Articles
Long-form research from the QuantAbundancia desk. Methodology, bubble deep-dives, and capital-flow analysis backed by live data from the platform.
Education series(46)
What every trader Googles - Fibonacci, mean reversion, walk-forward validation - but answered with QuantAbundancia's own backtests instead of textbook theory. Read in order; each piece is referenced by the next.
The AI Supercycle: is it a bubble? Why railways and dot-com already answered
Hyperscalers are spending hundreds of billions a year on AI compute while the dominant chipmaker helps finance its own customers. The history of bubbles says real technology and a real bubble are not opposites.
Assignment, exercise, and rolling - expiration mechanics - options trading, chapter 15
The boring plumbing that surprises beginners: auto-exercise thresholds, random assignment, early assignment before ex-dividend, pin risk, and how to roll a position.
Black Monday (1987): the worst day in Wall Street history that caused no depression
On 19 October 1987 the Dow fell 22.6% in a single day, the biggest one-day drop ever. No recession followed. The reason was code, and the response was a Fed that printed before lunch.
Building a trading playbook - turning rules into a repeatable process - trading basics, chapter 11
The market doesn't beat most traders - their own emotions do. A written playbook converts good intentions into rules you can follow under pressure: entry criteria, risk limits, a trade journal, and the discipline loop that improves them.
Buying calls and puts - directional bets with capped downside - options trading, chapter 9
Long options are leveraged direction bets with risk limited to premium. Why most out-of-the-money calls and puts expire worthless, how breakeven works, and how to size by premium-at-risk.
Calls and puts explained: the four basic option positions and their payoffs - options trading, chapter 2
Long call, long put, short call, short put - the four building blocks. Max gain, max loss, and breakeven for each, with numeric examples, and why buyers and sellers face opposite asymmetries.
Cash-secured puts - getting paid to set a limit buy - options trading, chapter 11
A cash-secured put sells a put while holding cash to buy the shares if assigned. How it lowers your cost basis, the full downside risk it carries, and how it feeds the wheel.
Correlation and diversification - why ten stocks can be one bet - trading basics, chapter 9
Owning ten stocks feels diversified. If they all move together, it's one position with extra steps. What correlation is, why narrative-driven clusters move as one, and how to diversify the risk that actually matters.
Covered calls explained - selling premium for income, not protection - options trading, chapter 10
A covered call sells upside for income against shares you own. Why it caps gains, why it isn't a hedge, how assignment works, and when neutral-to-mildly-bullish makes it fit.
Dollar-cost averaging vs lump sum: the math, the psychology, and when each wins
DCA spreads a fixed dollar amount over time; lump sum invests it all at once. The honest math says lump sum wins on average - but DCA wins where it counts: risk, regret, and the way real income actually arrives.
Dot-Com Bubble (2000): the internet was real, which is exactly why it was so dangerous
The Nasdaq ran from about 1,000 in 1996 to 5,048 in March 2000, then fell roughly 78% and erased about 5 trillion dollars. The trend was right. Picking the survivor was the hard part.
The Everything Bubble (2021): the bubble was not in the meme stocks, it was in the price of money
GameStop hit roughly $483, Bitcoin roughly $69,000, ARKK and SPACs fell 67% or more in 2022, and bonds had their worst year in modern history. The single asset everyone missed was the one priced at zero: interest rates.
2008: the bubble was not housing, it was leverage hidden inside complexity
The S&P 500 fell about 57% from its 2007 peak and Lehman filed the largest bankruptcy in US history. The shiny asset was houses. The real bubble was leverage, repriced AAA, and one model assumption.
How options are priced - the six inputs behind every premium - options trading, chapter 5
An option's price isn't one number you read off a screen - it's six inputs run through a model. What moves a premium, why time and volatility both add value, and why buying an option is a three-dimensional bet.
How stock markets actually work - exchanges, bid/ask, and liquidity - trading basics, chapter 2
Behind every trade is an order book matching buyers to sellers. What an exchange really is, why the bid/ask spread is a hidden cost, how liquidity decides whether you get a fair fill, and what market hours change.
How to read a stock chart - candlesticks, timeframes, and volume - trading basics, chapter 4
A candlestick chart is a record of who won each time slice, buyers or sellers. How to read a single candle, why the timeframe changes the whole story, and what volume confirms - without the pattern mysticism.
