CPI · Wed 12 Aug · BLS
July CPI-U +0.1% m/m SA; +3.4% y/y. Core (ex food & energy) +0.2% m/m, +2.5% y/y. Energy −1.5% m/m. Source: BLS CPI, USDL-26-1378.
68/100
Greed in the physical layer
AI equity complex · live brief
This week’s prints confirm the bottleneck has moved from GPU availability to interconnect, memory, power and financing — but NVDA is not confirming the satellite melt-up, Cisco is being marked down for mix/margin, and the 5-day leaders are already pricing a clean Aug 26 NVDA print.
1–4 week call. Respect the physical-layer confirmation; do not pay the 5-day leaders as if Aug 26 were already a beat-and-raise. Relative: NVDA / AVGO / TSM / ALAB / MU vs SMCI / NBIS / CRWV / P / CRDO into the event. Fade strength in names +20% vs 50d unless a new primary print extends the BOM.
01 · Session stamp
The tape this morning
Prices from yfinance 1.6.0 / Yahoo Finance, pulled 13 Aug 2026 at 11:05am EDT. SPY is at its 52-week high (779.37). SMH remains well below its 52-week high (671.83). Session volume vs 20-day average is not usable yet — cash has been open about 95 minutes.
SPX · ^GSPC
7,797.70
+0.63% day · +1.1% 5d
QQQ
732.34
+1.19% day · +2.5% 5d
SMH
595.15
+1.76% day · +4.1% 5d
SOX · ^SOX
12,654.84
+2.06% day · +5.0% 5d
| Benchmark | Last | Day | 5d | 1m (~21d) | 3m (~63d) | vs 50d | vs 200d |
|---|---|---|---|---|---|---|---|
| S&P 500 | 7,797.70 | +0.63% | +1.1% | +3.0% | +4.8% | +3.9% | +10.3% |
| SPY | 777.42 | +0.64% | +1.2% | +3.0% | +5.0% | +3.9% | +10.7% |
| QQQ | 732.34 | +1.19% | +2.5% | +2.0% | +2.6% | +2.7% | +12.8% |
| SMH | 595.15 | +1.76% | +4.1% | +0.7% | +4.0% | +0.4% | +29.2% |
| SOXX | 559.67 | +2.39% | +5.1% | +0.8% | +6.0% | −0.7% | +35.3% |
| SOX | 12,654.84 | +2.06% | +5.0% | +2.1% | +5.3% | +0.1% | +33.6% |
PPI · Thu 13 Aug, 8:30am ET · BLS
Final demand unchanged m/m SA; +4.7% y/y. Final demand less foods, energy and trade services +0.4% m/m, +4.7% y/y. Goods −0.7% (energy −3.1%); services +0.2%. Source: BLS PPI, USDL-26-1380.
Macro is not fighting the AI bid. Core CPI is contained; PPI goods deflation is energy, not a capex-kill. The AI complex is trading idiosyncratic earnings, not a Fed shock.
Calendar
NVIDIA Q2 FY2027 call Wed 26 Aug, 2:00pm PT / 5:00pm ET (quarter ended 26 Jul 2026). Source: NVIDIA Newsroom.
02 · Universe
What changed in the 2026 landscape
Adds are verified listed names pulled into the same tape. PSTG is dead on Yahoo — it was renamed, not delisted.
Adds. CRWV — CoreWeave, Nasdaq, IPO 28 Mar 2025, pure-play neocloud. NBIS — Nebius Group, second listed neocloud. IREN, APLD — listed AI-host / power-adjacent peers (high-beta, not hyperscalers). SMCI — AI server OEM, trading as a demand-proxy this week. CIEN, GLW, CSCO — optical systems / glass / campus-to-DC networking pulled into the same tape. SNDK — Sandisk (memory/storage; trades off the WDC complex). P — Everpure (ex-Pure Storage). Ticker changed PSTG → P effective 17 Apr 2026 (company IR / PR Newswire).
Dropped. PSTG (renamed/retickered, not delisted).
