Why most of the people who rush will end up poorer, why the tools do not save them, why the window is a few short seasons — and how to be the one who already knows where to dig.
Mainlode · Alsatronix Solutions · 2026
24 January 1848. James Marshall looks into the tailrace of a sawmill on the American River and sees something glint.
He is building a mill. He is not looking for gold.
Nobody believes him. San Francisco is a village of about a thousand people — and stays one for months.
Sources: Library of Congress; Polk, Fourth Annual Message; Starr & Orsi
30 November 2022. A laboratory releases a chatbot.
Eleven months to believe a carpenter. Two months to believe a chatbot.
History does not repeat — but it rhymes. (A line usually given to Mark Twain, who never said it. This story is about seeing what is actually there.)
Sources: UBS/Reuters, Jan 2023; company capex guidance, Q2 2026
The question in 1848 was never whether there was gold. It was: where is my claim — and what is already staked?
Sources: Stanford AI Index 2026; IFR; IRENA; BNEF; IEA; CSIS; DefiLlama; Stripe; Lilly; Deloitte; EssilorLuxottica; Roblox
Time to $100M in revenue: Lovable, 8 months after its first $1M.
An AI lab's annualised revenue: $9B → $65B in seven months.
Sources: company disclosures; lovable.dev, Jul 2025; Anthropic, Jul 2026
The forty-niners' phrase for what happened to them: you paid, you walked a long way to see the wonder — and what you mostly saw was the road.
"Oh, the pleasures of going to see the Elephant!" — Lucy Cooke, on the trail, 1852
"That desert is truly the great Elephant of the route." — Lucius Fairchild, 1849
"I see men every day making their fortunes, but I see five times as many more working twice as hard to keep from starving." — John E. Fletcher, June 1850
They named their camps accordingly: Poverty Hill. Skunk Gulch. Hell's Delight.
That is not a paradox. That is what a rush is.
Sources: Rodman Paul (1947) via Rawls; PBS; NY Fed; National Geographic — historians' standard estimates. Note: 1851 miner count estimated; 1854 daily take derived from ~$75/month.
Gold was found in August 1896. The news reached Seattle in July 1897 — eleven months later. By the time most stampeders arrived in summer 1898, almost 10,000 claims were already recorded.
Source: Pierre Berton, Klondike; NPS
The gold is real. The infrastructure gets built. Most of the people who rushed pay for it.
Sources: Odlyzko; Campbell (CEPR); MIT Energy Initiative; Meta filings; CoinGecko
The dot-com boom burned through $500 billion building internet infrastructure. Most of it was laid by companies that went bankrupt. Less than 5% of the fibre was ever lit.
The investors lost everything. The fibre stayed in the ground.
A decade later, YouTube, Netflix, and Zoom ran on that bankrupt fibre — for almost nothing. The infrastructure was real. The companies that paid for it were not the ones who profited from it.
Peak market cap $555B in 2000. Shareholders waited until December 2025 to break even. 25 years.
$500B+ laid. <5% lit at the time. Later used by every streaming service on Earth — at near-zero marginal cost.
The infrastructure gets built. The people who paid for it rarely profit. The people who use it cheaply — do.
The AI equivalent: $730B in compute capex in 2025. The apps that run on it cheaply in 2030 will be the ones that matter.
Sources: Odlyzko; FCC; Cisco annual reports; Nasdaq historical data
Sam Brannan learned of the gold when Sutter's workers paid in dust. Before the East had the news, his store was reportedly taking $36,000 in nine weeks. Richest man in California.
John Studebaker arrived in 1853 and never mined: $10 wheelbarrows, and $8,000 home to fund a wagon company. Wells & Fargo moved and banked the gold (1852). Huntington & Hopkins ran a hardware store on K Street — and built the western half of the transcontinental railroad. Leland Stanford kept a store for miners; the store became a railroad, the railroad a university.
Arrived in 1853 as a wholesaler of dry goods. The riveted trouser came in 1873. There were no jeans in 1849.
Average daily earnings collapsed from $20 in 1848 to $3 by 1852 — below a farm laborer's wage
The storekeeper who knew first. He never swung a pick.
The man who actually found the gold died poor in 1885, still petitioning for a pension.
Not the man who found the gold — Marshall died poor in 1885. Not the man who owned the land — Sutter died in 1880, still petitioning Congress.
Marshall finds gold. Brannan hears about it from Sutter's workers, who pay for goods in dust.
Brannan quietly buys every pick, pan, and shovel in San Francisco and Sacramento. Total cost: a few hundred dollars.
Brannan runs through the streets of San Francisco holding up a bottle of gold dust, shouting: 'Gold! Gold! Gold from the American River!' He is the only hardware supplier in the territory.
His store takes $36,000. He never mines a single ounce.
"He was the first man to get rich from the gold rush — and the last man to touch a shovel."
The AI equivalent: the companies selling compute, APIs, and tooling — not the apps built on top of them.
Arrived chasing gold in a territory that had fewer than 15,000 settlers in 1848
Below a farm laborer's wage back East. The cost of the trip alone was $500–$1,500.
