Private · Directions memo
Contrarian and early, in things you already read better than the tape. Buy what the crowd has left, be first to the nascent thing, and let the technical or consumer read do the work.
The frame
AMD near bankruptcy — a hated tech name you understood, held through the turn. Overstock — a distressed brand you read right when the market wrote it off. pump.fun — early to the nascent thing before it was a narrative.
The pattern under all three: early is the identity, contrarian is the engine, and understanding the technology or the consumer is the edge. Everything below is picked to that shape. Nothing here is oil, gas, hard assets, a discounted private-stake wrapper bought for yield, or an old revenue-coin you could name in your sleep. If a line reads like something the crowd is already chasing, it does not belong on this page.
Order is deliberate: crypto first and in the most detail, because that is where being early pays the most and where you want to be impressed, not managed.
01 · Crypto — new & early, in the trenches
No old coins. Every name is pre-TGE and farmable now, or a token that just went live and has not priced its story. Where the earliest seat is work (points, testnet, a node) I say so, because that work is the edge. Confidence is flagged honestly per name.
High on narrative Medium on which token wins
The crowd's AI-crypto trade is agents and inference — Virtuals, Bittensor, "AI x DePIN." That is consensus. The pre-consensus layer sits one step upstream: frontier models trained across permissionless GPU swarms, and the resulting weights owned as a cash-flowing asset. In 2026 this stopped being a whitepaper. Prime Intellect shipped INTELLECT-2 (32B, the first globally distributed RL run), Nous ran Hermes on its Psyche network, Pluralis proved out weight-durability. The line that writes itself next: you can own a piece of an open frontier model the way you owned a piece of a protocol in 2020. That framing is not mainstream yet. This is the closest thing on the board to DeFi summer before the summer.
| Name | Status | Earliest seat | Read |
|---|---|---|---|
| Nous Research | pre-token | Farm the Psyche node / testnet incentive program now | Highest "be early" score. $50M led by Paradigm at ~$1B private. |
| Pluralis | pre-token | Run a node, register the role ahead of TGE | Solving weight-extractability so swarms are durable. |
| PRIME | live (ETH) | Liquid handle — buy the leader, less asymmetry | Clearest technical leader (INTELLECT-1 and -2 shipped). |
| Gensyn | TGE Apr 2026 | Watch the post-unlock washout as the entry | a16z-backed, on-chain-verifiable RL compute. |
Honest risk. Distributed training can stay technically real yet commercially thin if centralized clusters keep winning on cost and speed. Token accrual to model weights is unproven in law and mechanism — the tech is real; the token capture is the open question.
Medium · high-variance, genuinely pre-attention
DePIN as decentralized wireless and maps (Helium, Hivemapper) is old consensus. The un-priced turn is machines — robots, drones, autonomous hardware — holding wallets, earning, paying, and being co-owned on-chain. It is the crypto tie-in to the humanoid-robot story the equity market already pays up for (Optimus, Figure), and crypto's version is nowhere near priced.
Beaten-down price, a real ecosystem. This is the clean AMD/Overstock shape in crypto: a category you understand, a hated price, and a turn you can see before it arrives. Buy it as the cheap infrastructure of a narrative that has not landed.
Identity, payment and coordination rails for robots — one of 2026's most-anticipated DePIN launches. A genuine get-in-at-TGE candidate, sized like a call option.
Honest risk. Timeline risk is severe. Robots-with-wallets could be a 2028 story and these tokens can bleed 80% waiting. Size like an option, not a position.
Medium-high on the theme · trench-level per name
The cleanest "be early, farm real usage, get a real airdrop" trade in the market. A pack of new perpetual venues are doing genuine volume with points programs and, in most cases, no token yet. This is the contrarian bet against the crowded leader (the fade below): the challengers are quietly eating share. The airdrop meta has moved to points earned by real on-chain activity — volume, open interest, liquidity — which also means wash-farming is rife, so the game is generating organic flow, not looped noise.