Power context (not core table). VRT, CEG — grid/power is a binding constraint; tape today is not treating it as the leader.
03 · 72-hour evidence
What’s printing
Six independent reads in a week: financing duration, optical conversion, neocloud backlog, networking mix, fabric silicon, and a memory/HDD bid. This is a physical BOM confirmation, not a model-release headline.
Mon 10 Aug · NVIDIA Newsroom
NVIDIA × Wall Street — $500bn of MoUs, not a funded check
NVIDIA announced MoUs with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to “mobilize over $500 billion of third-party capital” for AI infrastructure over time. Partnerships remain subject to execution of final agreements. Reuters confirmed the same day. Huang (via Reuters/X) said NVIDIA has the option to backstop up to 25% / ~$125bn.
This is an addressable-financing target, not capital called. Credit is being asked to treat GPUs as long-duration collateral; residual-value risk is the whole debate (FT, 12 Aug). Demand-duration, not a funded check.
NVIDIA Newsroom, 10 Aug 2026Tue 11 Aug · Lumentum IR
LITE blowout — CPO lasers, in-rack optics, $1.25bn guide
FYQ4 ended 27 Jun: revenue $1,006.3m (+109.3% y/y, +24.5% q/q); non-GAAP GM 50.4%, operating margin 36.6%, EPS $3.23. GAAP net loss $7.16bn driven by a one-time non-cash $7.76bn loss on convertible-note equitization — not an operating miss.
Q1 FY27 guide: revenue $1.225–1.275bn, non-GAAP OM 39.5–40.5%, EPS $4.05–4.35. CEO: CPO lasers, first ELS-module order, NPO engagements; optics “starting to penetrate in-rack connectivity.”
Lumentum IR, 11 Aug 2026Wed 12 Aug · Coherent IR
COHR print then fade — copper-to-optical, then −2.6%
FYQ4 ended 30 Jun: revenue $2.05bn (+34% y/y); non-GAAP EPS $1.74. CEO: “AI datacenter architectures increasingly transition from copper to optical.” Q1 FY27 revenue guide $2.2–2.4bn (8-K summaries; confirm vs company tables).
Stock −2.6% this morning after the print — beat-and-fade, not a demand miss. High 21d realized vol (131%). If this is de-grossing not a miss, it can bounce; if the guide is “not enough,” it is a trap.
Coherent IR, 12 Aug 2026Tue 11 Aug · CoreWeave IR
CRWV — sold-out capacity, $104bn backlog
Q2 revenue $2,575m (vs $1,212m y/y). Backlog ~$104bn as of 30 Jun 2026, excluding >$25bn net new commitments in early Q3. FY26 revenue guide raised to $12.4–13.2bn; capex $35–39bn; year-end ARR $18.5–19.5bn (earnings-call transcript summaries of company remarks).
This is a capacity-and-financing story, not a demand story. Sold-out capacity is bullish for pricing and bearish for execution, power interconnect, and refinancing.
CoreWeave IR, 11 Aug 2026Wed 12 Aug · Cisco Newsroom
CSCO — AI order book arrived; non-GAAP gross margin did not
Q4 FY26 revenue $17.3bn (+18%); networking revenue $9.79bn (+28%). Hyperscaler AI infrastructure orders $4bn in Q4, $9.3bn FY26; company expects ~$7.5bn hyperscaler AI revenue in FY27 (vs ~$4bn recognized FY26). FY27 guide: revenue $72.2–73.4bn, non-GAAP EPS $5.05–5.11.
Non-GAAP total GM 66.3% vs 68.4% y/y. Inventories $5.69bn vs $3.16bn y/y. Stock −7.4% this morning. AI orders are not in dispute; GM and inventory are.
Cisco Newsroom, 12 Aug 2026Tue 4 Aug · Astera Labs · plus memory tape today
ALAB fabric ramp, and a memory/HDD bid this morning
Astera Labs Q2: revenue $392.4m (+104% y/y, +27% q/q); Q3 guide $540–560m. Scorpio fabric switches expected to become largest product family one quarter early. Stock is −7.4% vs 50d — the multiple de-rated into the print.