How to read an earnings report - results vs expectations, guidance, and the call
A stock reacts to earnings versus what was priced in, not to the raw numbers. How to read revenue/EPS against estimates, why guidance moves the stock more than the quarter, and why good prints still drop.
How to read an options chain - the columns that matter and the ones that trap you
An options chain looks like a wall of numbers; it's a structured table. What each column means (bid/ask, volume, open interest, IV, the Greeks), how to spot illiquid strikes, and how to pick an expiration and strike.
Implied volatility explained - why being right can still lose - options trading, chapter 6
IV is the market's forecast of movement, backed out of an option's price. What high vs low IV means, how IV rank works, and why IV crush makes you lose on a correct earnings call. The beginner's biggest trap.
Iron condors and strangles - defined-risk range trades - options trading, chapter 13
Strangles and iron condors are how you bet that a stock stays quiet. Why the iron condor caps your risk, when high IV makes it pay, and a full four-leg numeric example.
Nifty Fifty (1972): the quality bubble, where the companies were great and the price was the mistake
In 1972 about fifty 'buy and never sell' growth stocks traded at 50 to 90x earnings, some above 100x. Then the 1973-74 bear market cut the S&P roughly 48% and many of them 60 to 90%. The companies were excellent. The price was the bubble.
Option premium: intrinsic vs extrinsic value, and why OTM options decay to zero - options trading, chapter 4
The premium splits into intrinsic value (already in the money) and extrinsic time value (which decays to zero by expiration). Moneyness, a worked example, and what an option is worth at expiry.
The Greeks: delta and gamma - your option's directional exposure - options trading, chapter 7
Delta is how much your option moves per $1 in the stock - and doubles as a rough probability and a shares-equivalent. Gamma is how fast delta itself changes. The two Greeks that govern direction, with worked examples.
The Greeks: theta and vega - why direction isn't enough - options trading, chapter 8
Theta is the daily time decay you pay as a buyer and collect as a seller - and it accelerates near expiry. Vega is your exposure to IV swings, the engine behind IV crush. The two Greeks that beat correct directional calls.
Options risk management - the capstone checklist - options trading, chapter 16
The risks unique to options and the rules that contain them: premium-at-risk sizing, defined-risk structures, liquidity, and an eight-point pre-trade checklist.
Order types explained - market, limit, and stop orders - trading basics, chapter 3
The three orders every trader uses, and the slippage trap that empties beginner accounts. When a market order fills at a price you didn't expect, why limit orders trade certainty for control, and how stops actually work.
Position sizing and risk - the 1% rule that keeps you in the game - trading basics, chapter 7
The reason most beginners blow up isn't bad picks - it's bad sizing. How the 1% risk rule works, why your stop distance sets your share count, and the math that lets you survive a losing streak you will absolutely have.
R-multiple and expectancy - does your trading system actually make money? - trading basics, chapter 8
Win rate is the most over-rated number in trading. R-multiples measure every trade in units of risk, and expectancy tells you what you make per trade on average - the one number that decides if a system is worth running.
Railway Mania (1845): the technology was real and the equity was still a bubble
In 1846 Parliament passed 272 railway acts and authorised capital near Britain's entire annual GDP. Most of those lines lost half their value or were never built. Real does not mean safe.
1929: the crash that gets blamed for the Depression it did not actually cause
The Dow ran from about 63 in 1921 to 381 in September 1929, then fell roughly 89% by 1932. The crash was real. The thing that turned it into a decade was the policy response, not the selloff.
South Sea Bubble (1720): the trade-riches story was a cover for a debt swap
South Sea shares ran from about 128 pounds in January 1720 to roughly 1,000 by August, then crashed to 100-200 by December. The real business was never trade. It was converting the national debt.
Straddles and earnings plays - why the move isn't enough - options trading, chapter 14
A long straddle bought before earnings often loses even when the stock gaps. The reason is IV crush. How the expected move is priced in, and why earnings is a volatility trade.
Strike price and expiration: how to read an option chain - options trading, chapter 3
The strike is the fixed price you can transact at; expiration is when the contract dies. Weeklies vs monthlies vs LEAPS, American vs European exercise, cash vs share settlement, and the chain.
Support, resistance, and Fibonacci retracement - finding levels that matter - trading basics, chapter 5
Support and resistance are price levels where the buyer/seller balance has flipped before. What makes a level real, why Fibonacci retracements work as a self-fulfilling map, and how to use levels to define risk instead of guessing.
