Have you noticed how the scariest sentences in tech now arrive right before someone wants a sky-high listing price? I have. Every few weeks another senior researcher or chief executive tells the world that the next model might cook up a bioweapon, slip the leash, or wipe out a chunk of office work. The language is grave. The timing is rarely accidental. When a company is preparing to sell a story to public markets, fear can look a lot like prestige.
Why AI Doom Talk Should Not Drive Your Portfolio
I am not here to pretend the technology is a toy. Large language models can write code, translate, summarize, and surprise people who have not used them much. That is real. What is not automatically real is the leap from “this tool is useful in structured tasks” to “civilization has ten years left.” Investors keep getting invited to treat that leap as settled science. It is not. It is a narrative with a price tag attached.
A basic habit of adult money management is to ask who benefits. Founders who want a valuation measured in the trillions are not neutral witnesses. Researchers who work inside the same firms are not sitting in a monastery. Media desks love a dramatic clip. Put those incentives in a room and you get a feedback loop. The loop sounds like prophecy. It behaves like marketing.
The Convenient Calendar Of Fear
Watch the calendar. Warnings intensify when capital is being raised, when a listing window opens, or when a rival product looks cheaper. That pattern is older than this cycle. In earlier years, a lab delayed a weaker model and framed the delay as public protection. The model was not a superintelligence. It struggled with simple reasoning. The press still ran with danger. The brand still collected attention.
This year the volume has been turned up. A resignation tweet from a little-known account can travel like a stadium wave. Influential accounts amplify it. Suddenly the public is discussing extinction probabilities as if they were quarterly guidance. I find that odd. Genuine risk work tends to be boring, technical, and slow. Viral fear is none of those things.
Fear sells. Outrageous claims travel farther than a careful note about error rates in unstructured text.
Call it doom trolling if you like. The phrase is a bit cheeky, but it captures the mechanic. Phone users doom-scroll. Companies can doom-post. A dark story makes the product feel world-historical. World-historical products attract world-historical multiples. That is the quiet arithmetic underneath the sermons.
What The Models Actually Do Well
Start with the unglamorous inventory. In domains with tight rules and quick feedback, these systems look impressive. Coding has compilers. Translation has parallel texts. Customer support has tickets and scripts. Failure is visible, so teams can sand down the worst edges. That is why enterprise pilots keep clustering there.
Step outside those fences and the shine fades. Journalism, strategy memos, legal reasoning without retrieval, and open-ended research still collide with a stubborn habit: the model will invent a citation, a number, or a person with complete confidence. I have watched otherwise bright people treat that fluency as proof of understanding. Fluency is not understanding. It is next-token competence wearing a nice suit.
Investors sometimes forget that markets do not pay trillions for a slightly better spreadsheet. They pay that kind of money for a platform that rewrites cost structures across the economy. Maybe that arrives. Maybe it does not arrive on the timetable baked into current prices. The gap between those two sentences is where portfolios get hurt.
- Structured work with clear pass-fail tests is the sweet spot.
- Open-ended knowledge work still leaks errors at scale.
- Hallucinations are not a quirky footnote. They cap reliability.
- Cost of inference can erase gross margin fantasies.
- Switching costs are lower than the branding suggests.
The IPO Weather System
Listings like weather. They need a warm front of attention. A sequence of high-profile offerings in space, foundation models, and adjacent platforms creates a season. Each deal wants to look like the defining asset of the decade. If the product is still, at heart, a probabilistic text engine with rising compute bills, the story has to carry extra voltage.
Apocalypse talk supplies voltage. So does the claim that a model is too dangerous to ship, followed by a ship date that arrives anyway. So does a public letter about slowing down while training runs continue in the background. I am not accusing every engineer of bad faith. Plenty of people inside these labs are sincerely uneasy. Sincerity and strategy can share a building. Markets should price the strategy.
There is also the unpretty possibility that the arms race is getting expensive in a way that no longer flatters a growth story. Training and serving costs climb. Quality gains get smaller. That is a hard slide to show investors who were promised a vertical line. A moral pause sounds better than a budget pause. I cannot prove that motive in every case. I can say the incentive is sitting on the table in plain sight.
Regulatory Theater And Who It Helps
Calls for rules often arrive dressed as humility. Look closer and you may see regulatory capture in work clothes. Heavy compliance favors firms that already have lawyers, lobbyists, and giant clusters. It is less kind to lean open-source teams and foreign competitors that sell capability at a discount. If you cannot beat a cheaper model on price, you can try to fence it out of rich markets.