Memory/storage tape today: MU +5.0%, WDC +8.1%, SNDK +6.9%, STX +3.9%, P (Everpure) +6.2%. Spot-tightness / AI-HDD / HBM-adjacent bid, consistent with Cisco calling out component-cost pressure and with LITE/COHR converting optics. No primary Micron print in this window; treat the tape as a read-through, not a company event.
Astera Labs, 4 Aug 202604 · First principles · next 1–4 weeks
Bull case
Demand is converting, not just being guided
Three independent layers printed in 72 hours: optical components (LITE 2× y/y, guide +24% q/q at the midpoint), neocloud contracted backlog (CRWV $104bn + $25bn), and hyperscaler networking orders (CSCO $4bn / qtr). That is a physical BOM confirmation, not a model-release headline.
The constraint set has rotated
Interconnect + memory + power + capital. LITE’s CPO/NPO/in-rack commentary and Coherent’s copper-to-optical language match the industry claim that scale-up/scale-out optics, not GPU wafer starts, now bind cluster delivery. The MU/WDC/SNDK/STX bid is the memory/HDD expression of the same BOM. If that rotation is real, CRDO, LITE, CIEN, ANET, ALAB, MU keep relative bid into NVDA.
Off-balance-sheet capital is being built
The $500bn MoU complex exists because hyperscaler cash-flow duration and GPU useful-life duration do not match. If even a fraction funds before 26 Aug, it extends the duration of GPU offtake and is NVDA-bullish at the print. The point is to keep the capex cycle from choking FCF multiples.
Macro is a tailwind, not a tax
CPI 3.4% / core 2.5% and PPI unchanged reduce the odds of a hawkish surprise hijacking the NVDA event window. The complex is trading earnings, not a Fed shock.
NVDA itself is not extended vs the satellites
+8.9% vs 50d, +0.3% today, 21d realized vol 40% vs NBIS 202% / CRWV 144%. Positioning into 26 Aug is in the periphery, which historically squeezes the core if the print is clean.
Bear case
Financing is an MoU, and MoUs are a tell
NVIDIA is recruiting private credit to fund its customers and may backstop 25%. That is economically closer to vendor-financed duration than to organic hyperscaler FCF. Residual value on a 2–3 year GPU in a 7–10 year SPV is the credit question. A single poorly received term sheet, or a rating-agency note, hits CRWV/NBIS/SMCI first, then NVDA.
Cisco is the quality check
Orders +35%, AI book $9.3bn — and the stock is −7.4% because non-GAAP GM compressed 210 bps y/y and inventories nearly doubled. If the AI BOM is inflating component costs (memory, optics, substrates) faster than ASPs, the stack’s earnings power is not what the multiples imply. That is a 1–4 week problem into NVDA GM commentary.
Hyperscalers are not acting like unconstrained buyers
META −13.5% / 1m and −6.1% vs 200d; GOOGL −6.7% / 1m. MSFT is the exception (+25.6% / 1m, +20.8% vs 50d) and is itself stretched. If the buyers’ equities de-rate, the suppliers’ order books get marked for duration risk, not cancelled — but multiples compress anyway.
5-day moves have already spent the NVDA-beat premium
NBIS +40.5% / 5d, SMCI +43.7%, CRWV +29.3%, P +35.6%, CRDO +23.6%, IREN +22.7%. A in-line NVDA print is a mean-reversion event for those names even if the cycle is intact.
Leverage and delivery risk at the neocloud layer
CRWV guided $35–39bn 2026 capex against Q2 revenue of $2.6bn. Sold-out capacity is bullish for pricing and bearish for execution, power interconnect, and refinancing. Power names (VRT −4.8% / 1m) are not confirming an unconstrained build.
Event cluster
NVDA 26 Aug, then Jackson Hole (Fed symposium calendars place it immediately after). Two macro-adjacent events in 48 hours. Vol will be bid; crowded longs pay the variance.
