Signal vs Noise: 6 Filters to Invest in the AI Era (2026)

How to separate what is real from what is loud and position accordingly

Quick Answer: Separating signal from noise requires three maps, drawn in this order physical first, financial second, narrative third. Signal lives in physical reality: power grids, minerals, water, infrastructure. Noise lives in consensus belief: headlines, earnings calls, billionaire speeches. The gap between the physical map and the narrative map is always where the exploitable insight is. This article gives you the Three Maps framework and six filters to apply it then shows you the biggest mispriced physical map of the AI era: Africa.

Key Takeaways

• Signal requires physical reality. Noise only requires belief. A cobalt deposit either exists or it doesn’t. A stock thesis only requires consensus.
• The Three Maps framework: Physical map first (resources, infrastructure, geography) → Financial map second (capital flows, what is mispriced) → Narrative map third (press consensus, public opinion). The gap between map one and map three is the signal.
• Six filters: Ask what cannot be faked. Ask who benefits from you believing this. Track velocity mismatches. Look for what nobody is talking about. Separate the instrument from the underlying. Please build the physical map before proceeding with the analysis.
• Aschenbrenner’s trade: He didn’t build AI. He identified energy as the physical constraint every AI company needs regardless of who wins. That is the Rockefeller pipeline move applied to the AI era.
• Africa is the most underpriced physical map of the AI era. The DRC holds approximately 70% of global cobalt reserves (USGS, 2025). East Africa’s Rift Valley is one of the largest geothermal energy sources on earth. Neither is correctly priced by Western financial models, which are still reading the narrative map.

Ken Griffin, the $48 billion Citadel CEO who called AI “garbage” as recently as January 2026, went home one Friday in May fairly depressed. This was not because markets were down. Because he had just watched AI agents complete PhD-level financial research work, his firm typically assigned to master’s and doctorate holders over weeks or months in hours or days.

“These are not mid-tier white-collar jobs,” he said at the Stanford Leadership Forum. “These extraordinarily high-skilled jobs are automating agentic AI.”

Days later, Jeff Bezos told CNBC something that sounded like the polar opposite. AI, he said, will not destroy jobs. It will create a labor shortage. Productivity will soar. Food will become cheaper. Housing will get cheaper. People with two-income households will voluntarily drop out of the workforce.

“If you’ve been digging out a basement for your house with a shovel and somebody’s about to hand you a bulldozer,” Bezos said, “you should be so happy.”

Then there is a third figure, quieter, less covered, and more instructive than either of them. Leopold Aschenbrenner did not build artificial intelligence. He did not write the foundational transformer paper, did not train GPT, and did not design a chip. What he did was read the physical map. Then he positioned himself accordingly, not in the AI companies everyone was watching, but in the energy infrastructure that all of those companies depend on to function at all.

He did not bet on who would win the race. He bet on what every racer would need to run it.

So who is right, Griffin, Bezos, or Aschenbrenner?

That is the wrong question. The right question is: which of these three reads the signal, and which two read the noise?

The Noise Is Loud Right Now

Here is what the visible layer tells you: AI is either the greatest threat to human livelihoods in history or the greatest gift. The debate is everywhere. The takes are brisk, confident, and contradictory.

Here is what the physical layer tells you, if you look.

As of May 2026, tech companies have laid off more than 142,000 workers, roughly 993 people per day. According to Nikkei Asia, nearly half of the cuts in Q1 2026 resulted directly from AI or automation. Meta fired 8,000 people in May, the first installment of what could reach 20% of its global workforce, while simultaneously committing to spend over $100 billion on AI data centers this year alone. Amazon has eliminated roughly 30,000 corporate roles since October 2025, its largest workforce reduction in 31 years.

Meanwhile, Microsoft, Amazon, Google, and Meta combined are on track to spend $725 billion in capital expenditure in 2026, a 75% increase over 2025, almost entirely on AI infrastructure.

People are being fired. Concrete is being poured. Both things are true simultaneously.

The debate about whether AI will “take jobs” is the noise. The physical infrastructure being built to run AI at scale is the signal.

The Three Maps Framework

To understand what is actually happening, draw three maps. In this order only.

The physical map first. Where is the power going? Where are the data centers being built? What minerals does the hardware require? What water is needed to cool it all? The physical map cannot be fabricated. A megawatt either flows or it doesn’t. A cobalt deposit either has minerals in the ground or it doesn’t. This is the only honest map.