That does not mean every safety proposal is a cartel move. Some risks around cyber misuse and social engineering are ordinary and current. Those deserve boring standards, audits, and liability rules. Extinction rhetoric is a different animal. It pulls the debate into metaphysics, where evidence is thin and branding is thick.
If you cannot beat them on cost, the next best play is to make their product look illegitimate.
Investors should separate two files in their heads. File one: practical harm that already exists, from scams to brittle automation in customer operations. File two: speculative end-of-species scenarios used as a halo. File one belongs in risk models. File two belongs in the marketing appendix.
How Hype Distorts Valuation Math
Valuation is a story about cash that has not shown up yet. When the story is “this might save or end the world,” discount rates get sloppy. People stop asking simple questions. What is the gross margin after inference? Who owns the customer relationship once the novelty fades? How fast do open alternatives close the gap? What happens if enterprises cap spend after the first wave of pilots?
I have found that the most useful habit is to translate miracle language into unit economics. Tokens cost money. Talent costs money. Data center power costs money. Sales cycles in big companies are slow. Procurement teams do not buy the apocalypse. They buy a reduction in ticket volume or a faster first draft that a human still has to check.
| Claim In The Air | Investor Translation | What To Check |
| Model may go rogue | Brand wants exceptional status | Release cadence versus pause talk |
| Work will vanish | TAM is being stretched | Actual task automation rates |
| We must slow down | Costs or returns may be stalling | Capex, energy, and model deltas |
| Only we can be trusted | Moat via regulation | Open model performance gaps |
None of this requires cynicism as a personality trait. It requires a refusal to outsource judgment to people who get paid when the multiple expands.
A Personal Filter I Use On Tech Narratives
In my experience, the stories that age well share a dull virtue. They survive contact with a spreadsheet. The stories that age badly share a theatrical virtue. They survive contact with a camera. AI doom sits closer to the second camp right now, at least in its public form.
I still want the tools. I use them. I also keep a sticky note in my head: if a claim cannot be falsified this quarter, it is not an input for position sizing. You can believe long-run transformation and still refuse to pay a price that assumes the transformation is already in the bag. Those two ideas are not enemies.
Perhaps the most interesting aspect is how quickly the conversation slides from product quality to moral theater. That slide is useful for attention. It is less useful for anyone trying to decide whether a listing is a business or a relic of a fever.
What Ordinary Investors Can Do Instead Of Panic
First, treat mega listings in a crowded theme as a crowded trade. Crowds are not always wrong. They are often expensive. If everyone you know already agrees the asset is destiny, the easy money may have already introduced itself to someone else.
Second, diversify like you mean it. Not the poster-board version where you own twelve names that all rise and fall with the same compute cycle. Real spread across sectors, geographies, and cash-flow styles. When a narrative breaks, correlation has a habit of going to one at the worst moment.
- Write down the cash-flow case without the end-of-world adjectives.
- Stress the model for slower capability gains and higher energy costs.
- Ask what an open alternative does to pricing power in three years.
- Cap any single theme so a mood swing cannot wreck the plan.
- Revisit the thesis when the company ships, not when it tweets.
Third, keep your head. That sounds like a greeting card. It is still the whole game. Markets punish people who outsource their nervous system to a feed. If a probability of human extinction is being discussed like a product feature, step away from the screen and look at invoices, margins, and customer retention.
The Difference Between Risk And Atmosphere
Risk is specific. A model can be used to draft phishing copy. A chatbot can leak a prompt. A company can burn cash while chasing a benchmark that customers do not buy. Atmosphere is the fog around those facts. Atmosphere says the species is on the clock. Atmosphere is hard to hedge. Risk can be underwritten, limited, and sometimes ignored if the price is wrong.
I keep coming back to cigarettes and the old playbook of suppression. That industry hid harm. This industry advertises a grander harm. Why the inversion? Because the product needs to feel larger than a writing assistant. If it is only a writing assistant with plugins, the multiple compresses. If it is a civilizational event, the multiple inflates. That is a blunt reading. Blunt readings are underrated.
White Collar Work Is Not A Movie Scene
Will some tasks shrink? Yes. Drafting, first-pass analysis, boilerplate, and search-like synthesis are already changing. That is a labor-market story with winners and losers. It is not automatically a mass-extinction story. Companies still need people who can tell when the output is wrong. That checking layer is messy and expensive. It does not vanish because a keynote said so.