05 · Fragile
Fast movers at risk
Fragility = large recent % move + news/event crowding + (high realized vol or extreme SMA distance). Prices as of 11:05am EDT. These are the names most likely to give back a 5-day melt-up on an in-line NVDA print.
SMCI
42.23
Server-OEM proxy on CRWV/LITE read-through. +57% / 1m, +36% vs 50d. 21d RV 127%. Crowded momentum into NVDA.
+12.3% day · +43.7% 5d
NBIS
266.60
+83% vs 200d. 21d RV 202%, 21d beta vs SPY 7.4. Earnings spike; no Yahoo beta (short history).
+2.9% day · +40.5% 5d
CRWV
110.34
Backlog print is real; $35–39bn capex and customer-concentration/financing risk are not priced as a 1-week story. 21d beta vs SPY 6.8. +43% / 1m, +21% vs 50d.
+2.4% day · +29.3% 5d
P Everpure
118.32
+67% / 1m, +52% vs 50d. Storage-AI re-rate on a name that just changed ticker. Easiest mean-reversion in the storage sleeve.
+6.2% day · +35.6% 5d
CRDO
284.78
+19% vs 50d, +67% vs 200d. Optical/SerDes momentum piggyback on LITE. Yahoo beta 3.23.
+6.2% day · +23.6% 5d
IREN
46.59
Yahoo beta 4.30; 21d RV 161%. Power/host beta, not a hyperscaler.
+6.7% day · +22.7% 5d
LITE
939.31
Fundamental print is clean. Still +15% vs 50d, +49% vs 200d, 52w range 111–1,086. Any “sold-out lasers” disappointment vs the $1.25bn midpoint is a 10% tape.
+0.7% day · +12.1% 5d
CSCO
114.72
Fragility already expressing. AI orders are not in dispute; GM and inventory are. Further downside if NVDA/MU commentary validates component inflation.
−7.4% day · −5.1% 5d
COHR
346.27
Beat-and-fade the morning after. High 21d RV (131%). If this is de-grossing not a miss, it can bounce; if guide $2.2–2.4bn is “not enough,” it is a trap.
−2.6% day · +3.7% 5d
AAOI
140.98
+29% / 1m but −37% / 3m. High-beta optical (Yahoo 3.79, 21d beta vs SPY 8.4). Whippy, not a core holding into NVDA.
+2.1% day · +13.5% 5d
06 · Stretch vs cheap
Mean reversion
SMH is only +0.4% vs 50d but +29% vs 200d: the semi complex has digested the 50-day and is not a short-term stretch. The stretch is in single-name satellites.
Stretched · rich vs own 50d / 200d or vs the stack
200d extremes (cycle-long, not 1–4 week signals by themselves): MU +74%, STX +67%, CRDO +67%, SNDK +60%, MRVL +63%, AMD +53%, ALAB +53%, INTC +51%, ARM +48%, LITE +49%. These are bull-market distances; they fail if NVDA GM/guide disappoints, they do not mean-revert on a quiet Tuesday.
Lagging · cheap vs the stack, not necessarily vs history
META −13.5% / 1m, −6.1% vs 200d — buyer-multiple compression, not a GPU-supply story. GOOGL −6.7% / 1m, −2.3% vs 50d. ALAB −7.4% vs 50d after a 104% growth print and $550m midpoint guide — multiple already reset; less fragile than CRDO/LITE into NVDA.
WDC −11.4% vs 50d, SNDK −13.4% vs 50d despite today’s +8% / +7% bounce. Storage was sold; today’s bid is a squeeze, not a completed re-rating. AMD −3.4% vs 50d, ARM −8.6%, MRVL −4.5%, INTC −3.6% — the XPU complex is not participating in the 5-day optical/neocloud melt-up. That divergence either closes on a strong NVDA (catch-up) or confirms NVDA share / custom-ASIC fears (lag persists).
NVDA itself is the relative laggard of the week (+0.3% day, +2.6% / 5d) despite the $500bn headline. That is the core vs periphery split.