The financial map is second. Where is capital actually flowing, rather than where spokespeople claim it is flowing? What is priced in and what isn’t? Griffin’s own firm, Citadel, poured over $4.2 billion into AI and tech positions in Q3 2025 even while Griffin was publicly calling AI “garbage” at Davos. The financial map reads actions, not words.

The narrative map is third. What does the press believe? What is the consensus story? Bezos says abundance. Griffin says depression. Both are narrating from the visible layer. The narrative map is always the least reliable; it is the most recent, most widely shared, and therefore the most thoroughly priced. A narrative that everyone is already reading rarely contains original ideas.

The gap between the narrative map and the physical map is always where the signal lives.

The Third Man: What It Actually Looks Like to Read the Physical Map

Most people who follow AI know Nvidia. Many know OpenAI. Few paid serious attention to Leopold Aschenbrenner, a former OpenAI safety researcher who, in 2024, published a document called “Situational Awareness” that the financial press largely treated as a technology thesis.

It is not a technology thesis. It is a physical constraint thesis.

The argument, stripped to its core: the AI model race is a software competition running on top of a hardware competition, which runs on top of an energy competition, which runs on top of a land, water, and mineral competition. The model that wins is not necessarily the model with the best architecture. It is the model whose infrastructure owners can guarantee the power, the cooling, and the physical presence required to train and run at scale.

Aschenbrenner did not build chips. He did not start an AI company. He moved toward energy, the physical prerequisite that every AI company needs, regardless of which one wins the software race.

This is the Rockefeller trade, repeated. Rockefeller did not own every oil well. He owned the pipelines, the only route through which every oil well’s output could reach the market. The leverage was not in the commodity. It was in the physical constraint on the commodity’s movement.

Aschenbrenner’s version of that trade: the GPU is the visible layer. The data center is the visible layer. The benchmark is the visible layer. The power plant is the physical map. The transmission line is the physical map. The cooling water source is the physical map.

Griffin is watching the output of the PhD work being replaced. Bezos is narrating the outcome of the abundance to come. Aschenbrenner positioned himself in the substrate that makes any of it physically possible.

One of these three things is a trade. The other two are commentaries.

Six Filters for Separating Signal from Noise

Filter 1: Ask What Cannot Be Faked

Noise can be manufactured. Signal cannot.

Every piece of information you receive sits in one of two categories: things that require physical reality to be true and things that only require belief to appear true.

Chip stocks going up requires belief that AI demand will continue, margins hold, and competition doesn’t erode pricing. A substation either exists or it does not. A cobalt deposit either contains minerals in the ground or it does not. Physical reality does not respond to narrative.

Your filter: Does this narrative require physical reality to be true, or only consensus? Consensus-dependent information is noise until physical reality confirms it.

Filter 2: Ask Who Benefits from You Believing This

Jeff Bezos predicts a labor shortage and abundant deflation. Jeff Bezos is also spending $200 billion on AI and has eliminated 30,000 of his workers. Amazon CEO Andy Jassy told staff in June 2025, “We will need fewer people doing some of the jobs that are being done today.”

This is not cynicism. It is a filter. When a sender’s interest aligns with you believing something regardless of whether it is true treat the information as noise until the physical layer confirms it.

Who benefits from this framing? What does the physical resource map look like underneath the stated reason? The gap between the stated reason and the physical map is where the actual signal is.

Filter 3: Track Velocity Mismatches

The signal moves slowly. Noise moves fast.

A tweet that moves a market 3% in an hour is noise. A demographic shift in workforce skills is a signal it moves at one data point per year for two decades before becoming visible as a trend.

The most important AI signal is not the quarterly earnings call. It is the 18–36 month construction cycle for power grids, data centers, and transformer manufacturing. These physical processes don’t respond to Federal Reserve statements. They move at the speed of concrete and copper wire. That slow velocity is precisely what makes them signal.

Aschenbrenner read slow variables: power grid construction timelines, data center permitting cycles, and transformer manufacturing lead times. None appear on earnings calls. All of them determine whether AI at scale is physically possible.

Griffin’s own reversal from “garbage” in January to “depressed” in May is itself a velocity signal. When a committed skeptic of that stature turns in four months, something real shifts in the physical layer.

Filter 4: Look for What Nobody Is Talking About

In 2020, nobody was pricing power grid capacity constraints as an AI limitation. The conversation was entirely about chip architecture and model size. The power constraint was clearly visible in public utility filings, data center permit applications, and transformer manufacturer order books. All public. Zero coverage.

