⚖️ Both Beat. Both Raised. Twenty-Four Points Apart.
Palantir beat revenue by 5%, raised the year by $498M, and went up 15%. AMD beat revenue, beat EPS, doubled data centre, guided Q3 above consensus — and fell 9%. The difference isn't the quarter.
📊 The 30-Second Read
┌───────────────────────────────────────────────────────────┐
│ BEAT? RAISED? REACTION │
│ $PLTR YES YES +14.97% AH │
│ $AMD YES YES −8.94% AH │
│ ──────────── │
│ ~24 points apart │
│ │
│ PLTR: the acceleration was in the PRINTED NUMBERS │
│ AMD: the acceleration is still in the GUIDE │
└───────────────────────────────────────────────────────────┘
🔮 PALANTIR — What We Said, And What Landed
We said: “Palantir is going to beat. That’s not the trade. The event is the FY guide.”
That was right, and the internals show exactly why.
Revenue ............ $1.935B vs $1.841B est +5.1%
Adj EPS ............ $0.41 vs $0.35 est +17.1%
Revenue YoY ........ +93% · +19% QoQ
GAAP EPS ........... $0.41 (from $0.13)
GAAP net income .... >$1B — first time in company history
Rule of 40 ......... 155 (93% growth + 62% adj op margin)
FCF margin ......... 63%🎯 The guide was the event, and here’s the proof:
FY guide BEFORE .... $7.650B – $7.662B (71% growth)
FY guide AFTER ..... $8.150B – $8.158B (~82% growth)
─────────────────────────────────────────────────────────
THE RAISE .......... $498M
from Q2 beating its own guide ... $136M (27%)
from HIGHER H2 EXPECTATION ...... $362M (73%)🔍 Seventy-three percent of the raise is revenue Palantir hasn’t reported yet. Management didn’t just bank an outperformance — they marked up the back half. That’s a fundamentally different signal from a beat, and it’s why the stock moved on the guide rather than the EPS line.
📌 Q3 guidance came in at $2.160–$2.164B against a $2.038B consensus (+6.1%), and the new full-year range sits above the $7.859B the Street was carrying.
✅ The Number We Flagged — It Delivered
We said: “Government is appropriation-dependent and lumpy. Commercial is what a 71x sales multiple actually requires. Watch the US commercial line specifically.”
US commercial revenue YoY
Q1 2026 ............ +133%
Q2 2026 ............ +149% ← ACCELERATING
US commercial TCV .... $2.13B, +153% YoY — highest quarter ever
US comm FY guide ..... raised to >$3.42B from $3.22B
Deals ≥$1M ........... 220 (98 ≥$5M · 73 ≥$10M)🚨 Consensus expected US commercial to decelerate. It accelerated by 16 points. That was the single largest driver of the reaction, and it’s the line we said the multiple depended on.
⚖️ The bear notes, stated fairly — because they’re real: stock-based compensation grew faster than operating income on a quarter-over-quarter basis, which dilutes holders even without cash leaving. France’s domestic intelligence agency is reportedly replacing Palantir with a French rival, and a UK police contract has faced legal pushback. International remains the weak line.
⚪ The Call We Didn’t Make — And What It Cost Us
We passed on Palantir’s vol. Here’s the honest accounting.
Implied move ......... ~11.6%
REALIZED ............. +14.97% ($125.65 → ~$144.45)
ratio ................ 1.29x the impliedWe passed because two base rates were circulating — 7.39% and 17.80% — a 2.41x spread, and we refused to pick the one that produced a tidier call.
Against 7.39% ........ implied was 1.57x → RICH → would have LOST
Against 17.80% ....... implied was 0.65x → CHEAP → would have WON🚨 The figure we dismissed as stale was the more accurate one. We flagged the 17.80% eight-quarter median as fourteen months old and discounted it. It described this print better than the current-but-undefined “historical average” we couldn’t pin a window to.
