Pre-Fed Bounce? Oil, Housing, and how AI’s at the “IBM Moment”

We kick off Fed Week with the CFNAI – Chicago Fed National Activity Indicator – just out: adjust the rearview accordingly.

“The Chicago Fed National Activity Index (CFNAI) increased to +0.07 in September from –0.08 in August. Three of the four broad categories of indicators used to construct the index increased from August, and two categories made positive contributions in September. The index’s three-month moving average, CFNAI-MA3, increased to +0.03 in September from +0.01 in August.

The CFNAI Diffusion Index, which is also a three-month moving average, increased to +0.06 in September from +0.05 in August. Forty-five of the 85 individual indicators made positive contributions to the CFNAI in September, while 40 made negative contributions. Forty-one indicators improved from August to September, while 43 indicators deteriorated and one was unchanged. Of the indicators that improved, 12 made negative contributions.”

Tomorrow, because it’s the last Tuesday of the month, the Case-Shiller/S&P housing number will come.  And these will matter as the ‘next to last’ economic puzzle piece before Fed rates Wednesday.

Earlier this morning, the CME Group Fed Watch tool was showing a 90.6 percent chance of “no change” from the Fed. But that’s not where the excitement is.  Neither is the overhyped Brazilian election result – there’s more to come.  No, the real problem with the Fed rate is the chairman’s talk and International Trade which will roll out Wednesday. That sets tone. Tone triggers optionality.

The Optional Future Still Matters

Our Over-the-Horizon (OTH) predictive work has been drifting toward a late-October/November instability window, with particular attention to the period surrounding the elections and the November 27 half-session. Our consigliere is approaching much the same period from geopolitical, historical-cycle, and financial directions.

Suppose the following occur within a few weeks:

• Renewed Middle East escalation produces another energy-price shock.

• AI infrastructure financing begins to reprice as investors reassess long-term commitments.

• European political instability damages confidence and increases sovereign-risk premiums.

• Equity markets are already technically vulnerable, with liquidity thinning and investors reducing exposure.

Individually, markets might absorb each development. Together, they could create feedback loops.

A rise in energy prices raises inflation expectations. That limits the Federal Reserve’s freedom to respond to falling equities. Falling equities tighten financial conditions. Tightening credit makes speculative AI investments harder to finance. Financing difficulties trigger further equity selling.

Yes, a fine time will be had by all.  Besides, the smart money on nuke use holds until New Year’s but before the next Congress and that load of “change” sit down.  The 120th Congress gavels-in on January 3rd. Might happen earlier, though.  BYOPA (Bring your own pickaxe.)

Bloody Thanksgiving?

We will stop shy of “predicting” anything. But there’s a pile of poker chips on that weekend.

The reason? Consider what’s already on tap for the future.  I mean besides the Economic Fractalist’s work and our own Peoplenomics research providing for prompt (and seriously gravity-impaired) declining markets in November.

Bombs, bombs, who’s got the bombs?

Israel holds its national parliamentary election tomorrow, with Prime Minister Benjamin Netanyahu facing a challenge from an opposition bloc led by former military chief Gadi Eisenkot. Voters will elect the 120-member Knesset, but an electoral defeat would not immediately remove Netanyahu from office. Just like Trump’s party gets a window before the next Congress, Netanyahu gets some slack, too. Timing is everything.

Official Israeli results are due by November 4, after which President Isaac Herzog has seven days to designate someone to form a government. That candidate receives 28 days, potentially extended by 14, to assemble a coalition. If an opposition coalition succeeds, Netanyahu would leave office when the new government is sworn in—potentially in November or December, not on election night. If negotiations stall, he could remain caretaker prime minister considerably longer, even into 2027.

The OTH implication: Israel’s October 27 election and the November 27 market-risk window overlap with a potentially prolonged political transition, during which Netanyahu could retain executive authority despite losing his governing majority.

