Tesla September seasonality analysis with multi-factor expansion

Expanding the commentary — TSLA, NVDA, AMD into week 37

Note on data vintage: prices are live intraday as of 04 Sep 2026, 14:18 ET. TSLA's last earnings call transcript available is the Q1'26 call (22 Apr 2026), so the "delivery mechanism" framing is from April, not from the July quarter.

---

1. The seasonal claim checks out — but it is one week wide, and the month around it is ugly

Seasonality computed from 01 Jan 2011, weekly buckets, 15 samples for week 37. Next week (07–11 Sep) is week 37.

Ticker	Wk 36 (current)	Wk 37 (next)	Wk 38	Sep month
TSLA	−1.8%, 44% up	+4.5%, 73% up	+2.6%, 67% up	+1.8% avg, 38% up (16y)
NVDA	−2.7%, 38% up	+3.0%, 67% up	−0.8%, 27% up	−0.3% avg, 63% up
AMD	−2.6%, 19% up	+2.6%, 67% up	−1.6%, 27% up	−4.4% avg, 25% up
SPY	—	—	—	−1.5% avg, 50% up

The TSLA figure you cited (+4.5%, ~75% win-rate) is exactly right. Three caveats that matter:

• Sample size is 15. One 20%+ week moves the average by ~1.4pp on its own. The median is not in the data set, and the gap between "73% positive" and "+4.5% average" strongly implies a fat right tail rather than a reliable grind higher.
• The window is one week only. For NVDA and AMD, week 38 immediately reverses to 27% win-rates. Week 37 in these names looks less like a "bright spot in September" and more like a mid-month bounce inside a bad month.
• September as a whole is genuinely weak for TSLA on hit-rate, not on average. 38% of Septembers were positive versus a +1.8% average — again outlier-driven, and the worst hit-rate of any month for the stock except October (33%). AMD's September is the cleanest negative: −4.4% average, 25% win rate, worst month of the twelve.

---

2. Technicals: none of the three is actually leading right now

Computed from daily closes; last price is live 04 Sep 2026.

Metric	TSLA	NVDA	AMD	SPY
Price	353.70	230.31	471.91	769.70
1-day	−6.02%	+0.81%	+3.45%	−0.45%
1-week	+1.42%	+5.86%	+1.36%	+0.05%
1-month	+8.05%	+8.66%	−9.00%	−0.21%
3-month	−15.47%	+5.33%	−9.80%	+1.67%
Since 03 Mar 2026	−9.87%	+27.90%	+147.14%	+13.13%

Moving averages and momentum (04 Sep):

• TSLA: below SMA50 (357.95) and 11.5% below SMA200 (399.56). RSI(14) 50.9. ATR(14) 15.27 = 4.3% of price. It is 16.8% below the 01 Jul close of 425.30.
• NVDA: 9.4% above SMA50 (210.57), 17.1% above SMA200 (196.68). RSI 60.4. The only one of the three in a clean uptrend.
• AMD: 5.5% below SMA50 (499.16) but 38.4% above SMA200 (340.97). RSI 48.0. Down 18.8% from the 30 Jun close of 580.91.

Relative strength percentile vs the S&P 500 (quarterly): TSLA 15.2, AMD 12.2, NVDA 52.0 (03 Sep). TSLA was at 80.8 on 10 Jul and bottomed at 1.4 on 12–13 Aug. AMD was at 99.4 on 15 Jul. So calling these three "tech leaders" going into next week is a backward-looking label — two of the three are in the bottom fifth of the index on a 3-month basis, and only NVDA is above median.

---

3. Fundamentals: TSLA's income statement is being deliberately dismantled

Latest reported quarters:

Metric	TSLA Q2'26	NVDA Q2'27	AMD Q2'26
Revenue	$28.2B (+25.5% YoY)	$96.2B (+105.9%)	$11.5B (+50.1%)
Gross margin	16.8%	75.0%	53.8%
Operating margin	1.41%	66.2%	17.3%
Free cash flow	−$1.09B	$21.4B	$1.6B
Capex / revenue	20.5%	2.8%	7.0%
ROIC (quarterly)	0.43%	33.8%	3.0%

TSLA's Q2'26 operating margin of 1.41% is the weakest of the last seven quarters (prior range 2.06%–6.16%), gross margin 16.8% is the second weakest, and it is the only negative FCF quarter in the set. Capex/revenue doubled from 11.1% in Q1'26. That is not a surprise — it is the plan management laid out.

Valuation versus industry peers (same-industry medians, latest quarter):

Metric	TSLA (auto mfg)	Peer median	NVDA (semis)	AMD (semis)	Peer median
P/E ttm	389.4	−1.25	26.1	134.1	−0.10
P/S ttm	16.03	0.96	16.49	20.61	4.98
EV/EBITDA	152.6	6.51	24.9	89.4	1.61
P/B	19.12	1.18	21.81	12.66	3.95

Three things jump out:

• TSLA is the maximum value in its entire industry on P/E, P/S, EV/EBITDA and (essentially) P/B simultaneously. Its P/S is 16.7x the auto-manufacturing median.
• AMD is now more expensive than NVDA on sales (20.6x vs 16.5x) while earning a 17.3% operating margin against NVDA's 66.2%, and its EV/EBITDA (89) is 3.6x NVDA's (25).
• NVDA's trailing P/E of 26.1 is the lowest of the last seven quarters (it was 49.4 a year ago) — earnings have compounded faster than the multiple. That is the single most under-discussed number in this trio.