Don’t ask who wins the visible competition. Ask: what is the invisible prerequisite for anyone to compete at all?

Applied to the AI labor debate: nobody is asking what happens to mid-career professionals in the 18-month gap between when their role gets automated and when the new roles Bezos promises actually materialize. That gap, the transition cost no one is modeling, is the signal nobody is covering.

Filter 5: Separate the Instrument from the Underlying

Nvidia is an instrument. AI compute demand is the underlying. They are not the same and do not move together indefinitely.

The “AI will destroy jobs” narrative and the “AI will create abundance” narrative are both instruments. The underlying process is simpler and more physical: capital is being redirected, at extraordinary speed, from human labor costs to AI infrastructure. Griffin and Bezos are describing different time horizons of the same underlying reality, not different realities.

Griffin is describing the present: PhD-level work is being automated now. Bezos is describing a possible future: if productivity gains flow through to the economy, abundance follows. Aschenbrenner is positioned in the gap between those two time horizons in the physical infrastructure the present depends on and the future requires even more of.

When the gap between instrument price and underlying value gets large enough, the instrument corrects. That gap is information.

Filter 6: Build the Physical Map Before You Build the Analysis

Before you panic about AI automation or celebrate AI abundance, map the physical reality first.

US data center spending is estimated to top $500 billion in 2026. Power generation infrastructure is being built at a pace not seen since the electrification of America in the early 20th century. The minerals required cobalt, lithium, and rare earths, are concentrated in geographies that most financial models treat as risk factors rather than prizes.

Which brings us to the part of the physical map that almost no one in the West is drawing correctly.

The Unmapped Layer: Africa and the Physical Substrate of the AI Era

The Democratic Republic of Congo contains approximately 70% of the world’s known cobalt reserves, according to the US Geological Survey (2025). Cobalt is a critical input to the lithium-ion batteries that power the devices running the applications running on the AI models that Griffin watched automate his researchers and that Bezos says will free up the global workforce.

The AI era does not run on mathematics alone. It runs on physical materials, through physical supply chains, on physical infrastructure. And roughly 70% of one of the most critical physical materials is in one country.

The DRC is not a risk factor in an African investment thesis. It is the Saudi Arabia of the AI era, sitting in the substrate below the entire conversation, underpriced in every Western financial model because those models are reading the narrative map (conflict, governance, political risk) while missing the physical map entirely: irreplaceable mineral concentration in a geography the next two decades of technology depend upon.

East Africa compounds this. The Great Rift Valley is one of the most significant geothermal resources on earth. Ethiopia, Kenya, Tanzania, and Rwanda sit on top of sustained, renewable heat sources capable of generating electricity at a cost and carbon profile that makes them directly competitive for exactly the kind of energy-intensive, continuous-load computing that AI infrastructure requires. According to the International Energy Agency, East Africa’s geothermal potential represents one of the most underdeveloped renewable energy opportunities on the planet.

Data centers need cheap, reliable, low-carbon power. East Africa has it beneath its feet.

The market has not connected those two sentences into a thesis because it is reading the narrative map “emerging market,” “frontier market,” and “political risk“ instead of the physical map. The gap between the narrative map and the physical map is precisely as wide and precisely as exploitable as the gap in British telegraph cables in 1870 or Standard Oil pipelines in 1890.

Then there is water. The Nile basin runs through ten countries and provides the primary water source for a significant fraction of the continent. Water is not simply a humanitarian variable. It is a cooling variable. It is a data center variable. No financial model correctly prices the Nile basin’s water dynamics because no financial model draws the physical map first.

Aschenbrenner did not build chips. He positioned the power that chips require to function.

The equivalent African trade is not building apps. It is owning infrastructure that is locally powered, locally controlled, and built on the geothermal and mineral substrate that is already in the ground that the entire AI era runs on.

The cobalt is there either way. The geothermal heat is in the rock either way. The question is whether the gap between the narrative map and the physical map gets priced before or after the rest of the world notices what is sitting in the substrate.

What This Means for You

Whether you are an investor, a professional, or someone who wants to make better decisions in a moment of genuine uncertainty, the framework is the same.

Do not trade on the debate. The Griffin-vs-Bezos argument is noise. Both men have interests. Both are partially right, operating from different time horizons. Trading on their words is trading on the narrative map.

Trade on the physical constraint. What is the bottleneck every AI company will need, regardless of which model wins? Power. Water. Cobalt. Physical infrastructure. These cannot be fabricated by a press release or a Stanford speech.