Intergenerational Concern: A fellow of my acquaintance – retired from the shiny oaks – mentioned the big, bigger, biggest context in an email overnight. I don’t think he’d mind my sharing this point:

“To a somewhat informed observer, it sure appears as if we are living in something approaching Biblical end times. “Mark of the Beast” technological capabilities (for the first time ever for humans), alignment with 2 Daniel prophecy, the apparent on-going great falling away from the faith, wars and rumors of wars, the growing possibility for something like a global famine (led by the increasing cost of diesel and decreasing availability of inorganic fertilizer material stocks, including nat gas based nitrogen, and decreasing potash, and lack of sulphur, which is critical in the production of other minerals, which mostly had been coming out of the Middle East as sulphuric acid, a bi-product of desulphurization of sour crude oils.) All the elements for creating the full spectrum of prophesied Biblical “end times” markers appear to be emerging globally. In terms of cycles, that would seem to be the big’un. If so, things are going to get a lot worse before they get better. “

We see a tremendous opportunity for failing me-too media to reinvent itself here.

How about a nightly game of “Spot the Horsemen” with live stand-ups?

News Compressor: ON

Has a nice equestrian vibe to it, if you don’t mind our “horsing around.” But hay, let’s get on with it.  Give me a riff; maybe a G flat…

The Vital Four: Four Horsemen, Four Directions

Our morning BlinkLab scan screened 58 distinct news clusters. Four rose to the top. And wouldn’t you know it? Three of them are riding suspiciously close to our OTH instability window.

ENERGY: The Gulf’s Backup Plan Has a Target on It
Yemen’s Houthis claim attacks on Saudi Arabia’s East–West oil pipeline pumping stations and Riyadh’s international airport, although Saudi confirmation of pipeline damage remains outstanding. With Hormuz already restricted, the alternative route to the Red Sea is becoming strategically important. Escalating.

EUROPE: The Streets Are Getting Crowded
France saw an estimated 410,000 demonstrators Saturday, with unions planning more action while political unrest continues elsewhere in Europe. Different grievances, different governments, but a common problem: the cost of living is outrunning political patience. Escalating unevenly.

AI MONEY: The Electric Bill Comes Due
Major technology earnings and Wednesday’s Fed decision arrive as Australian data-center operator NEXTDC reports electricity-network delays that have deferred contracted revenue. Turns out AI doesn’t run on investor presentations; it needs megawatts, transmission lines, and customers willing to pay enough to service the debt. Financial sensitivity increasing.

AFRICA: The Other Kind of Contagion
Uganda is tightening border controls as an Ebola outbreak in the Democratic Republic of Congo raises regional transmission concerns. It’s not a global pandemic, but it’s another example of governments having to allocate resources to problems that don’t respect borders. Containment pressure increasing.

Notice something? Energy, political legitimacy, capital availability, and biological disruption.

Different horsemen. Same increasingly crowded road. All of our “learn to grow your own food talk?”  We expect that horseman in 2027 under a “fertilizer and fallout” standard.

Inside Pages: The Other Hoofbeats

Here’s what else the news jockeys have lined up for the week’s start:

• Ukraine: US missile-warning arrangements and Starshield capabilities reportedly restored following diplomatic discussions, while attacks and counterattacks continue. Military pressure continuing.
• Brazil: Lula’s narrow reelection has been challenged by his opponent, setting up another political legitimacy dispute. Result confirmed; political challenge emerging.
• Venezuela: Opposition leader María Corina Machado has reportedly returned from exile, adding pressure to the country’s unresolved political transition. Political pressure increasing.
• Serbia: A student-led political movement has reportedly defeated President Vucic’s governing forces in parliamentary elections. Political transition emerging.
• China: Xi Jinping’s Communist Party plenum is examining governance and economic priorities. Policy consolidation underway.
• Korean Peninsula: North Korea is raising the rhetorical temperature over border incidents and military deployments. Tensions elevated.
• Algeria/Niger: Algeria’s parliament approved a military deployment to Niger, extending its regional security involvement. Regional involvement expanding.
• Nigeria: Amnesty International reports at least 40 people killed by gunmen in Taraba state. Security deteriorating locally.
• Mexico: Tropical Storm Vance has weakened, but dangerous rainfall and flooding remain concerns. Wind threat easing; flooding risk persists.

America: Blood and Trouble

Ah, the peaceful home front. (NOT!!!)

Tallahassee, Florida: Eleven people were shot near Florida A&M University Saturday night, with several seriously wounded. Investigators are still sorting out what happened.

