Too early to call: only ~1.3 independent months of non-overlapping evidence (28 closed days, 2026-06-09 → 2026-08-03). The researcher is behind by 4.4 pts/pick so far - but that's still inside the noise. Watch this as more picks mature.
Post-close 9/11/2026: AI capex is still the tape's engine (TSMC record monthly revenue, Broadcom's multi-year doubling guide, Nvidia near an all-time high), while a broad healthcare selloff and a Mideast oil spike add crosscurrents. Today's picks favor verified, dated forward legs over already-moved headlines, which is why the +26% one-day oil trade and the muddled Lilly reaction are left out.
regime:chop· benchmark efficiency ratio 0.06· mean-reversion publishes, momentum is benched
Not investment advice. Racks Picks is automated research - an AI analyst plus mechanical screens (GBM Monte Carlo, Black-Scholes implied vol, half-Kelly sizing) over public data. Models assume the future resembles the past; it often doesn't. Positions are never held; the scoreboard below tracks every past pick, including the losers.
#1Broadcomdeep research
$332.13+3.1% today▼ -8.0% since pickconv 8/10
Shares fell about 14.6% to roughly $361 after a fiscal Q4 revenue guide of $34.8B landed less than 1% below consensus, an overreaction that ignores an upgraded multi-year AI outlook. CEO Hock Tan guided AI revenue to roughly $115B in FY2027 and $230B in FY2028 (doubling each year) with FY2026 AI revenue lifted to about $58B, and Q3 AI chip sales already jumped 221%. That is a stock punished on a rounding-error miss while the core AI thesis got bigger.
Click for full thesis, catalyst, risk & the quant card →
horizon: monthsSemiconductors / AI
ThesisShares fell about 14.6% to roughly $361 after a fiscal Q4 revenue guide of $34.8B landed less than 1% below consensus, an overreaction that ignores an upgraded multi-year AI outlook. CEO Hock Tan guided AI revenue to roughly $115B in FY2027 and $230B in FY2028 (doubling each year) with FY2026 AI revenue lifted to about $58B, and Q3 AI chip sales already jumped 221%. That is a stock punished on a rounding-error miss while the core AI thesis got bigger.
CatalystFY2027 AI-revenue doubling guide plus custom accelerator and networking ramp with hyperscaler and AI-lab customers (including a custom-silicon program with OpenAI); next hard print is fiscal Q4 in December 2026.
RiskCustom-silicon revenue is concentrated in a few customers whose capex timing can slip, and a low-20s forward multiple leaves little cushion if the FY2027 doubling gets pushed out.
#2Revolution Medicinesdeep research
$194.23+3.2% today▼ -4.5% since pickconv 8/10
The FDA approved RASONQUE (daraxonrasib), the first broad RAS-targeted medicine, for metastatic pancreatic cancer, and did so months ahead of schedule on data showing it roughly doubled survival. This is a genuine first-in-class launch into a cancer with essentially no effective targeted options, giving Revolution a durable franchise rather than a one-day headline.
Click for full thesis, catalyst, risk & the quant card →
horizon: monthsBiotech
ThesisThe FDA approved RASONQUE (daraxonrasib), the first broad RAS-targeted medicine, for metastatic pancreatic cancer, and did so months ahead of schedule on data showing it roughly doubled survival. This is a genuine first-in-class launch into a cancer with essentially no effective targeted options, giving Revolution a durable franchise rather than a one-day headline.
CatalystUS commercial launch is now underway; forward legs include expansion into additional RAS-mutant tumors (non-small-cell lung and colorectal), pivotal RASolute readouts, and a pending EMA decision.
RiskThe stock trades near its 52-week high after roughly a 5x run off the lows, so the approval is largely priced; launch uptake, pricing, and real-world durability of the survival benefit now have to deliver.
#3Nvidiadeep research
$206.69+2.3% today▼ -5.3% since pickconv 7/10
Coming off a record fiscal Q2 (revenue +106% year over year to $96.2B) with a roughly $108B Q3 guide, Nvidia is widening its moat beyond silicon: it agreed to acquire model hub Hugging Face for about $13B and is investing $3.5B in MediaTek while pushing NVLink Fusion into custom accelerators. The stock sits just under its record high with the next product cycle still ahead.