07 · High beta
What the beta is actually betting on
Yahoo info.beta where available; 21d realized beta vs SPY computed from yfinance daily returns (last 21 overlapping sessions). SPY 21d realized vol ≈ 13.8% ann. NVDA realized β has compressed vs NBIS / CRWV — the market is parking the variance in the new book.
| Ticker | Yahoo β | 21d β vs SPY | 21d RV ann. | What that beta is underwriting |
|---|---|---|---|---|
| APLD | 5.77 | 5.82 | 114% | Power/host leverage; not GPU cycle purity. |
| IREN | 4.30 | 6.69 | 161% | Same sleeve; a rates/power/crypto-AI hybrid. |
| ARM | 3.91 | 4.07 | 96% | Royalty duration on AI units; de-rates if custom silicon slows ARM attach. |
| ALAB | 3.84 | 5.67 | 119% | PCIe 6 / Scorpio fabric attach per GPU. High because the product ramp is binary-feeling at this multiple. |
| AAOI | 3.79 | 8.37 | 152% | Small-cap optical; crowding + thin liquidity. |
| CRDO | 3.23 | 4.67 | 105% | SerDes / AEC / optical DSP attach. |
| AMD | 2.49 | 3.79 | 82% | MI-series share vs NVDA; currently not getting the satellite bid. |
| MRVL | 2.25 | 4.76 | 94% | Custom / DSP / electro-optics. Realized β > Yahoo β — tape is treating it as an optical/AI high-beta. |
| INTC | 2.24 | 4.24 | 83% | Foundry + Gaudi optionality; 3m still −12%. |
| NVDA | 2.22 | 1.87 | 40% | Realized β has compressed vs Yahoo β. The market is parking the variance in CRWV/NBIS/optics, not in NVDA. That reverses on a miss. |
| MU | 2.21 | 4.36 | 107% | HBM tightness. Realized β >> Yahoo: this is now an AI-cycle name, not a PC-DRAM name. |
| WDC | 2.22 | 3.12 | 108% | Nearline HDD / AI storage. |
| COHR | 2.11 | 5.79 | 131% | Photonics; realized β is event-vol, not structural 2.1. |
| AVGO | 1.47 | 2.17 | 41% | Custom XPU + merchant networking. Lower vol, still the silent compounder. |
| AMZN | 1.45 | 1.96 | 64% | AWS capex + Trainium. |
| P | 1.41 | n/a | — | Yahoo β understates the 1m move; treat as high-idiosyncratic, not high-index-beta. |
| GOOGL | 1.24 | 1.75 | 50% | TPU / GCP / Search. |
| META | 1.24 | 0.61 | 44% | Realized β has collapsed — stock is trading company-specific (capex ROI / cloud rumors), not as an SPX high-beta. |
| MSFT | 1.10 | 2.49 | 60% | 1m +26% pulled realized β up; Azure/OpenAI duration. |
| CSCO | 1.00 | 0.84 | 39% | Index-like until yesterday; today’s −7% is idiosyncratic GM. |
| CRWV | n/a | 6.82 | 144% | No Yahoo β (IPO 2025). This is the high-beta expression of GPU offtake + leverage. |
| NBIS | n/a | 7.38 | 202% | Same. Highest realized vol in the universe this month. |
| SNDK | n/a | 6.67 | 155% | Memory spin; trades like a levered MU. |
Positioning implication. The “AI high-beta” book that used to be NVDA is now NBIS / CRWV / IREN / AAOI / COHR / MU. NVDA is trading like a mega-cap (RV 40%, 21d β 1.9). A vol-expansion event on 26 Aug hits the new high-beta book first.