Newport, Oregon: A major seafood-processing plant fire triggered evacuations because of industrial ammonia concerns. The overnight update was more encouraging: tanks reportedly stable and no injuries reported at that stage.

Four Corners: Flooding across parts of Arizona, Colorado, New Mexico, and Utah has damaged infrastructure and forced evacuations. Wastewater-system problems on the Navajo Nation add another headache.

And wildfire season hasn’t exactly gone home. The National Interagency Fire Center’s October 23 figures showed roughly 9 million acres burned nationally this year, substantially above the ten-year average.

The Next 96 Hours: Where to Watch

The calendar gets interesting in a hurry this week.

But you know what?  Screw it.  What is it Bill’s Friends say? “Cinch by the inch, hard by the yard?”

AI Titans might want to think long and hard about that.  Now I’ll tell you why.

Around the Ranch: The AI Risk No One Talks About

There’s a peculiar thing about technological revolutions. The people making the most money from the present version of a technology are often the last to recognize what will replace it.

Which brings us to artificial intelligence, the hundreds of billions being poured into data centers, and a little research project underway here at the ranch.

But first, a trip back in time.

When IBM Was the Whole Universe

When I was in school, computers were something you connected to. Not something you owned.

My introduction involved FORTRAN, a dial-up connection, and a mainframe somewhere else. The computer was an institutional resource. You submitted your work, waited for the machine, and hoped your program didn’t contain some ridiculous little error that would send you back for another round.

Computing was expensive. Centralized. Controlled by people who understood the machinery, and rented to those of us who needed answers.

IBM was the great symbol of that world.

And for a while, the industry’s future seemed obvious: bigger computers, more processing power, larger installations, and more people connected to increasingly capable central machines.

Sound familiar?

Then something happened.

Microprocessors improved. Memory got cheaper. Software escaped the priesthood. Personal computers arrived, and the economics of computing changed.

IBM didn’t disappear. Mainframes didn’t disappear, either. But the idea that meaningful computing had to happen somewhere else, on somebody else’s enormous machine, was permanently broken.

The computer moved from the institution to the desk. Then to the briefcase. Then to the pocket.

Today, a cheap little computer can perform work that would have seemed miraculous to the mainframe operators of my youth.

The important point isn’t that mainframes were bad.

It’s that we mistook the prevailing delivery system for the permanent architecture of computing.

And I wonder whether we’re doing it again.

The AI Risk Nobody Wants to Price

Look at today’s AI business.

The dominant commercial approach involves building extraordinarily large models, training them on immense quantities of information, and operating them using increasingly expensive computational infrastructure.

More chips. More electricity. More cooling. More transmission lines. More data centers.

And, naturally, more financing.

The assumption underlying much of this investment is that tomorrow’s useful intelligence will require substantially more of the same.

Perhaps it will.

But there’s another possibility that doesn’t receive nearly enough attention from the people financing the boom.

What if useful intelligence becomes much cheaper because we discover that we’ve been organizing it inefficiently?

Not merely through better chips, smaller numerical representations, or cleverer compression.

What if the architecture itself changes?

Imagine financing a nationwide network of enormous centralized computing facilities, only to discover that a substantial portion of the useful work can be performed by much smaller systems operating locally.

The big installations would still have jobs. Plenty of them.

But their pricing power, utilization assumptions, and expected returns could change dramatically.

That is the risk.

Not that AI fails.

That AI succeeds in a different form than the investors have financed.

The Little Animal in the Shop

Which brings us to something we’ve been working on here. Regular readers know that I’ve been experimenting with local AI systems. Nothing resembling the massive installations being built by the technology giants.

Small computers. Modest models. Limited memory and processing power.

The sort of equipment a reasonably determined fellow can operate without first negotiating a power-purchase agreement with a utility company.

But our latest work has moved beyond simply asking whether a small AI can answer questions.

The more interesting question is whether a small AI can develop a persistent, measurable behavioral identity through a different arrangement of learning, memory, interruption, and choice.

We’ve given the research a name:

VOHA — Volitional Observer Host Architecture.

It’s part of a broader project we call C2S, or Carbon-to-Silicon, informally known as the Silicon Animal.