Track slow variables. The 142,000 tech jobs cut in 2026 are fast-moving, loud, and already priced. The decade-long shift in what physical assets underpin the AI economy is the slow signal worth positioning for. Aschenbrenner found it in energy. The next version of that trade may be sitting in the Rift Valley.

Watch what the money does, not what it says. Griffin called AI garbage and bought $4.2 billion of tech stocks. Bezos predicts a labor shortage and fired 30,000 people. The physical actions are the signal. The words are the narrative map.

The Permanent Lesson

In 1917, Arthur James Balfour wrote 67 words — not a treaty, not legislation, just a letter — and restructured a region for over a century. Not because of what the words said, but because of the physical strategic interest they were applied to. Britain wanted the Suez Canal’s eastern flank. The communication was the vehicle. The physical geography was the prize.

Rockefeller understood this. The story was oil. The signal was the pipeline.

Aschenbrenner understood this. The story is artificial intelligence. The signal is power.

The continent that understands this technology next and positions itself in its own physical substrate before the narrative map catches up will have done what Britain did with telegraph cable, what Rockefeller did with pipelines, and what Aschenbrenner did with energy: read the physical map first, while everyone else was still arguing about the visible layer.

The signal is always in the physical layer. It is never in the headline.

Frequently Asked Questions

What is the difference between signal and noise in investing? Signal is information grounded in physical reality that cannot be manufactured: infrastructure that exists, minerals in the ground, power that flows. Noise is information that only requires belief or consensus to appear true: stock price movements driven by sentiment or executive statements about future outcomes. The Three Maps framework helps you identify which is which before acting on information.

What is the Three Maps framework? The Three Maps framework is a strategic analysis method requiring three maps drawn in order: (1) the physical map, where are the resources, infrastructure, and irreplaceable geography; (2) the financial map, where does capital flow and what is mispriced; (3) the narrative map, what does the market consensus believe? The gap between the physical map and the narrative map is where exploitable signal lives.

What did Leopold Aschenbrenner invest in and why? Aschenbrenner, author of the 2024 document Situational Awareness, founded the hedge fund Situational Awareness LP in September 2024 with roughly $225 million in seed capital backed by Stripe co-founders Patrick and John Collison, among others. He positioned himself in energy infrastructure and data center capacity rather than AI chip companies. His largest disclosed position is Bloom Energy (NYSE: BE), a solid-oxide fuel cell company providing off-grid power for data centers, which returned 176% from his Q4 2025 entry point to May 2026 (Motley Fool, May 2026; ECIKS, May 2026). The fund returned 47% after fees in the first half of 2025 alone versus roughly 6% for the S&P 500, according to the Wall Street Journal. By early 2026, the fund’s publicly disclosed US equity exposure had grown from $225 million to approximately $5.5 billion, and its most recent SEC 13F filing in May 2026 showed total assets of $13.7 billion (Fortune; ECIKS, May 2026). Top holdings across the portfolio returned between 100% and 800%+. His thesis: AI development is a software competition sitting on top of an energy competition. The physical constraint is power not chips, not algorithms. He applied the same logic Rockefeller used with oil pipelines: control the prerequisite, not the commodity. Rather than buying the companies everyone could see winning, he bought the infrastructure that every company would need to compete at all.

Why is Africa underpriced in AI-era investment models? Western financial models apply a narrative map, political risk, and frontier market classification to African assets while ignoring the physical map. The DRC contains approximately 70% of global cobalt reserves (USGS, 2025), a critical AI hardware input. East Africa’s Great Rift Valley is among the world’s largest geothermal energy sources, directly competitive with AI data center power requirements. These physical facts exist independent of the narrative applied to them.

What is Ken Griffin’s position on AI job automation? At the May 2026 Stanford Leadership Forum, Citadel CEO Ken Griffin stated that AI is automating PhD-level financial research in hours that previously took human teams weeks or months. He described going home “fairly depressed” after observing the capability. Citadel nonetheless invested over $4.2 billion in AI and tech positions in Q3 2025, illustrating the gap between public narrative and financial action.

What is Jeff Bezos’s prediction about AI and employment? Bezos stated in May 2026 on CNBC that AI will create a labor shortage rather than mass unemployment, predicting that productivity gains will lower the cost of food and housing. Amazon CEO Andy Jassy separately acknowledged in June 2025 that the company would need fewer people in roles being automated, illustrating the gap between the optimistic narrative and the physical employment reality.