| Venue | Angle | Token | Earliest seat |
|---|---|---|---|
| Ostium | Arbitrum perp for synthetic real-world assets (commodities, FX, pre-IPO); RWA × perps | OSTIUM | Trade real size; under the radar vs the leader |
| Paradex | Starknet appchain, unified spot/perp/options margin | pre (DIME) | Season 2 XP farming open — earliest of the seats |
| Lighter | Zero-fee ZK order-book perp | LIT — live | Season 3 (25% of supply) unannounced = live farm window |
| edgeX | Institutional-grade order-book perp | pre | Open Season points; ~20–35% of supply to users |
| Hibachi | Privacy-first perp (Arbitrum/Base), sub-10ms | pre | Small (~$204M vol) = better points math per dollar |
| GRVT | Licensed hybrid CEX/DEX — the "compliant challenger" | TGE Jul 2026 | Season 2 (180M tokens) still farmable post-listing |
Two pre-TGE farms sit alongside the basket and rate highest on pure asymmetry. Pacifica — Solana's #1 perp by volume, ~$100B cumulative, and the tell that matters: self-funded, no VC, the exact Hyperliquid setup, so no insider float dumps into your bid at TGE. Its anti-sybil engine retroactively slashes wash traders, which is the "distrust inflated volume" filter you want. Variational — Arbitrum derivatives, 50% of supply committed to the community and a smaller crowd, so early size counts for more; the points window closes by end of Q3 2026.
Honest risk & the base rate, out loud. Of 113 tokens that launched 2024–26 and cleared $100M cap, only ~7% traded above their launch price; median return was −95%. Being early to the farm is smart; blindly holding new tokens through unlocks is how you get destroyed. Ex-FTX leadership at Pacifica is real headline risk; perp volume can evaporate in a risk-off week.
Highest-conviction single view here
The most crowded long in crypto. At peak it held ~71% of decentralized-perp volume — the trade is fully discovered and universally loved. The tape is already rolling over: down ~28% from the June 2026 ATH ($76.67 → ~$54 by late July), volume at the year's lightest, momentum negative. On the honest denominator it is ~52x FDV, and the business has never been tested through a 50%+ drawdown — it launched straight into a bull market.
The contrarian read is exactly the crack a tech-literate person sees before the crowd: consensus is most parked here, the second-wave challengers above are eating share, and "the leader is unproven through a downturn" is a real structural hole, not a slogan. Fade the consensus long; farm and hold the challengers. Softer adjacent fade: KAITO — the attention-token narrative already broke when X cut the API in January; that top is in.
High the rails matter Low there is a token to own
x402 revives the dead HTTP 402 "Payment Required" code into a machine-readable rail: an AI agent hits an endpoint, gets a 402, pays USDC over HTTP, gets the data or compute — no account, no API key, no billing cycle. By April 2026 it had done 165M+ transactions across ~69,000 agents, ~$600M annualized, with Coinbase, AWS Bedrock, Visa and Mastercard behind the standard. This is the first crypto rail whose primary users are software, not people.
The trap, said plainly. x402 is an open standard with no token and USDC is the settlement asset — "buy x402" is a category error. The biggest builders (Coinbase, Google, Visa) have no token either, so value may accrue entirely to them. The only tradeable expressions are thin facilitators; PAYAI is the largest by real volume on Solana, and it is small and unproven.
Honest risk. You could be right on the narrative and have nothing good to buy. Agent volume can be inflated by loops and self-dealing. This is a farm-and-watch, not a ticker.
02 · Distressed tech & consumer brands
A recognizable brand the crowd has left for dead, where you read the technology or the consumer better than the tape, and there is a specific turn a normal person can see before Wall Street re-rates it. Zero risk is stated where it is real.
Best risk/reward on the page
Why it's hated ~8x forward earnings, near the bottom of its range. The Street models branded checkout in permanent decline — Apple Pay and Shop Pay were supposed to eat the button — and pays nothing for the rest.
The turn you can see Branded volume stopped bleeding (+2% sequential, Q1); "Pay with Venmo" grew ~44% and went global across 90 markets; Q2 beat on both lines. The real optionality is agentic commerce: when an AI agent checks out for you it needs a trusted, tokenized wallet with a stored funding source and an existing consumer relationship. PayPal plus Venmo is ~400M funded accounts and a merchant button already embedded across the web — the default endpoint an agent plugs into. The market prices a declining PSP; a tech reader prices the installed wallet.
Honest risk. Could be a slow-melt terminal-value story if checkout share resumes falling — huge FCF and buybacks, but it can dead-money you for years.