📌 The pass cost nothing in the ledger. But the process has a lesson in it, and we’d rather log it than bury it: a dated figure with a stated methodology beat a current figure with none. Next time we hit a base-rate conflict, the tiebreaker should be which number tells you how it was measured — not which one is freshest.
🔥 AMD — What We Said, And What Landed
We said: “The straddle prices an ordinary day.” And: “The Q3 guide and the MI450 timeline are the event, not the Q2 revenue line.”
The second half was right. The first half didn’t get paid.
Revenue ............ $11.54B vs $11.28B est +2.3% RECORD
Adj EPS ............ $1.66 vs $1.61 est +3.1%
Revenue YoY ........ +50% · +13% QoQ
Non-GAAP GM ........ 56% · GAAP GM 54%
Non-GAAP op income . $3.1B · GAAP $2.0B
Non-GAAP EPS $1.66 vs GAAP EPS $1.38 — a 20% gap
Q3 GUIDE ........... ~$13.0B ± $300M (~41% YoY)
vs consensus ....... ~$12.5B (+4.0%)A beat and a raise. The stock rallied 7% in the session and then fell 8.94% after hours to $472.20.
🔑 The Number We Flagged — It Held, And Nobody Broke It Out
We said: “Data Center contains AI accelerators, EPYC server CPUs, GPUs, APUs, DPUs, FPGAs and Adaptive SoCs — all in one reported line. A strong Data Center print can be carried by either one, and the headline won’t tell you which.”
Data Center ........ $6.7B +107% YoY 58% of revenue
AMD's attribution .. "EPYC processors AND Instinct GPUs"📌 Still not broken out. The company credited both, in one sentence, exactly as anticipated. Pre-print analyst work put the split near $4.0B server CPU and $2.5B AI accelerator — implying roughly 62% of Data Center is CPU.
🚨 And Lisa Su said the quiet part on the record at the Advancing AI event in July:
“We’re expecting that the CPU ratio will actually be greater than 1. So maybe we get to the point where it’s 2 CPUs for 1 GPU.”
🔍 The CEO is explicitly guiding toward a more CPU-heavy mix — inside a multiple being paid for the GPU story. That’s not a criticism of the business. EPYC is an excellent, high-margin, share-gaining franchise. It’s an observation about what the reported number does and doesn’t tell you.
⚠️ A New Finding: The +107% Laps A Damaged Quarter
Buried in AMD’s own release:
Q2’25 results included $800 million in inventory and related charges due to the U.S. Government’s export control on AMD Instinct MI308 data center GPU products.
🧮 The 107% Data Center growth is being measured against a quarter that policy broke. Export controls on MI308 forced $800M of charges and blocked sales into China. That doesn’t make the growth fake — demand is clearly real — but it means the year-over-year percentage overstates the underlying improvement, and nobody quoting “data centre revenue more than doubled” is mentioning it.
⚖️ It’s the eighth time this cycle we’ve found the same shape: reg credits at Tesla, client incentives at Visa, price/mix at Coke, inventory drawdown at Boeing, a tariff gain at Ford, unbranded volume at PayPal, adjusted EBITDA at SpaceX — and now a comp flattered by a policy shock.
📋 Grading Our AMD Call
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
OUR CALL ......... CHEAP · 🟡 LEAN
IMPLIED .......... ±12.28%
RULE ............. correct >16.0% · wrong <8.6% · push between
─────────────────────────────────────────────────
Regular session ... +7%
After hours ....... −8.94%
STATUS ............ ⏳ PENDING — grade needs the Aug 5 close
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━📌 We measure close-to-close, so the after-hours print isn’t the grade. But we’re not going to pretend this is looking good. If roughly −9% holds through Wednesday’s close, it lands in the push band and the CHEAP call doesn’t get paid.
🧮 The thesis was that a 12.28% straddle was cheap against a tape printing 8–13% on ordinary days. The tape didn’t deliver the move it needed. We flagged the specific failure mode in the preview — “if last week’s volatility was a one-off panic that has now cleared, the four-quarter base rate is the right reference and this straddle is fairly priced” — and that appears to be roughly what happened.