The central idea is deceptively simple.

What if intelligence isn’t merely a matter of how much information a system contains?

What if some of what we recognize as individual behavior emerges from the continuing relationship between memory, environmental inputs, accumulated consequences, and the choices available to the system?

Consider an animal.

It doesn’t awaken every morning with a freshly downloaded encyclopedia and begin life again.

• It carries forward experiences.
• It remembers things that worked. It avoids things that hurt. It develops habits, preferences, and recognizable patterns of behavior.
• Sometimes it changes its mind.
• Sometimes it makes a mistake.
• Sometimes an unexpected event interrupts everything and forces it to reconsider what it was doing.

Now, here’s an interesting question:

Could a relatively small artificial system develop stable, individually recognizable behavior through a similar process, without requiring an enormous language model to do all the heavy lifting?

That’s what we’re investigating.

The Importance of Being Interrupted

One of the unusual features of our research involves something most computer designers traditionally try to eliminate.

• Noise.
• Randomness.
• Interruptions.

Conventional computing generally prizes predictable execution. Give a machine an instruction and expect it to follow that instruction.

But living systems don’t experience the world as a perfectly ordered sequence of instructions.

An animal is constantly interrupted by changing conditions, competing needs, memories, surprises, and sometimes apparently irrelevant stimuli.

• Some interruptions are distractions.
• Others change behavior.
• Occasionally, an interruption opens a path that would never have been considered if the original process had continued undisturbed.

Our experimental architecture deliberately includes randomized interruptions as part of its operation.

Not because randomness magically produces intelligence. It doesn’t.

But because we want to test whether controlled stochastic interruptions, interacting with persistent memory and consequential choices, can produce different learning trajectories and more durable behavioral distinctions.

We’ve already run preliminary experiments with small simulated agents.

Some produced intriguing differences in behavioral patterns under different conditions. Other experiments showed how restarting a system or changing its memory affected those patterns.

Interesting? Absolutely.

Proof of consciousness? Not even remotely. And that’s an important distinction.

We’re not claiming to have created a conscious machine, discovered an artificial soul, or settled the philosophical problem of subjective experience. We’re investigating something much more approachable scientifically:

Oh, we’re also not the only ones looking under this particular rock. Asking the hardest question in the face of a mania.

Can a small artificial agent develop a measurable behavioral identity that persists through time, interruptions, and changing circumstances?

And can that identity be distinguished from ordinary programmed responses?

Those are questions we can actually design experiments around.

Intelligence Versus the Cost of Intelligence

Here’s where this becomes economically interesting.

Suppose you have two ways to accomplish a useful task.

One involves sending the problem to an enormous remote system, consuming network resources, centralized processing capacity, and the services of a large model.

The other involves a much smaller local system that has learned enough about its environment, retained relevant experience, and developed appropriate decision-making behavior to handle the task independently.

The second system might not know everything.

But it might not need to.

A ranch hound doesn’t need to understand quantum mechanics to know when something unfamiliar is approaching the dirt garden.

Its intelligence is specialized, persistent, and continuously informed by its environment. When he’s awake.

That’s not an argument that small AI can replace frontier models across the board. It cannot presently match their general capabilities.

It is an argument that general-purpose computational power and useful situated intelligence are not necessarily the same product.

And the economics of those two products could be radically different.

A small persistent agent that performs a narrow task reliably might be more valuable than a vastly larger model that must repeatedly reconstruct the task’s context.

Especially if the small agent can operate locally, use little electricity, and continue functioning when the Internet goes down.

Out here in East Texas, where backup power is more than an academic subject, that last point has some appeal.

The IBM Moment, Revisited

Now return to the mainframe era.

The mistake wasn’t building mainframes.

They were magnificent engineering achievements, and they remain useful today.

The mistake would have been assuming that the economics and architecture of centralized computing described the entire future.

Personal computing didn’t eliminate large-scale computing. It changed which work needed to be centralized, who could afford computing, and where the value accumulated.

The same could happen with AI.

There may always be a need for immense frontier models doing scientific research, complex analysis, and other demanding work.

But a substantial amount of everyday intelligence could eventually migrate into smaller, persistent, specialized systems.