Tulip Mania (1637): the first bubble, and why the famous version is mostly myth
In 1637 a single Dutch tulip bulb could change hands for the price of an Amsterdam canal house, then the market fell roughly 99% in a week. What actually happened is more useful than the legend.
Vertical spreads explained - defined risk, defined reward - options trading, chapter 12
A vertical spread buys one option and sells another of the same type and expiry. How debit and credit spreads cap both loss and gain, with worked max-profit and max-loss math.
What is a moving average? SMA vs EMA, crossovers, and what they actually tell you
A moving average smooths price by averaging the last N closes, recalculated each bar. SMA vs EMA, the 20/50/200-day lengths, golden and death crosses, and why MAs lag instead of predict.
What is a stock? Ownership, price, and why it moves - trading basics, chapter 1
A stock is a fractional ownership claim on a real business, not a lottery ticket. What you actually own, why the price moves second to second, and the one mechanic - buyers vs. sellers - behind all of it.
What is an option? Contracts that transfer risk, not lottery tickets - options trading, chapter 1
An option is a contract giving the buyer the right - not the obligation - to buy or sell 100 shares at a fixed strike before expiration. The mechanics, the two sides, and why options exist.
What is Fibonacci retracement? A quant's 3-year backtest on 48 live levels
Textbook Fibonacci is 23.6 / 38.2 / 50 / 61.8 / 78.6%. We backtested 48 user-curated levels across 12 themes - basket PF 1.76, Sharpe 1.42, +23.7% over 3 years. Here's what works, what doesn't, and on which name classes.
What is mean reversion? Three flavors backtested - and only one wins consistently
Mean reversion is a class, not a strategy. We walk-forwarded three implementations across 35 thematic names: regression-channel wins 22 of 35, Fibonacci basket wins at PF 1.76, flat-mean Bollinger wins 1 of 35. The structural reason is which mean you assume.
What is short selling? Borrowing shares, unlimited risk, and short squeezes
Short selling means borrowing shares, selling them, and buying them back lower. The risk profile is inverted - gains cap at 100%, losses run unbounded - plus borrow fees and squeezes.
What is walk-forward validation? 104 strategy-ticker pairs tested - only 54% survived
Walk-forward validation separates in-sample fluke from real edge. 104 (strategy, ticker) pairs tested on the QA universe - 56 ROBUST, 20 STABLE, 18 LUMPY, 10 no-trades. Here's the procedure, the verdicts, and why most retail backtests quietly fail it.
What moves a stock price - fundamentals, flows, and narrative - trading basics, chapter 6
Three forces move every stock: fundamentals (what the business earns), flows (who's forced to buy or sell), and narrative (the story the crowd believes). Why the third one drives bubbles, and how to tell which force is in control.
Why most trading strategies fail - curve-fitting and out-of-sample testing - trading basics, chapter 10
A beautiful backtest is the easiest thing to fake and the easiest way to lose money. What curve-fitting is, why a strategy that worked on the past can be worthless, and the walk-forward bar that 46% of QA's tested strategies failed.
Your first trade - a pre-flight checklist - trading basics, chapter 12
The whole beginner course distilled into the steps you run before clicking buy. A ten-point pre-trade checklist covering thesis, levels, risk, sizing, and the exits that have to exist before you enter.
NVIDIA structural deep-dive(7)
Six-part breakdown of the NVDA thesis - Blackwell→Rubin cadence, the CUDA moat, HBM bottleneck, customer concentration, the networking second-business, and custom-ASIC competition.
Nvidia & AMD's China chip loophole just closed - and the demand math barely moves
Commerce extended AI-chip licensing to Chinese firms' overseas subsidiaries on May 31, 2026, closing a year-old loophole that leaked hundreds of thousands of Blackwell, Rubin and MI350x chips through Malaysia. Why the structural read isn't what the headline says.
NVIDIA vs custom ASICs - what TPU, Trainium, and MTIA actually do (and why most of them target inference, not training)
Google TPU, AWS Trainium, Meta MTIA, Microsoft Maia. Four hyperscaler custom-silicon programs in flight, three of them inference-first, one (TPU) that has reached training parity at scale. The technical reason matters: training is where CUDA's moat lives; inference is where it's leakiest. This is the actual competitive landscape under the headline.