⚖️ So Why Did One Go Up 15% And The Other Down 9%?
Both beat. Both raised. Twenty-four points of reaction between them. The quarters don’t explain that. The location of the growth does.
PALANTIR — the acceleration is in the printed numbers
US commercial ....... 133% → 149%, when consensus said slow
Rule of 40 .......... 155
FCF margin .......... 63%
GAAP net income ..... >$1B, first time ever
Capital required .... none. It's software.
AMD — the acceleration is in the guide
H1 2026 average ..... ~$10.8B/quarter
Consensus H2 needs .. ~$14.1B/quarter
Company language .... Data Center "accelerates in the second half"
Helios ramp ......... runs BELOW corporate gross margin
The +107% comp ...... laps an $800M charge quarter
Capital required .... enormous, and ongoing🎯 Palantir showed it. AMD promised it. In a tape that spent last week repricing AI financing rather than AI demand, showing beats promising by roughly twenty-four points.
📌 And this is exactly the filter we flagged in the week-ahead: AI demand gets rewarded, AI capex-dependent stories get punished. Microsoft +16% and Meta −10% last week were the same trade. Palantir sells enterprise software at a 63% free cash flow margin with nothing to build. AMD has to fabricate the product, finance the fab capacity, and ramp Helios at below-average margin before the revenue arrives.
⚖️ The honest caveat on that framing: AMD’s business is not in trouble. Fifty percent revenue growth, a record quarter, a Q3 guide 4% above consensus and a data centre business that genuinely doubled is a good outcome by any normal standard. It was punished relative to an extraordinary set of expectations, not relative to reality. Those are different things and the distinction matters if you’re holding it.
📒 The Ledger
TICKER OUR CALL TIER RESULT
$PLTR — ⚪ PASS no grade (by design)
$AMD CHEAP 🟡 LEAN ⏳ pending Aug 5 close
tracking toward PUSH📌 A note on the pass. We’ve now passed three times in this cycle — UPS, Palantir, SpaceX — each for a different reason: no edge, no usable base rate, and no history at all. Palantir is the first one where passing visibly left money on the table. We’d still take the pass, because picking between two contradictory numbers to justify a call is how a ledger stops meaning anything. But the lesson about which number to weight is real and it’s logged above.
⚠️ Caveats
📋 Market commentary and education — not financial advice.
🔷 Disclosure repeated: Anthropic is reported among AMD’s customers; Anthropic builds the model used to produce this research. No positions.
⏳ The AMD grade is not final. Our rule measures close-to-close (Aug 4 close → Aug 5 close). The −8.94% after-hours figure is indicative only. We’ll post the settled grade once Wednesday closes.
🔢 Consensus figures vary by vendor. PLTR revenue is measured against $1.841B (ChartMill); other sources used $1.812B, which would make the beat 6.8% rather than 5.1%. AMD is measured against $11.28B and $1.61. The PLTR EPS beat is 17.1% against $0.35 or 20.6% against $0.34.
📊 The Palantir guidance decomposition ($136M from Q2 outperformance, $362M from raised H2 expectation) is third-party analysis, not a company disclosure. The 73%/27% split is our arithmetic on those inputs.
🧮 The AMD Data Center split (~$4.0B CPU / ~$2.5B accelerator) is a pre-print analyst estimate reported by MarketWatch, not a company-reported figure. AMD did not break the segment out. Treat the 62% CPU share as indicative.
📉 AMD’s H2 requirement (~$14.1B/quarter) derives from MarketScreener consensus of $12.5B Q3 and $15.7B Q4, published before the print. AMD’s actual Q3 guide of ~$13.0B is above the Q3 portion of that.
⚖️ We grade our vol calls win or lose, and we publish the ones that go against us with the same prominence as the ones that don’t.
📬 Both beat. Both raised. One went up fifteen and one went down nine. The quarter was never the variable. Reply with your read.