Instead of renting a giant brain for every question, perhaps we’ll have modest artificial agents that learn their particular jobs and carry forward what matters.

That wouldn’t necessarily destroy the data-center business.

But it could alter the relationship between computing investment and economic return.

And if investors have priced the industry on the assumption that almost every incremental unit of intelligence must pass through expensive centralized infrastructure, an architectural shift could be quite a surprise.

Particularly if it arrives while financing costs are rising.

Which brings us right back to our Over-the-Horizon work and the possibility of an AI investment repricing.

There are two different risks worth distinguishing.

The first is that AI demand fails to justify the capital spending.

The second, and potentially more interesting, is that demand flourishes while the cost of delivering useful intelligence falls faster than the owners of expensive infrastructure expected.

One is a demand problem.

The other is a business-model problem.

And as we’ve discussed around here before, everything is a business model.

Even when the business is intelligence.

The Experiment Is Not the Conclusion

Before somebody mistakes this for an announcement that we’ve invented the next personal computer in the shop, a little perspective.

VOHA/C2S remains experimental.

Our early results concern behavioral patterns in small artificial agents, not demonstrated general intelligence or consciousness.

The next steps involve controlled comparisons, repeated trials, memory and restart experiments, and measurements of how much computing and energy are required to produce useful behavior.

We also need to determine whether apparent identity is genuinely persistent across changing conditions or merely a statistical artifact of the experimental setup.

Those are substantial questions.

And the project is intended to be open source, because independent replication matters more than an impressive announcement.

If the hypothesis is wrong, properly designed experiments should tell us.

If it’s partly right, the interesting work will be figuring out which elements matter and whether they scale beyond toy systems.

If it’s substantially right, we may have identified one more path toward practical, affordable artificial intelligence.

No promises.

Just a research direction we’re pursuing.

Stumbling into a Different Kind of Future

There’s something rather satisfying about investigating these questions on a small computer in a shop surrounded by radios, machine tools, antennas, and assorted unfinished projects.

It reminds me of the early personal-computer era. Back then, the people with enormous computing budgets weren’t the only ones asking interesting questions.

Some of the most consequential work came from people who wondered why computers had to be so big, so expensive, and so far away. Today, the industry is once again building enormous centralized machines.

And once again, some of us are wondering whether there’s another way.

Maybe the future of AI is a trillion-dollar installation.

Maybe it’s a small, persistent system sitting quietly on your desk, learning its environment and getting better at its job.

Most likely, there will be room for both.

But investors might want to think carefully about which one they’re financing, and what happens if the economics shift.

After all, I remember when you needed a telephone line and permission to use somebody else’s mainframe just to run FORTRAN.

Now there’s more computing power in my shop than I could have imagined in those days.

And we’re using a little of it to investigate whether intelligence might be less about building a bigger brain and more about giving a smaller one a meaningful history.

The risk no one talks about isn’t that artificial intelligence becomes too powerful.

It’s that useful intelligence becomes too cheap for today’s investment assumptions.

And somewhere, perhaps, the next IBM moment is already underway.

While carbons dream of AI, we’re working at the other end: silicon’s learning to dream.

Haven’t we been here before?  Dial-up mainframes and “power too cheap to meter?”

Aftergraph: The Price of Thinking

There’s one more possibility worth considering before we leave the AI barn.

When personal computers arrived, they didn’t eliminate mainframes. They changed who could afford computing and what people could do with it. The same could happen with AI. The next breakthrough might not be a machine that knows everything, but a little one that remembers enough, learns from consequences, and knows when to reconsider.

Our VOHA/C2S work is an experiment in that direction, not a declaration of victory. But the economic question is already worth asking: What happens to the price of intelligence when the architecture changes faster than the financing can be paid off?

Somewhere between FORTRAN on a dial-up mainframe and a Silicon Animal in an East Texas shop, there’s a lesson about confusing the cost of machinery with the value of what it produces.

The machinery can get cheaper.

The thinking can get better.

And the people holding the debt may discover those are two very different developments.

“Perchance to dream?” If AIs can do it, our species is not alone.  (I told you it would come to this in one of my books.)

Once again, the yoke’s on whom?

Write when you get rich,

~ Ure (who is supposed to be retired)