal for A.I. and many of the limitations - like needing a data center, limit current A.I. effectiveness and widespread adoption.
The issue facing America, China and other sovereigns today is not how to scale A.I. It is how to scale the crappy 1980s tech infrastructure - to support A.I.
The answer, as you will come to learn here, with live customers implementing Fractal - is you need an A.I. purpose-built tech stack - and we have one.
When you define the problem that way, “what tech stack enables A.I. to scale” - clarity ensues.
If you need data centers on every corner - if Microsoft needs a nuclear power plant to run A.I. models - if the rural, small town bucolic landscape must be plowed into data center construction - you hit a wall.
The infrastructure - by definition - cannot scale to meet the A.I. problem.
That which cannot scale, won’t.
Let’s look at some examples.
Take database to start.
The conventional tech stack - yeah, the same one Gartner said was obsolete in 2025 - that one - is relational database. Whether it’s Oracle, PostgreSQL, Mongo, SQL they are derivative of the relational tech stack.
One of the biggest problems in A.I. today - we know this because corporations and government agencies tell us - is merging multiple databases, each written in different decades, with different indexing (ISAM, VSAM, embedded pointer, network etc.) which have relevant data - but they cannot be easily merged for A.I. to query.
Need proof?
Our favorite government techie - Elon Musk - said D.O.G.E. was hamstrung because it might find Phineas was ineligible for his Social Security checks, but there was no way to also check if Phineas was milking Medicare and 20 other government programs.
So Elon gave Palantir - who uses ancient tech at scale - with lots of help from pals inside the government who will later work for Palantir - Elon gave Palantir a no bid, $30 million contract to “API interface” government systems.
That is a 1980s approach - which sucked in 1981 and will only require another massive data center - with those 5 figure a month electric bills and 400,000 gallons of water - today.
They likely cannot get it done for $30 million.
It will take forever and it will be a very limited solution the moment it ever gets done.
All someone has to do is change one letter in Phineas’ last name, move one digit over, and the entire interface thing collapses.
Welcome to relational technology APIs.
Elon’s guys - like Sam Corcoss and the Levels guys - are doing great things because they and only they have access - unlimited access to the most screwed up tech infrastructure in the world - the federal government.
And their solutions are 1980s interfaces we would be embarrassed to even consider.
Elon is the guy to get us to Mars. He is not the guy to integrate complex databases - as the Fractal team shows about every day.
Fractal solves this problem - and it isn’t some trivial problem - it’s a couple of years work to merge a few large DBs with conventional tech - probably with 1980s API technology.
Fractal jumps this limitation by moving A.I. down to the ETL (that’s a term meaning extract the data, transform it to make all the sets play well together, and load it up for A.I.) layer.
So in Fractal, we can take databases which are in totally different formats, written over different decades by people who never spoke to each other - for very different purposes - and we can in a couple of days - as in a weekend - make them a single integrated database.
No API.
No data center.
No expensive conversion.
No 1,000 Indian outsourcing guys paid $65 an hour to do the work.
One engineer, over a weekend.
Go to our micro site, TheFractalGovernment.com and read about how we took the voter rolls for two dozen states, plus their property tax databases, plus a bunch of other databases, and merged them into a 2.7 billion record database - running without a data center - at silicon speed - accessible from a phone.
One engineer did it over a weekend.
These were two dozen production-level databases - each created by guys who did not know the guys in the neighboring state, over 40 years, some using database technology that isn’t even supported any more.
Then there were property tax records - totally different file structures - across dozens of zip codes - each county calling a single family house something different.
Try that with conventional technology.
Do it in a weekend.
Run it in production - on a computer 4 - inches by 4 - inches.
Do it for 25 states.
No data center!
Before A.I. can take on the big problems, foundational problems must be solved.
If you cannot merge databases, in real time, you hobble the roll out of effective A.I.
Elon and Palantir are into APIs - and they rigged a no-bid government contract for $30 million bucks to API a bunch of government computers. Good luck with that.
Fractal could merge those government database - in a month or less - likely in days - without a data center, and give Trump $25 million back.
We won’t get the chance because a no-bid contract is there for a reason.
If you have to build massive data centers, which take a couple of years, and you need to build a power plant, which takes 5 years or more - you are pretty much screwed on how much A.I. you are going to apply to a problem.
