Forecasts as they were recorded.

Loans on apartment buildings keep going bad, topping 9% by late 2027, as cheap interest-rate protection from 2021-22 expires and breaks the soft landing.

Airlines keep flying older planes through 2027 as new-jet deliveries lag, so the money flows to engine repair shops and spare parts, not new aircraft.

Turning uranium into reactor-ready gas stays expensive through 2027, because only three Western plants do it and they can't catch up to the nuclear revival.

By 2027 most new AI servers ship with liquid cooling, because the hottest GPU racks can't be air-cooled, making plumbing mandatory data-center gear.

A full monthly dose of a brand-name weight-loss drug drops to $199 cash or less by 2027, as cheap-to-make pills crack the old injectable supply chokehold.

Some AI data-center loan gets downgraded or a GPU-cloud startup goes bust by 2027, as used-chip values crash and 3-year loans come due against shrunken collateral.

Nickel likely doesn't break $20,000 a ton before 2028, because the Gulf acid squeeze meant to choke battery-nickel supply is already reversing.

Cobalt probably stays below $75,000 a ton through 2027, because stockpiles outside Congo keep cushioning the export-cap shortage longer than the bet allows.

By 2027 a market forms for the right to plug huge data centers into the PJM grid, and those energized hookups trade at over twice the price of bare land.

Cheap gas stays stranded in Appalachia at a steep discount through 2028, because the binding limit is pipeline capacity and permits, not the gas itself.

Electrician pay rises faster than electrical-equipment prices through 2028, because skilled trades, not gear, are the real bottleneck on building out the grid.

Prefab and factory-built electrical gear probably won't relieve the electrician shortage before 2028, so the labor crunch likely stays tight.

Starlink still runs most of the world's low-orbit internet satellites through 2028, because it locked up the spectrum and head start that latecomers can't buy back.

China's exports of the powerful rare-earth magnets in EVs and wind turbines drop at least a quarter below 2023 by 2028, as it hoards the one step it dominates.

Even with 'too much nickel' headlines, the special battery-grade nickel gets scarce: an acid shortage in Indonesia chokes the supply, pushing its premium past $4,500 a ton by mid-2027.

Loans on apartment buildings keep going bad, with the delinquency rate topping 9% in some month by late 2027, as cheap 2021-22 floating-rate deals blow up despite the 'soft landing' talk.

A brand-name weight-loss shot from Lilly or Novo will list at $199 a month or less cash for a full ongoing dose by end-2027, as price competition cracks the high-margin franchise.

The idea that GPUs make rock-solid loan collateral breaks by end-2027, when a chip-backed AI data-center loan gets downgraded or a debt-loaded cloud startup restructures or files bankruptcy.

Nickel tops $20,000 a ton at least once before end-2027, because a Gulf sulfur-acid shortage starves Indonesian battery-nickel plants and flips the 'too much nickel' story into a shortage.

Cobalt metal spikes to $75,000 a ton or more for at least a month by end-2027, once Congo's hard export cap drains the stockpiles that kept prices flat through early 2026.

The first storage squeeze AI builders hit is enterprise SSDs, the big data-center hard drives, which stay scarce through 2027 because the exact qualified, server-approved ones run short even after headlines cool.

The memory crunch spreads past the fancy AI chips into ordinary system memory too, so regular DRAM gets tight and pricey, not just the sold-out AI stuff everyone already knows about.

Fast storage won't get treated as core AI infrastructure until vendors show hard production numbers, and that proof stays missing for now.

Washington edges toward a government backstop for catastrophic cyberattacks the way it backs terrorism insurance, but the formal plan stays unfinished by 2028.

Consumer AI moves onto always-on devices packed with sensors near you, where battery life, heat, and privacy decide the winner, not a standalone gadget.

AI data centers get ranked by hard proof they can actually get powered, because a bad power assumption can strand land, chips, and billions in debt.

Getting grid power for a data center starts costing like booking reserved capacity, making the AI-power crunch trackable through public utility filings instead of guesswork.

AI data centers get rationed by who can plug into the power grid, not by chips. The bottleneck is substations and energization permits, not GPUs.

Once AI makes designing new microbes cheap, the hard part becomes scaling them into real factories: cost, contamination, yield, and approvals. That industrial bridge is where the money gets stuck.

Phones and gadgets hit a heat-dissipation wall, so each new model feels barely better. Annual upgrades lose their point and replacement cycles stretch out by 2030.

Spy-satellite radar resolution as the new scarce strategic chokepoint is a clever idea, but satellites are scaling fast, so this one probably doesn't bind.

One company makes the machines that write chip-design blueprints, and that caps how many new AI chips the industry can launch each year by 2031.

America's mining boom will stall because there aren't enough certified geologists to sign off on deposits, and banks won't fund a mine without that sign-off.

The real money in data-center power shifts to a new official, sellable right to plug a giant load into the grid, even though today it is still buried inside the price of plain powered land.

AI servers run so hot that liquid cooling becomes mandatory, and Vertiv keeps cashing in, topping $12B in revenue, though rivals may eat into the margins.

With new jets scarce, airlines fly old planes longer, so GE makes most of its money fixing and supplying engines, pushing revenue past $45B.

The real bottleneck for powering AI is the giant gas turbines, and GE Vernova's order slots are sold out to 2028, pushing its power revenue past $20B.

The shortage of skilled electricians probably won't ease by 2028, because prefab building and training programs can't close the wage gap fast enough.

Electricians keep getting paid far more than the gear they install, all the way through 2028, because the training pipeline takes years to catch up.

The squeeze on filling vials with sterile drugs probably won't ease by 2028, but new contract factory lines commission quicker than ships, so relief is possible.

Factories that fill injectable drugs into vials stay jammed past 90% busy with year-long waits through 2028, because new capacity comes online too slowly.

Contract factories that fill injectable shots stay jammed past 90% capacity with year-long waits through 2028, so the money flows to whoever owns the filling lines, not the drugmakers.

Rare medical isotopes, made just grams a year, stay the real bottleneck on cancer radiation drugs through 2029, not the targeting molecule. New sources might ease it, but probably not in time.

Western rare-earth refining outside China probably won't catch up before 2029, because new subsidized plants take years to qualify and the China price premium holds.

Rare earths refined outside China stay scarce and pricey through 2029, because the new plants ordered now mostly don't come online until 2028 or later.

Will OpenAI reach its profit cap for the first round of investors by 2035? (~35% we're right)

If at least one frontier AI model has been trained by a Chinese firm before 2027, will a C (~40% we're right)

Prices for high-voltage grid-connection gear stay high through 2027, because the AI-and-electrification building boom keeps the bottleneck tight while everyone watches GPUs instead.

The rubber stoppers and parts inside every weight-loss injection pen keep minting money, pushing West Pharmaceutical past $3.3 billion by 2027, even as some GLP-1 drugs go to pill form.

The company that enriches uranium fuel keeps growing past $500 million by 2027, because the nuclear-reactor revival needs Western enrichment, and Centrus is the only licensed US source.

Micron keeps spending $15 billion-plus a year building memory chips, because the new AI model designs are hungry for memory, not raw compute. This is a lasting shift, not a blip.

South Korea, the fastest-aging rich country, hits one retiree for every 2.5 workers by 2037, putting heavy strain on pensions and the shrinking workforce.

Germany ages to just two workers per retiree by 2052 as the boomers finish retiring, squeezing the labor base that runs its industry.