NVIDIA's networking moat - why NVLink, Spectrum-X, and the Mellanox acquisition are the second product the market doesn't talk about
NVIDIA's $7B Mellanox acquisition in 2019 was framed as a defensive move. It's now the second-largest product line at the company - networking revenue at ~$13B run-rate, growing 50%+ year-on-year. NVLink, NVSwitch, Spectrum-X, and InfiniBand together form the fabric that makes thousands of GPUs look like one machine. Custom-ASIC clusters still buy NVIDIA networking. This is the part of the moat that survives even if the silicon moat erodes.
NVIDIA's customer concentration - five buyers, half the revenue, and what happens if one of them in-houses
Microsoft, Meta, Alphabet, Amazon, Oracle. NVIDIA's top five direct customers represent roughly 45-55% of data-center revenue depending on the quarter. The concentration is not a footnote - it's the single biggest structural bear case on the stock that the CUDA moat does not defend against. What the 10-K language actually discloses, what each hyperscaler's custom-silicon program is doing, and why Google's TPU is the real comp.
NVIDIA's HBM bottleneck - why Blackwell shipments depend on three Korean fabs
Blackwell B200 needs 8 stacks of HBM3E per GPU. Each stack is fabbed by Samsung, SK Hynix, or Micron - period. NVIDIA's revenue ramp through 2026-2027 is gated not by demand, not by TSMC fab capacity, but by HBM3E wafer allocation at three companies, two of which are in Korea. This is the supply-side constraint that no CUDA moat can fix.
The CUDA moat - why NVIDIA's software ecosystem defends the hardware monopoly long after AMD catches up
AMD's MI300X and MI350X are technically competitive with Blackwell on raw FLOPs. AMD's data-center GPU revenue is still ~1/10 of NVIDIA's. The gap isn't the chip - it's 18 years of CUDA libraries, every PyTorch optimization, every framework integration, every kernel hyperscalers don't want to rewrite. This is what an actual software moat looks like priced into a $3T market cap.
NVIDIA's Blackwell → Rubin roadmap - the annual cadence that priced the next $2T of market cap
NVIDIA went from a two-year product cycle to annual. Hopper 2022, Blackwell 2024, B300/GB300 mid-cycle 2025, Rubin 2026, Rubin Ultra 2027, Feynman 2028. Each cadence step is roughly 2-3× FLOPs and a new HBM generation. The market is implicitly pricing the cadence holding - which means a delay or yield issue at any node would compress the multiple sharply. What the actual roadmap milestones are and what would break them.
Bubble validations(4)
Editorial bubbles measured against the tape. Which ones are real co-movement blocs and which ones are SPY in a costume.
Datacenter power stocks: the second-derivative AI trade
VST, CEG, NRG, GEV, ETR, SO, AEP - the IPPs and utilities feeding hyperscaler AI campuses. ~0.55 residualized correlation. Pure exposure to GW-scale PPAs, not to NVDA's margin trajectory.
Hyperscalers: the failed AI bubble (and why we publish failures)
MSFT, GOOGL, AMZN, META, ORCL look like an AI capex bloc. The data says no. Raw correlation 0.65 collapses to ~0.05 under residualization. SPY in a costume.
Quantum: the most empirically real AI-adjacent bubble
QBTS, RGTI, IONQ, ARQQ, QUBT - five pre-revenue stocks with the highest residualized correlation in the AI taxonomy. Why they trade as one, and how to size the bet.
The 12 AI bubbles, ranked by empirical realness
Twelve editorial bubbles in the AI supercycle. We measured all of them with 252-day residualized correlations. Half are real. Half are SPY in a costume.
Smart-money tape(6)
13F + OGE 278-T disclosures mapped onto the QuantAbundancia bubble taxonomy. What the funds actually own and where the AI exposure clusters.
Bloom Energy (BE) explained - solid-oxide fuel cells, datacenter power, and the 2.8 GW Oracle deal
Bloom Energy signed a 2.8 GW master agreement with Oracle in April 2026 (1.2 GW initial). Aschenbrenner's largest long ($879M position). The on-site fuel-cell pitch: deploy in months when the grid needs years. Here's how the stack works.