Click for full thesis, catalyst, risk & the quant card →
horizon: monthsSemiconductors / AI
ThesisComing off a record fiscal Q2 (revenue +106% year over year to $96.2B) with a roughly $108B Q3 guide, Nvidia is widening its moat beyond silicon: it agreed to acquire model hub Hugging Face for about $13B and is investing $3.5B in MediaTek while pushing NVLink Fusion into custom accelerators. The stock sits just under its record high with the next product cycle still ahead.
CatalystFiscal Q3 earnings in mid-November 2026 (implying roughly 89% growth), the Vera Rubin product ramp, and hyperscaler capex guided from about $800B in 2026 toward $1.3T in 2027.
RiskChina remains largely closed (DeepSeek is ordering 160,000 Huawei chips instead), and at a $5T-plus market cap any hint of capex digestion or a growth deceleration is punished hard.
#4Taiwan Semiconductor Manufacturingdeep research
$404.8+1.5% today▼ -5.4% since pickconv 7/10
August revenue hit a record NT$514.8B, up 53.3% year over year and 10.1% month over month, the fourth straight monthly increase, with eight-month revenue up 39.3%. As the sole leading-edge foundry for Nvidia, Broadcom, and Apple, TSMC converts the AI capex boom directly into its own top line, and it is the cleanest way to own the trend without single-product risk.
Click for full thesis, catalyst, risk & the quant card →
horizon: monthsSemiconductors
ThesisAugust revenue hit a record NT$514.8B, up 53.3% year over year and 10.1% month over month, the fourth straight monthly increase, with eight-month revenue up 39.3%. As the sole leading-edge foundry for Nvidia, Broadcom, and Apple, TSMC converts the AI capex boom directly into its own top line, and it is the cleanest way to own the trend without single-product risk.
CatalystQ3 2026 earnings and updated capex outlook in mid-October 2026, ongoing monthly revenue prints, and 2026 capex expanding toward roughly $64B.
RiskThe ADR sits near 52-week highs and carries Taiwan geopolitical headline risk; one soft monthly print or an AI capex-digestion scare can trigger quick profit-taking.
#5Energy Fuelsdeep research
$7.24-2.0% today▼ -46.9% since pickconv 5/10
Energy Fuels closed its roughly $300M acquisition of Australian Strategic Materials on August 28, building an integrated mine-to-magnet rare earth platform alongside its existing US uranium production. That positions it squarely in Western supply-chain reshoring just as China tightens exports of rare earths and chip materials, a policy tailwind the market is not yet paying for.
Click for full thesis, catalyst, risk & the quant card →
horizon: monthsMaterials / Critical Minerals
ThesisEnergy Fuels closed its roughly $300M acquisition of Australian Strategic Materials on August 28, building an integrated mine-to-magnet rare earth platform alongside its existing US uranium production. That positions it squarely in Western supply-chain reshoring just as China tightens exports of rare earths and chip materials, a policy tailwind the market is not yet paying for.
CatalystRare earth separation and magnet-metal production milestones from the combined platform, potential US government or defense offtake agreements, and uranium sales into a tight market.
RiskThe deal has already closed and the stock is still sliding (down about 8% on the month); this is a pre-scale, capital-intensive story exposed to weak rare earth prices and dilution, so it needs external policy or offtake catalysts to rerate.
On the radar - strong news signal, not picked today
The names with the most Racks-network news momentum that the researcher considered but didn't publish. Tap any to chart it - these are watchlist candidates, not picks.
The screens' records - and the gate
Walk-forward backtest, past year, today's 57-name universe: every 5 trading days the screen picks its top 5 (using only data available that day), equal-weighted, measured 21 trading days forward. A screen only publishes when its edge clears the gate - positive excess vs SPY whose studentized-bootstrap p survives a Benjamini-Hochberg correction across all five screens (false-discovery rate 10%). benched means it didn't, today. - the test windows overlap heavily, so the honest independent count is the eff figure (n·step/fwd), not n; the p-value already corrects for that overlap. The universe is news-selected (stocks get into the news by moving, which flatters these results) and no trading costs are modeled. The deep-research model has no backtest; its record is the live scoreboard below.
benched⚠ thin
benched⚠ thin
benched⚠ thin
benched⚠ thin
benched⚠ thin
⚠ thin
Scoreboard - every past pick
✓ = graded once at the pick's stated horizon, then frozen; unmarked picks are still open and track the live price.
◎ Truth Board — Racks Picks is one line in the live forward ledger. See how all four books grade together on /truth (ensemble · Racks/HODL · PRDCT/GMBL).