The mispricing is a comprehension gap — the cleanest on the page
The 2024 app redesign broke customers' speakers, cost the CEO his job, and made "Sonos" shorthand for a botched company. The damage was software, and software is fixable — the hardware moat, patent stack and installed base survived. FY26 is healing: Q3 revenue $375M (+9%, accelerating), net income back to ~$30M, first positive Q2 adjusted EBITDA in four years, real buybacks, seven straight quarters of hitting guidance. The crowd conflates "the app was broken" with "the company is broken." Anyone who knows the difference sees a temporary, separable discount. Risk: thin hardware margins, discretionary spend, and Amazon/Google/Apple can undercut forever.
| More, ranked | Px | Conf. | The turn a tech reader sees |
|---|---|---|---|
| U Unity | ~$43 | Med | Blew itself up on the Runtime Fee; Vector (ML ad network) is the real inflection — Grow reaccelerating to ~50%, guiding to GAAP profit by Q4. |
| XYZ Block | ~$84 | Med | Cutting bitcoin fees to own the customer is smart; USDC now live inside Cash App — a 50M-user stablecoin wallet the market isn't paying for. |
| INTC Intel | ~$100 | Med thesis / low entry | The emotional archetype, but the bottom trade already ran (up from ~$18). Only own it for the second leg: a marquee external customer on 18A/14A foundry. Not contrarian-cheap anymore. |
| WOLF Wolfspeed | ~$38 | Low (4/10) | The genuinely distressed one — post-Chapter-11 equity, 200mm SiC fab, datacenter-power leverage. Real zero risk; size tiny. |
The cleanest Overstock analog on the board
Down ~95% from the 2021 mania; the market still tells the "fad bike, bleeding subs" story. What it is mispricing: the recurring-revenue engine under the declining hardware. Peloton just posted its first full-year net profit ($63M) and, more important, $378M of free cash flow, having already made the hard cut from growth-at-all-costs to a cash machine. Churn near 2% on a premium subscription is loyalty, not a fad. At a $2.4B cap that is ~6–7x FCF for a globally known brand, with a new CEO repositioning it as an AI health-and-wellness platform. Risk: subs keep declining and the platform pivot stalls — then it's a melting annuity with ~$1.3B debt. The bet is profitable-and-stabilizing beats declining-and-cheap.
| More, ranked | Px | Conf. | The turn a consumer reader sees |
|---|---|---|---|
| ETSY Etsy | — | Med-high | The most spreadsheet-provable one: Depop/Reverb divestitures ($1.4B cash) masked an Etsy-only core growing 7.5% at a 29% EBITDA margin. The mispricing is an accounting optical illusion. |
| BMBL Bumble | — | Med | Founder back, running a deliberate quality reset that depresses revenue on purpose while margins hit records. The bet: the swipe-free/IRL pivot answers what young users actually hate. |
| UAA Under Armour | ~$6 | Low-med | Right archetype (founder + distressed icon), painful-but-correct premium reset — but no inflection on the tape yet. Small and patient only. |
| GPRO GoPro | ~$0.68 | Low / option | Iconic brand, near-death, delisting notice, board running a strategic-sale process plus a pro-camera/defense pivot. Pure takeout optionality — buy like a call. |
Two traps to not confuse with the above: W (Wayfair) — right shape but the trade already ran ~30% in August, the crowd arrived. BYND (Beyond Meat) — famous logo, but near-zero gross margin and authorized shares proposed to 3 billion; that's a dilution spiral, not a turnaround. Recognizable ≠ investable.
03 · Overlooked industries
Screened against your filter: contrarian, early, in your wheelhouse. Where the crowd already came back, it's cut.
The pure-plays are the crowd's favorite hype trade, not an overlooked one. IONQ trades at ~90x sales, RGTI at ~500x on $4.4M of quarterly revenue, QBTS at ~$11B on bookings off a tiny base. These ran 5–10x on almost no revenue and a wave of retail plus government-funding headlines. That is narrative-priced — the opposite of contrarian-cheap. Buying here is chasing what the crowd is chasing, and it fails your core test.