Aschenbrenner 13F Follow-Up: Long Book Down 10-19% in 48 Hours
48 hours after Situational Awareness LP's Q1 2026 13F filed, the $3.86B long book broke first: Bloom Energy -17%, IREN -18%, Applied Digital -19%, CoreWeave -14%. The $8.5B chip-short stack is working - but the longs are bleeding harder.
Aschenbrenner Q1 2026 13F: $8.5B chip short (SMH/NVDA puts)
Situational Awareness LP's Q1 2026 13F-HR (filed 2026-05-18): 42 positions, $13.7B notional, $8.46B in puts shorting SMH, NVDA, AVGO, AMD, ORCL, ASML. Largest chip short ever disclosed in an AI-thematic hedge-fund book. Long side: tripled SanDisk, kept Bloom Energy, held 9 Bitcoin-miner-to-HPC names.
Trump's Q1 2026 portfolio: what the OGE filings actually disclosed
President Trump's two OGE Form 278-T filings disclosed 3,711 securities transactions executed Jan 6 → Mar 30, 2026, cumulative value $220M-$750M. Every confirmed buy and sell mapped onto the QuantAbundancia bubble taxonomy - semis, AI hardware, crypto, enterprise SaaS down 30-45% YTD, and three Mag-7 names sold in size on a single February day.
Q1 2026 13Fs: Buffett, Ackman, Druckenmiller, Tiger, TCI, Gates
Seven of the most-watched institutional portfolios filed Q1 2026 13Fs this week. Every top holding mapped onto the QuantAbundancia bubble taxonomy - where the smart money is clustered, where it diverges, and which AI bottlenecks the tape actually owns.
Aschenbrenner portfolio (pre-13F): SNDK, LITE, INTC, COHR, BE
SNDK, LITE, INTC, COHR, BE - the 5 AI-bottleneck stocks Leopold Aschenbrenner's Situational Awareness LP held before its first 13F filing. Superseded by the Q1 2026 13F (42 positions, $8.5B chip short); read this for the original thesis.
Memory & supply chain(9)
The companies most retail traders skip because they don't list in New York - HBM, foundry capacity, DRAM share shifts - and why they gate the AI cycle.
The SK ecosystem - one chaebol, three ways to own the AI memory leader
SK hynix is ~60% of the HBM market. SK Square owns 20.5% of it at a ~42% NAV discount. SK Telecom owns none of it, despite what retail assumes. The map of the SK Group family tree, and which ticker actually carries the exposure you want.
SK hynix (HXSCL) - the HBM leader is heading to a US listing
SK hynix owns ~60% of HBM and the lion's share of NVIDIA's HBM4 Rubin orders, yet US investors can't cleanly buy it. That changes with a US ADR listing filed for end-2026. Forward P/E ~5.4 vs Micron's ~8.4. The structural read.
NVIDIA qualified all three HBM makers for Vera Rubin. The fight is now allocation.
Jensen Huang confirmed SK hynix, Samsung, and Micron all passed HBM4 qualification for Vera Rubin, now in full production for H2 2026. Qualification was never the question. The allocation split, roughly 60-70% to SK hynix, is. Here's the structural read.
The AI bottleneck nobody's watching: T-Glass and Nitto Boseki (3110)
Every NVIDIA AI server runs on an IC substrate reinforced with low-CTE glass cloth. One Japanese company - Nitto Boseki (3110) - has a near-monopoly on the production-grade version. Why T-Glass is a real AI chokepoint, who it serves (NVDA/MSFT/GOOGL/AMZN), and the valuation trap behind the parabola.
Micron (MU) explained - DRAM, HBM, NAND, and the only US-listed pure-play memory bet
Micron crossed $1T market cap May 26, 2026 - the first memory pure-play in history to do so. ~25% HBM share, 'preferred supplier' status on NVIDIA's most advanced nodes, and a $70B revenue projection for FY26. Here's the structural read.
Memory cyclicality - the supercycle that still has a cycle
DRAM has cycled five times in the last decade. Every up-leg ended in a 50-75% drawdown within 18 months of the peak. Micron just printed +1,873% off the lows and traded through $1T market cap. Is AI a structural break - or the biggest mean-reversion setup in the sector's history? Here is the framework, the historical record, and the eight signals that turn a parabolic chart into an exit signal.