Estimated share of forty-niners who returned home poorer than they left, after costs
Roughly the odds of finding a claim that paid more than the cost of getting there
The gold was real. The math was not.
Sources: Starr & Orsi, Rooted in Barbarous Soil; Rohrbough, Days of Gold; Library of Congress
The Miners' Ten Commandments — a letter sheet sold to miners. Nearly 100,000 copies in a little over a year.
"Thou shalt have no other claim than one." — Commandment I, Placerville, 1853
Three ways to win a rush, then and now:
Information
Infrastructure
To the phase when the gold becomes an industry
Remember those three. We will come back to them.
Sam Brannan — who knew first, who sold the shovels, the richest man in California through the 1850s and 60s — lost it all to a divorce settlement and bad investments, sold pencils door to door in his late sixties, and died in 1889 without money for his own funeral.
Knowing first is not enough if you stop knowing.
Which brings us to the reason most people lose. It is not greed, and it is not stupidity.
It is biology.
To make a human being
Brain builds 2–3× more connections than it will keep, then spends ~25 years deciding which to delete. That is what education is, physically.
Processing speed peaks
Working memory peaks
Reading a face — knowing what someone across a table actually feels — keeps improving
Vocabulary peaks
When we say wisdom comes late, we are not being poetic. It is measurable — and it is late.
Sources: Petanjek et al., PNAS 2011; Hartshorne & Germine, Psychological Science 2015; Grossmann et al., PNAS 2010
Every generation must climb a taller ladder before it can add a rung.
Max Planck: a new truth wins because "its opponents eventually die, and a new generation grows up that is familiar with it."
Tested in 2019 on 452 eminent scientists who died early: after each death, work by outsiders in that field rose 8.6%.
Fields advance one funeral at a time. So do markets.
Sources: Jones & Weinberg, PNAS 2011; Planck, Scientific Autobiography; Azoulay, Fons-Rosen & Graff Zivin, AER 2019
25,000 publicly traded North American companies, 1950–2009: a public company's half-life is about 10 years, in every sector — and its chance of dying does not fall as it ages.
Of the 1955 Fortune 500, 52 were still on the list in 2019.
"It takes all the running you can do, to keep in the same place." — the Red Queen
Sources: Daepp et al., J. R. Soc. Interface 2015; Innosight 2018; AEI 2019
Built the first digital camera in 1975. Management: "That's cute — but don't tell anyone about it."
Was offered Netflix for $50M in 2000 and "just about laughed us out of their office."
Had ~51% of the world's smartphones at the end of 2007 — the market, the engineers, the money — and six years later had none of the three.
Each had wisdom. Each was too old by the time it used it.
The base rates:
Sources: BLS; Ghosh (HBS); Flyvbjerg & Gardner 2023; McKinsey/Oxford 2012
The mean age of the founders of the fastest-growing 0.1% of new American companies.
A 50-year-old founder is 1.8× more likely to build a top-growth company than a 30-year-old.
Age is not the problem. Speed and wisdom live in different bodies — different generations, different companies.
The whole game is to put them in the same body. To be born with the map.
Source: Azoulay, Jones, Kim & Miranda, AER: Insights 2020
A quarter of YC's batch had codebases that were 95% written by AI
Went from $1M to $500M in revenue faster than any software company in history — and was bought for $60B four years after it was founded
More than a quarter of its new code is generated by AI
Nine months of gestation is now six weeks. You can be born in six weeks.
Sources: TechCrunch, Mar 2025; Sacra; Alphabet Q3 2024 call
Enterprise GenAI pilots with no measurable P&L impact
AI projects that reach production; 8 months from prototype
Agentic AI projects that will be cancelled by 2027
Companies abandoning most AI initiatives (up from 17% in one year)
Show no tangible value; leaders follow 10% algorithms / 20% tech & data / 70% people & process
AI initiatives that delivered the expected ROI
RAND, 2024: Five root causes of AI project failure — four are not technical.
Experienced developers using AI tools took 19% longer — and believed they had been 20% faster
More AI adoption, lower throughput (−1.5%), worse stability (−7.2%)
Two-thirds of developers on AI code: "almost right, but not quite"
Sources: METR, Jul 2025; DORA 2024; Stack Overflow 2025
Instagram copied Snapchat's Stories — and had more daily users than all of Snapchat eight months later
Google shipped Bard
DeepSeek erased $589 billion of Nvidia's value
Whatever is expensive to build today is cheap to copy next year.
Sources: Instagram; Snap S-1; Google; Forbes, Jan 2025
385 with a stated reason, most with more than one.
Timing explained more of success than team, idea, model or funding (Bill Gross, 200 companies). 70% of startups scale before they should.
Companies that bought or partnered for AI reached deployment ~67% of the time. Internal builds: ~33%.
Ideas are free now. Surveyed claims are not.
Sources: CB Insights, Mar 2026; Gross, TED 2015; Startup Genome; MIT NANDA 2025
California's pick-and-pan window. After 1850 the surface gold was largely gone, even as miners kept arriving.
Gold was a capital business — hydraulic mining, ditch systems, companies.
Fewer than two dozen stock companies owned it.
Claims gone eleven months after the news.