So the Fractal way was to build the foundational stack first - the tech substructure so merging databases and building data centers are not necessary - so when we apply A.I. we are starting on third base.
And we did that.
Another issue in A.I. is scale.
Actually it isn’t AN issue, it is THE singular issue in A.I. today.
Current technology - A.I. or not - cannot scale.
Proof is that the legacy software companies need bigger data centers to run A.I. models.
The problem isn’t A.I., it the I/O intensiveness of the current underlying tech stack - the one Gartner said is obsolete in 2025 - that one.
A.I. relies on that tech stack - so the limitations of the stack set the boundaries of A.I.
Data centers are mausoleums for I/O intensive legacy applications and vendors.
1980s tech vendors could outrun Moore’s Law until chips couldn’t be made much smaller. That game is now over. Thus the data center madness.
You would think before people went nuts with A.I. for everything, they might ask the question - can we scale the heck out of A.I. with our current infrastructure?
Had that question been asked by the big tech vendors, which it surely was, their product managers had a very predictable answer - “if we do that, all our stuff - driving all our revenue and jobs - is obsolete.”
Some guy at Palantir or Oracle probably said that - or will soon.
“So, NO, we have to go A.I. with what we have, not with what the customer needs.”
The legacy answer is to flood the press - like the hapless Wall Street Journal tech writers and bloggers - about how legacy vendors are supporting bigger models - called LLMs. (large language models).
The WSJ and other tech press actually believes the current tech stack can scale - while admitting it needs nuclear power to do so.
That is the mindset of the tech press - that’s why you need to read our stuff - and if you are a qualified prospective customer, try our tech out!
Tech disruption never comes from the past - from old vendors hanging on to licensing models and carrying a book of business - in the bazillions of dollars a year - dependent on their current stuff. Why would it?
If you ever took an economics course - in 101 or 201 they taught you about Joseph Schumpeter. Old Joe pretty much called the ball in 2025 - innovation from newbies will eat the old guys alive.
He says it a bit more eloquently - over 450 pages, but that’s the summary.
According to Schumpeter, those with the market share do not innovate - they hold on for dear life - often because of product management (he did not say product management, we added that), they hold on for dear life until they are swept away.
Remember Kodak?
One of their guys invented a digital camera - no film.
Oops, Kodak is a film company, so they tried to ignore the guy - but he split, went to Sony or another Japanese company and Kodak became toast.
The A.I. story is starting to play out and there are some surprises for all of us - the Fractal guys included. We are blown away with the overnight realization that America needs data centers coast to coast.
We just figured everyone would get the fact that A.I. can be hugely I/O intensive and different underlying tech stacks would emerge. They don’t seem to have, though.
Quantum computing was supposed to be the answer but it isn’t coming any time soon - and even if it did, it is highly limited to very niche applications - like science stuff and crypto - the security type not the currency type.
Fractal runs quantum speed on current hardware so some of those applications will come, or have already started to come, to our tech stack.
As the A.I. madness continues we are watching the preposterous levels of investment in new data centers - we watch from the sidelines, smiling.
Elon Musk and the D.O.G.E. guys will continue to make headlines - as long as nobody else is allowed to compete. They survive on closed off access and no bid contracts.
Those days are numbered.
Those data centers will be about half built when the critical mass hits - and they will be abandoned in droves.
We would short those companies (it’s a stock thing) but right now we are putting all our dough into just getting more A.I. projects out the door without a data center.
We are working with some pretty cool big time companies taking Fractal to their customers - we prefer selling via partners - follow us here for some case histories over the near future.
Follow us as well to watch the collapse of the “data center” madness - over the next 18 months or so. We are pretty confident because our customers make us so.
We are more confident as Wall Street types beg us for inside info on upcoming news releases so they can buy puts on Palantir and Microsoft.
The journey is fun and we thank you for coming along for the ride.
Thanks for your support.
We made a sizable donation to a Texas animal rescue and later in the year we will make another to our local donkey rescue.
FractalComputing Substack is a newsletter about the journey of taking a massively disruptive technology to market. We envision a book about our journey so each post is a way to capture some fun events.
Subscribe at FractalComputing.Substack.com
Fractal Website: Fractal-Computing.com
Fractal Utility Site: TheFractalUtility.com
Fractal Government Site: TheFractalGovernment.com
Portions of our revenue are given to animal rescue charities