TSMC has a structural lock-in on AI compute - and most retail traders are buying the wrong way
Every AI accelerator that ships in 2025-2027 goes through TSMC. There is no second-source for leading-edge logic. This isn't a 'TSMC is a good stock' article - it's a structural breakdown of the lock-in mechanics, the $TSM ADR vs the Taipei primary tradeoff, the CoWoS packaging bottleneck, the Taiwan geopolitical tail, and the honest position-sizing implications most retail TSMC takes ignore.
HBM is the tightest bottleneck in the AI cycle - and three companies own it
High-Bandwidth Memory is the constraint that gates whether each new GPU generation can actually run at spec. Three companies - SK Hynix, Samsung, Micron - control essentially 100% of supply. The compute side gets the headlines; the memory side decides which compute ships. Here's the structural map and what's actually trade-able from a US-retail account.
CXMT enters Corsair DDR5 modules - what China's first Western-brand retail design-win means for the DRAM bubble
ChangXin Memory (CXMT) shipped dies into Corsair's 16GB DDR5-6000 retail kits - the first time Chinese DRAM lands in a premium Western brand at the SKU level. CXMT did $7.4B in Q1 2026 (+719% YoY) and now holds 7.7% global share. Why this confirms - not breaks - the AI memory-bubble thesis, and what it means for MU / SK Hynix / Samsung.
Stack & operations(11)
The brokers, the bots, and the operational reality of running real money against this thesis. IBKR, OKX, what actually breaks at 3am.
Neoclouds ranked by risk-adjusted value: CoreWeave, Nebius, IREN, Applied Digital, SharonAI
The market treats 'neocloud' as one trade. It is five very different risk bets, separated by valuation, leverage, and customer concentration - ranked here from CoreWeave to SharonAI.
Why memory is the hardest thing to build - HBM, DRAM, NAND, and HDD ranked by difficulty
Ranked by how hard they are to build, memory goes HBM > DRAM > NAND > HDD. Difficulty maps onto market concentration and pricing power - the reason the DRAM deficit is more defensible than NAND's.
SK Telecom (SKM): not the SK Hynix backdoor everyone thinks - what you're actually buying
SK Telecom sold its SK Hynix stake to SK Square in 2021. What $SKM owns: a pre-IPO Anthropic stake (at the $965B round) and a gigawatt AWS datacenter buildout, in a discounted Korean telco.
The SK Group, fully mapped - the chaebol behind the AI memory supercycle
SK hynix is worth ~$1.07T. The holdco owning 20% of it trades ~42% under that stake, the parent above smaller still. The SK family tree, the money cascade, and the two Anthropic stakes hiding inside.
Arm Holdings (ARM) - the CPU IP underneath every AI chip, and how the v9 royalty stepup works
+213% YTD on a thesis that every AI chip - NVIDIA Grace, AWS Graviton, Apple M-series, Google Axion, MSFT Cobalt - pairs with an Arm v9 core. This piece walks the IP licensing model, the v9 royalty stepup, and where the bull/bear case actually breaks.
Nebius (NBIS): what it does, the $50B Microsoft-Meta backlog, and the AI-utility read
Nebius runs GPU data centers as an AI cloud - $1.92B ARR, a ~$50B Microsoft and Meta backlog, NVIDIA-backed. What the ex-Yandex neocloud builds, how it makes money, and the financing risk underneath the momentum.
Adobe (ADBE) explained - the AI-disruption trade, Firefly, and a 10x forward multiple
Adobe fell 43% to a ~10x forward multiple - the market pricing AI as an extinction event. Firefly ARR crossed $250M (+75% QoQ) even as AI cannibalizes Adobe's own stock library. The structural read.
Half the AI supercycle doesn't trade in New York
TSMC fabs in Taiwan. ASML lithography in Amsterdam. Samsung and SK Hynix HBM in Seoul. Tokyo Electron in Japan. The most critical AI supply-chain stocks list outside the US - and most US retail brokers can't actually buy them. Here's why we route the international leg through Interactive Brokers.
OKX vs Coinbase for crypto derivatives: where the perp tape actually lives
Coinbase is the default mental model for US-retail crypto. For derivatives, that mental model is broken. Where the deep perp liquidity, the working API, and the demo environment that matches live actually sit - and what we actually trade where, with the bot fleet to prove it.
IBKR Trader Workstation for systematic bots: the operational map nobody publishes
Interactive Brokers' docs cover the API, the symbols, the order types. They don't cover what actually breaks when you run a real bot against a paper account at 3am - IBC re-login windows, the port 4002 trusted-IP trap, the unhealthy-but-Up gateway state, the order-rejection wrapper you need but isn't documented. Here's what we learned shipping live bots against IBKR over 12 months.