Took 40 years to pay off — the decade belonged to factories that rebuilt around the motor first.
Perez: every revolution runs installation → frenzy → turning point → deployment. The golden age comes after the bubble bursts — and belongs to those who came through the frenzy with real businesses.
Sources: History.com; cprr.org; Paul David 1990; Atkeson & Kehoe; Carlota Perez 2002
Of all venture capital on Earth went to AI. That one quarter ≈ 70% of everything invested in all of 2025.
Half of all venture dollars went into 0.05% of deals. New funds forming at the slowest rate since 2016.
The length of task an AI can complete alone doubles every 3–7 months.
The cost of a given level of intelligence falls ~10× a year.
The surface claims — "the chatbot for X" — are gone. What is left is the hydraulic phase: vertical, regulated, data-heavy, integration-heavy. The kind that rewards a survey.
Sources: Crunchbase, Apr 2026; PitchBook-NVCA; METR; a16z
The industry needs $2 trillion of annual AI revenue by 2030 to pay for its compute — and is $800 billion short
Posted its first negative free-cash-flow quarter
The largest listing in history, trades 45% below its post-IPO high
Installations forecast to contract in 2026 for the first time in twenty years
None of this means the gold is fake. It means the frenzy is doing what frenzies do. In every prior crash the infrastructure passed to two kinds of people: those with real businesses when the music stopped, and those who sold the shovels.
Stake the claim before the correction, not after. Nobody wants to be buying a ton of goods in Seattle in 1899.
Sources: Bain Technology Report 2025; Alphabet Q2 2026; NPR, Aug 2026; SolarPower Europe
We are a survey office for the second gold rush. We research business ideas until they either die on the desk — or survive to become a brief you can buy.
100 spatial & AI ideas. 250 fintech & trading ideas. Each one took 40+ hours of research and two rounds of sourced verification.
Spatial library ideas are sold exclusively — one license, one builder. You are not competing with 10,000 other people reading the same trend report.
Each brief includes: problem, market, 4 competitors, 4 data sources, MVP scope, key risks, monetization, 5 GTM channels, 5 KPIs, 7 numbered sources.
The outfitter option: we build with you for 20% of what you build. We are paid mostly if you find gold.
The sample brief is free. mainlode.com
Nobody on that list sells the four things together: a surveyed claim, the model and the risks, a build path measured in weeks, and a partner paid mostly if you win.
Sources: Vendr median contract values 2025–26; vendor pricing pages
Rating factors published: market demand 25% · evidence 25% · competition 15% · complexity/moat 15% · assessed value 20%
Time to a first product: 5–7 weeks
More every month — and subscribers vote on where the surveyors go next.
Source: mainlode.com/ideas-250/sample
At this depth
Each with a full research report and a 90-day plan
40 hours of vetting and two rounds of sourced research each
Subscribe, and see every claim. An advisor that answers only from the catalog, and will tell you in writing when nothing fits your budget.
License one. In the spatial library each idea is sold to one buyer only. A first license starts at $10,000.
We build it with you: a named team, a 90-day plan, and 20% of what you build.
Sam Brannan sold shovels to anyone who walked in, whether or not there was gold where they were going.
We survey the ground first. And we are paid mostly if you find gold.
You are going to build something this year.
The only question is whether you spend your nine months of gestation on discovery — learning, expensively, what a hundred thousand other people are learning on the same creek — or on customers.
One that has already survived the research
In six to seventeen weeks
Not the road
The sample brief is free. Read it tonight. mainlode.com
A first license is $10,000. The sample brief is free. Read it tonight.
Eighty percent of venture money is on one creek, at prices that assume the frenzy never ends.
In every rush before this one, the durable fortunes were in the layer that supplied the diggers — and in the claims surveyed before the crowd arrived.
A pipeline of vetted, exclusive ventures across nine frontiers
20% of every outcome
Subscribers vote on where the surveyors go next
Q1 2026: 80% of all venture capital on Earth went to AI — at prices that assume the frenzy never ends
We take 20% of what you build. We are paid mostly if you find gold.
Start with three claims.
You have forty pilots and the same ninety-five percent as everyone else — because you bolted a motor onto the old line shaft.
Adjacent to what you already own
Nobody else can license them
With a team accountable for the outcome
"You have forty pilots and the same ninety-five percent as everyone else — because you bolted a motor onto the old line shaft."
MIT NANDA 2025: no measurable P&L impact. The problem is not the tool. It is the claim.
With a named team, a 90-day plan, and accountability for the outcome.
Stop paying for pilots. Buy claims.
Good advice in 1853. Good advice now: focus. But make it a surveyed one.
The gold is real. Most of the people rushing toward it will end up poorer — for reasons that have nothing to do with their tools and everything to do with what they chose, and when. The window for staking a claim is this season. And you do not have to pay the full price of the trip to see what is out there.
Surface gold lasts three seasons. The claims being staked now are the ones that will matter.
One surveyed claim. Problem, market, competitors, risks, build plan. Time to first product: 5–7 weeks.
Read it before you decide. mainlode.com
mainlode.com/elephant — read a sample brief
Seeing the Elephant