08 · Compact universe
The book, in one table
Last / day% = yfinance regularMarket* at 11:05am EDT. 1m% = 21 trading-day total return on adjusted close. Beta = Yahoo info.beta (n/a if missing). 50d dist = last vs computed 50d SMA. Swipe horizontally on a phone; ticker column stays put.
| Ticker | Last | Day% | 1m% | Beta | vs 50d |
|---|---|---|---|---|---|
| Hyperscalers / platforms | |||||
| MSFT | 496.89 | +0.9% | +25.6% | 1.10 | +20.8% |
| GOOGL | 345.93 | +0.7% | −6.7% | 1.24 | −2.3% |
| AMZN | 267.57 | +0.1% | +4.9% | 1.45 | +7.9% |
| META | 589.22 | +1.8% | −13.5% | 1.24 | −1.4% |
| Neocloud / AI cloud | |||||
| CRWV | 110.34 | +2.4% | +43.1% | n/a | +21.4% |
| NBIS | 266.60 | +2.9% | +33.7% | n/a | +20.2% |
| IREN | 46.59 | +6.7% | +21.6% | 4.30 | +2.6% |
| APLD | 31.34 | +0.6% | +8.0% | 5.77 | −8.9% |
| SMCI | 42.23 | +12.3% | +57.0% | 1.97 | +36.0% |
| XPUs / accelerators / foundry | |||||
| NVDA | 224.65 | +0.3% | +5.7% | 2.22 | +8.9% |
| AMD | 494.07 | +2.3% | −6.7% | 2.49 | −3.4% |
| AVGO | 423.95 | +1.9% | +7.5% | 1.47 | +8.1% |
| TSM | 433.53 | +1.0% | +3.3% | 1.26 | +1.9% |
| ARM | 287.03 | +5.6% | +3.6% | 3.91 | −8.6% |
| MRVL | 228.76 | +5.4% | +11.0% | 2.25 | −4.5% |
| INTC | 105.68 | +4.7% | +2.6% | 2.24 | −3.6% |
| Optical / photonics | |||||
| COHR | 346.27 | −2.6% | +15.7% | 2.11 | +2.0% |
| LITE | 939.31 | +0.7% | +24.9% | 1.51 | +15.2% |
| AAOI | 140.98 | +2.1% | +29.2% | 3.79 | +4.6% |
| CRDO | 284.78 | +6.2% | +25.7% | 3.23 | +19.3% |
| CIEN | 460.00 | +6.5% | +9.9% | 1.31 | +6.4% |
| GLW | 166.68 | −0.5% | −4.5% | 1.15 | −6.1% |
| Networking / fabric | |||||
| ANET | 208.53 | −0.9% | +21.3% | 1.61 | +20.2% |
| ALAB | 329.68 | +3.4% | −6.0% | 3.84 | −7.4% |
| CSCO | 114.72 | −7.4% | +2.6% | 1.00 | −2.4% |
| Storage / memory | |||||
| P (ex-PSTG) | 118.32 | +6.2% | +67.2% | 1.41 | +51.8% |
| WDC | 491.09 | +8.1% | −4.4% | 2.22 | −11.4% |
| STX | 912.40 | +3.9% | +10.2% | 2.10 | +2.7% |
| SNDK | 1,437.29 | +6.9% | −11.1% | n/a | −13.4% |
| MU | 957.16 | +5.0% | +5.8% | 2.21 | −0.6% |
| Power (context) | |||||
| VRT | 289.87 | +0.5% | −4.8% | 2.08 | −2.6% |
| CEG | 279.10 | +0.1% | +8.1% | 1.12 | +7.0% |
09 · Construction
How the 68 is built
68 = greed in optics/memory/neocloud, skepticism in quality-of-earnings (CSCO) and in the platform (NVDA lag). Weighted call lands in a 62–70 band.
| Factor | Weight | Score | Note |
|---|---|---|---|
| Physical demand conversion | 25 | 85 | Hard prints (LITE/COHR/CRWV/CSCO orders), not vibes. |
| Breadth vs NVDA confirmation | 20 | 45 | Satellites bid, NVDA +0.3%, AMD 1m −6.7%. |
| Positioning / stretch | 20 | 35 | 5d melt-ups, SMCI/P/NBIS extreme vs 50d. |
| Financing quality | 15 | 50 | $500bn MoU is duration-support and a credit tell. |
| Macro (CPI/PPI) | 10 | 70 | Not a hindrance; core contained. |
| Event risk (NVDA 8/26) | 10 | 40 | Variance will be paid by the new high-beta book. |
| Weighted | ~62–70 | Call: 68 |