I ran 12+ trading bots for a year. Here's what survived, what died, and what I learned.
Most 'I built a trading bot' content is success-bias narrative - the wins go viral, the failures get quietly deleted. This is the inverse: 12 months of running a real bot fleet across OKX crypto perps and IBKR tradfi equities, every failure class catalogued, every meta-pattern explicit. What survived is short. What that says about retail algo trading is the article.
Methodology(3)
How we measure what we measure. Residualized correlation, why narrative beats nothing but data beats narrative.
X-FAB and Sivers: two broken parabolas, and why we didn't buy the dip
One X post sent X-FAB up 77% intraday. A GlobalFoundries deal sent Sivers up 3,100% from its March low. Both then halved. Our five-gate entry check failed both — here's the arithmetic.
Why correlation > narrative in thematic investing
Editorial taxonomies tell you what stocks SHOULD trade together. Correlation tells you what stocks ACTUALLY trade together. Most of the time, those are different lists.
What is residualized correlation? (And why most thematic ETFs lie)
The market beta hidden in 'AI bubbles' makes correlation matrices useless until you strip it. Worked example: why Hyperscalers fail the test and Quantum passes.
Market events(7)
IPOs, ETF filings, single-day tape moves worth the standalone piece.
Roundhill Neocloud ETF ($NCLD): the first concentrated US bet on GPU-as-a-Service
Roundhill filed $NCLD on May 22 2026 - a concentrated Neocloud ETF tied to GPU-as-a-Service, high-density AI datacenters, and AI networking. Likely holdings, the private capacity leaders, comparable ETFs, and what to track before launch.
After SpaceX: the 2026 IPO pipeline, ranked by what's actually filed
SPCX priced at $135 and closed its debut at $161. The next-in-line list, ranked by filings instead of rumors: Anthropic's S-1 is in (window as early as October), OpenAI is a stage behind, and the Kraken/Revolut/Stripe tier is frozen, patient, or unhurried.
SpaceX IPO (SPCX): the $1.77T listing, what you're buying, and the space cohort
SpaceX prices the largest IPO ever on June 11 (SPCX, $135, ~$1.77T). What you're actually buying (launch + Starlink + Starship), the listing mechanics, and the public space names already trading the theme.
How to invest in the SpaceX IPO (SPCX), and what to buy if you can't get in
Most retail can't get SpaceX ($SPCX) at the $135 offer. The honest ways to get exposure: the open-market debut June 12, the index funds about to hold it, and the public space names (RKLB, ASTS, RDW) already trading the theme.
AI IPO lockup expirations: CoreWeave, Circle, Cerebras & Nebius dates
The lockup-expiration dates for the AI-era IPOs people actually search: $CRWV (expired Aug 14 2025), $CRCL (Dec 2 2025), $CBRS (~Nov 2026, the live one), and why $NBIS has no conventional lockup at all.
Roundhill Photonics & Optics ETF ($LYTE): the first concentrated US bet on optical interconnect
Roundhill filed $LYTE on May 20 2026 - a concentrated Photonics & Optics ETF tied to silicon photonics, co-packaged optics, and AI optical interconnect. Likely holdings, comparable ETFs, and what to track before launch.
Cerebras (CBRS) IPO: the first pure-play inference bet hits the tape
Cerebras (CBRS) opened +68% on its $5.55B IPO. Wafer-scale silicon vs NVIDIA, 86% UAE customer concentration, $24.6B backlog. The bull and bear cases.
Other research(14)
Pieces that don't fit a category yet.
GE Vernova (GEV) - what it does, how it makes money, and the AI power-crunch bet
GE Vernova carries ~25% of the world's electricity on its installed base and books gas turbines years out. Its three segments, the datacenter-power bottleneck, and the bull and bear cases.
The Iran war reaches the AI trade through the power bill, not the panic
US-Israel strikes on Iran pushed Brent up 65% and shut the Strait of Hormuz. The AI-name risk-off is noise - the lasting channel is datacenter-power economics and a defense rotation.
Adobe (ADBE) Q2: AI-first ARR triples past $500M, FY26 guide raised
Adobe printed $6.62B revenue (+13%) and $5.96 non-GAAP EPS, both above guide, with AI-first ARR tripling past $500M. The print pushes back on the AI-displacement de-rate that left it near 10x forward earnings.