The un-hyped adjacent that is overlooked: post-quantum cryptography. Unlike the computers, the migration is mandated on a clock — federal deadlines through 2030–35, only ~13% of organizations in production, ~60% not started, and "harvest-now-decrypt-later" means the spend starts before the threat lands. But be honest: the PQC pure-plays (ARQQ did $67K of half-year revenue, LAES, BTQ) carry the same narrative-over-revenue flaw. Own the theme through real businesses that bill for the migration — network and security incumbents embedding PQC (Cloudflare, Palo Alto, the HSM/PKI vendors) — and keep the pure-plays to a tiny lottery sleeve.
| Industry | Conf. | The setup & the turn |
|---|---|---|
| Independent ad-tech TTD · APP · PUBM · MGNI |
Med-high (cleanest fit) |
TTD down ~76% YTD, forward P/E collapsed ~150 → ~17. The open internet still needs one neutral buy-side that isn't also selling its own inventory; CTV budgets keep migrating off linear. A valuation reset on the category leader, not a broken business. The call: is the moat real or has Amazon's DSP cracked it — precisely your read. |
| App-layer SaaS after the "SaaSpocalypse" CRM · NOW · SNOW · IGV |
Med (highest tech-edge) |
$2T+ shed since January on "AI agents kill SaaS." The market painted every name with one brush. AI kills thin-UI-over-a-database; it rewards proprietary data, workflow lock-in and the governance layer agents run on top of. The discrimination between the two is the alpha — buy the moats, avoid the seat-priced thin tools. |
| Chinese tech BABA · BIDU · PDD · KWEB |
Med (window closing) |
AMD-near-bankruptcy in shape: earnings crushed on purpose (Alibaba's ~$56B AI/cloud capex), the AI buildout and robotaxi optionality near free. Cheap and hated — but the crowd is already trickling back, so buy it knowing the lonely part is ending. |
Don't mistake these for contrarian: biotech (XBI +25% YTD) and clean energy (+46% trailing) already ran — that trade was in April. And don't mistake edtech (CHGG, down 99%) for cheap — ChatGPT destroyed the product; a broken moat isn't a discount.
04 · Nascent tech to be early to
Early plus contrarian plus a technical read the market hasn't priced. Pre-attention categories with a dated catalyst that flips them from "nobody watching" to "narrative."
The enabler nobody owns, under the trade everybody owns
AI data centers are hitting a physical wall: copper between GPUs can't carry the bandwidth at acceptable power. The fix is moving the optics next to the switch chip — co-packaged optics — so signals travel as light. This is a bottleneck, not a fashion, and NVIDIA has said its next rack-scale platforms use it (Quantum-X early 2026, Spectrum-X photonics H2 2026). Everyone owns NVDA; almost nobody at retail owns the small firms making the optical guts NVIDIA now has to buy. POET (optical interposer, ~$5.6M production orders, Foxconn/Lumilens) is the clearest wheelhouse read — honest flag: it has already run and is story-priced on a productization ramp, not a cheap distressed name. LWLG is the earlier, lottery-ticket end.
| Vein | Conf. | The bet & the dated catalyst |
|---|---|---|
| Neuromorphic / ultra-low-power edge AI BRN (BrainChip) | High / binary | Chips that compute like neurons — AI at microwatts, on-device, no cloud. The contrarian inverse of the crowded data-center trade. Catalyst: AkidaTag to volume production Q3 2026. Highest variance here — it's been "about to matter" for a decade. |
| Prediction markets as a behavior shift | Medium | $50B+ volume around the 2026 World Cup, crushing sportsbooks, pulling in first-time and female users. Pure-plays are private; the public angle is the structural headwind to legacy sportsbooks, plus venture in the rails (market-making, settlement, compliance). |
| Ambient AI companion hardware | Speculative | Always-on wearable AI as emotional presence, not a tool (Friend, Omi, Plaud). Openly mocked — which is the contrarian tell. A pure taste call on whether the behavior sticks; venture-lane, lottery-sized. Own the ambient-capture platform layer, not the hero device. |
Deliberately not centered: smart glasses and humanoid robotics — the reads are right but both already hit inflection and the crowd has arrived. The only still-early sliver is the obscure component layer (waveguides, harmonic drives), not the obvious device.
05 · Where I'd point you
Early-crypto and distressed-tech-brand are where your edge is sharpest. In order.
Synthesis of your own trading history and independent research agents — not advice, and not a solicitation. Every figure here is a flag to re-verify, not a fact to trust: prices, market caps, volumes, TGE dates and points mechanics are as of roughly mid-August 2026 and this layer changes weekly. Confirm live quotes, token status, unlock schedules and liquidity the day you act. The crypto trench picks are high-risk: most new tokens fall ~95% from launch, farming pre-TGE guarantees no airdrop, and teams set the rules after the fact — the edge is farming where your cost is time, and selling into TGE hype rather than marrying the bag. Position sizes above are described as options and lottery tickets where that is what they are. Markets and method only.