Investors who confuse task automation with firm destruction will misread incumbents. A bank that uses models to speed KYC is not the same object as a lab that sells the model. A software vendor that embeds assistance into an existing workflow may keep the customer even if the underlying model is swapped. Distribution still matters. It always did.
I have sat through enough cycles to distrust phrases like “this time the old rules are dead.” Sometimes a rule is dead. More often the rule was napping while a slogan did cardio.
Diminishing Returns Hide In Benchmark Theater
Leaderboards make great screenshots. Businesses live on reliability, latency, and price. A two-point gain on a public test can mean little if the model still fabricates a contract clause. If each new training run costs more and moves the needle less, the growth story has a problem that no op-ed about alignment will fix.
That is why I watch second-order signals. Hiring freezes in applied teams. Quiet price cuts. Enterprises that keep pilots in a sandbox for a year. Energy contracts that look like a bet on volume the product has not earned. Those details are less cinematic than a rogue-agent anecdote. They pay better as research.
A working checklist I keep: Separate capability demos from paid usage. Separate paid usage from durable margin. Separate durable margin from listing narrative. Only the last item is optional for a healthy market. The first three are not.
Media Incentives Are Part Of The Machine
A calm article about token costs will not travel like a quote about human extinction. Desks know this. Executives know this. The result is a market for quotes. Once that market exists, the quotes appear on schedule. You do not need a smoke-filled room. You need aligned incentives and a public that treats confidence as evidence.
According to industry observers who study attention cycles, dramatic frames outcompete technical ones even when the technical ones are more relevant to cash flow. That should not shock anyone who has watched markets for a decade. It should still annoy anyone trying to allocate capital with a straight face.
So I read the scary posts. Then I look for the shipping notes, the pricing pages, the customer cohort data if it exists, and the competitive set that does not give interviews. The quiet competitors are often the ones that force the loud ones to talk about safety instead of price.
Keep Liquidity And Ego In Separate Drawers
There is a social payoff to sounding fluent in existential risk. It signals seriousness. Portfolios do not care about your dinner-table seriousness. They care whether you bought a narrative at a peak multiple because you did not want to look unsophisticated.
I would rather look unsophisticated and own a boring cash compounder than look visionary in a drawdown I cannot explain to myself. That preference is not anti-progress. It is pro-sleep. Progress will continue whether or not I overpay for the press conference.
If a future system does become unmanageable, the correct personal response is still not “I should have concentrated my net worth in the company that warned me the loudest.” That sentence only makes sense inside the marketing department.
A Cleaner Way To Think About The Next Twelve Months
Assume the tools keep improving in jagged steps. Assume customers keep buying the useful slices and ignoring the mythology. Assume energy, chips, and talent remain scarce. Assume open models keep nipping at heels. Under those assumptions, some firms will make real money. Others will list at a story price and spend years growing into it, or failing to.
Your job is not to settle the philosophy seminar. Your job is to avoid being the last buyer of atmosphere. That means smaller position sizes in pure narrative names, more curiosity about picks-and-shovels businesses with actual invoices, and a willingness to miss a rocket that was priced as if it had already reached orbit.
- Prefer evidence of repeated paid use over a single jaw-dropping demo.
- Treat pause letters as communications, not operating results.
- Remember that regulation can protect incumbents as easily as the public.
- Leave room in the portfolio for themes that are not on magazine covers.
- Rebalance when applause gets louder than cash collection.
The Mental Health Cost Of Living Inside The Pitch
There is a quieter harm that does not show up in a term sheet. People start to treat every product update as a referendum on the species. That is exhausting and, frankly, a bit silly when the same update still cannot keep a simple fact straight. Tuning out the doom channel is not denial. It is hygiene.
I do not know how long the fever lasts. Fevers end. When this one breaks, the damage will not be evenly spread. Concentrated believers will feel it first. Diversified skeptics will have duller parties and better options. I know which group I would rather join on a Tuesday morning.
Perhaps machines will one day outrun us. Today, the clearer threat to a private investor is paying a mythological price for a useful but finite tool.
So ignore the apocalypse as an allocation input. Watch the product. Watch the bill. Watch the customer. If those three still look extraordinary after you strip out the adjectives, you may have something. If they look like a better office suite wearing a black cape, walk on. There will be another cape next season. There always is.
Final Thought For Anyone Staring At A Subscription Button
Listings will come. Friends will send threads. Someone will say you do not understand the discontinuity. Smile. Ask for the unit economics. If the answer is a parable about the end of humanity, you already have your answer. Money is allowed to be unimpressed. In fact, that is one of the few edges a patient person still owns.