Penguin Solutions (PENG): what SMART Global became, AI factories vs memory
Formerly SMART Global (SGH), up 4.5x in 2026 on AI-factory demand while revenue shrinks 6%. What Penguin Solutions does, how it makes money, and the gap between the tape and the income statement.
Broadcom (AVGO): what it does, how it makes money, and why it owns custom AI silicon
A $2.3T chipmaker that designs ~70% of the custom AI accelerators hyperscalers run instead of Nvidia GPUs. What Broadcom does, how it makes money, and where AVGO sits in the AI-compute bubble.
Boost Run (BRUN): what it does, the $940M backlog, and the financing gap
A three-week-public neocloud with a $940M backlog, a $1.44B Dell GPU bill, and $9.7M of cash. What Boost Run does, how the de-SPAC works, and why the $2.2B price is really a financing bet.
What is TSMC (TSM)? How it makes money, and why every AI chip runs through it
A plain-English guide to Taiwan Semiconductor (TSM) - the pure-play foundry that manufactures nearly every leading-edge AI chip (NVIDIA, AMD, Apple, the hyperscaler ASICs). What a foundry is, the ~60% gross margins, HPC/AI now 61% of revenue, the CoWoS packaging chokepoint, and why TSMC is the master bottleneck of the AI supercycle.
IBM - the quantum foundry bet, and what a $278B incumbent actually does
IBM repriced from $222 to $296 in eight sessions on a $2B US-government quantum deal. Here's what IBM actually does, how it earns $67.5B in revenue, and where it sits among the quantum pure-plays.
Applied Optoelectronics (AAOI) - 800G transceivers and the hyperscaler bet
AAOI revenue +51% YoY to $151M in Q1 FY26 on 5 sequentially-accelerating quarters; $12.4Bn mcap at 24.5x P/S. Amazon ~50% revenue concentration, +50% YoY share dilution. The AI optical transceiver story, the customer concentration risk, and where it fits in QA's networking-optical bubble.
SanDisk (SNDK) explained - NAND, the AI memory cycle, and the $42B backlog that's trying to break the boom-bust
SNDK has run from $36 to $1,590 in a year on the AI NAND supercycle. Q3 2026 print: $5.95B revenue (+27% beat), $23.41 EPS (+61% beat). The structural read: $42B in multi-year hyperscaler contracts are management's bet against the historical NAND cycle.
How to invest in Anthropic - the proxy basket (AMZN, GOOGL, VCX, ARKV)
Anthropic closed its $965B Series H in May 2026 - reportedly its last private round, with an IPO targeted for H2 2026. The proxy basket while you wait: Amazon's stake, Google's, Fundrise VCX, ARK Venture, and the sleeper - SK Telecom (SKM).
Dell Technologies (DELL) Q1 FY27 - AI Servers +757%, FY27 guide raised to $60Bn
DELL Q1 FY27: $43.8Bn revenue (+23% vs consensus, +88% YoY), non-GAAP EPS $4.86 (+65% vs $2.94 consensus), AI-Optimized Servers +757% YoY to $16.1Bn. $24.4Bn AI orders booked. FY27 AI server guide raised to $60Bn. The ISG inflection thesis: confirmed.
Cerebras (CBRS) - wafer-scale inference, S&P 500 fast-track, and the 86% UAE concentration
S&P 500 fast-tracked Cerebras 12 days after its IPO. Passive funds now own a $48.8B name with 86% UAE revenue concentration. The wafer-scale inference thesis and the customer risk you buy through SPY.
Huawei's Tau Scaling Law and LogicFolding - what China's sanction-proof chip pivot actually means
At the 2026 IEEE symposium Huawei announced the Tau Scaling Law: instead of racing TSMC down the lithography curve it can't access, China optimizes for signal-delay reduction via 3D logic stacking. 381 chips already produced. Kirin 2026 ships this fall. Target: 400M transistors/mm² by 2031. SMIC closed +7.6% on the news. Why this reframes - but doesn't yet break - TSMC's lock-in on AI compute.
Go deeper
One email a day.
Daily digest of bubble shifts, alerts, and new research. Free during beta.
No spam. One email per day max. Pro adds Telegram trade alerts and higher AI-assistant limits.