---

4. Earnings track record and what's next

Ticker	Last report	EPS act/est (surprise)	Rev act/est (surprise)	Next report
TSLA	22 Jul 2026	0.33 / 0.44 (−25.0%)	28.2B / 25.2B (+11.9%)	21 Oct 2026
NVDA	26 Aug 2026	2.22 / 2.09 (+6.2%)	96.2B / 92.1B (+4.5%)	18 Nov 2026
AMD	04 Aug 2026	1.66 / 1.55 (+7.1%)	11.5B / 11.3B (+2.3%)	03 Nov 2026

TSLA has now missed EPS estimates in 3 of the last 7 quarters, including the most recent one by 25%, while beating revenue by 11.9% — the classic signature of revenue bought with margin. NVDA has beaten EPS in all 7 quarters shown (surprises +5.95% to +28.0%) and revenue in all 7. Consensus for NVDA Q3'27 is $109.0B revenue / $2.47 EPS; for AMD Q3'26, $13.0B / $1.87 (a step up from $11.5B).

Importantly: none of the three reports inside next week's seasonal window. Whatever week 37 does, it isn't earnings.

---

5. Segments and KPIs — where the money actually comes from

TSLA, quarterly:

Segment	Q2'26	Q2'25	YoY	% of Q2'26 rev
Total Automotive	20.5B	16.7B	+22.8%	72.7%
— of which Auto Sales	20.0B	15.8B	+26.6%	70.9%
— Regulatory credits	146.0M	439.0M	−66.7%	0.5%
Energy Gen & Storage	3.1B	2.8B	+10.7%	11.0%
Services & Other	4.6B	3.0B	+53.3%	16.3%

Your "$20 billion of traditional vehicles" is accurate — automotive sales revenue was $20.0B, 71% of the company. Two things sharpen the story:

• Regulatory credits, once pure margin, have collapsed to $146M from $595M in Q1'25. That is a −$449M annualised-margin hole the "delivery mechanism" has to absorb.
• Energy gross margin fell off a cliff: $640M on $3.1B = 20.6% in Q2'26, versus $952M on $2.4B = 39.7% in Q1'26. Management pre-announced this on the Q1 call — the 39.5% record included "one time benefits from certain tariff recognitions of more than 250 million." Bulls who annualised that 39.5% were annualising a tariff refund.
• Deliveries Q2'26 were 480,126 vs 384,122 (+25.0%), a record for the periods shown, and production 451,758 — deliveries exceeded production by 28,368 units, i.e. inventory drawdown.

NVDA, quarterly — the concentration is now near-total:

Segment	Q2'27	Q2'26	YoY
Data Center	89.0B	41.1B	+116.5%
— Hyperscale	48.7B	24.2B	+101.2%
— AI Clouds/Industrial/Enterprise	40.3B	16.9B	+138.5%
Edge Computing	7.2B	5.6B	+28.6%
Graphics	7.9B	5.4B	+46.3%
Supply & capacity purchase commitments	$279.0B	$45.8B	+509%

Data Center is 92.5% of NVDA's revenue. The number a commentator should be writing about is $279B of supply and capacity purchase commitments, up from $95.2B two quarters earlier and $45.8B a year ago — that is 2.9x NVDA's own quarterly revenue committed forward, and the most concrete evidence of the compute buildout anywhere in this data set. China (incl. HK) revenue also rebuilt to $7.9B from $3.1B in Q4'26.

AMD, quarterly:

Segment	Q2'26	Q2'25	YoY
Data Center	6.7B	3.2B	+109.4%
Data Center op income	2.1B	−155M	swing to profit
Client	3.1B	2.5B	+24.0%
Gaming	779M	1.1B	−29.2%
Embedded	977M	824M	+18.6%

AMD's Data Center went from a $155M operating loss to $2.1B of operating income in four quarters — that is the whole equity story. Gaming is shrinking and Client & Gaming operating income is flat-to-down (582M vs 767M a year ago).

---

6. Management's own framing (Q1'26 call, 22 Apr 2026)

The exact source of your "delivery mechanism" line — CFO Vaibhav Taneja:

> "With all this in mind, we have evolved our vehicle sales strategy where we now emphasize FSD as a product and vehicle as only the delivery mechanism."

On the capital intensity, Taneja again:

> "As Elon mentioned, we are in a very big capital investment phase which is going to start now and would last a couple of years. So based on that our current expectation for 2026 is over 25 billion of capex... While this may seem a lot and we will have the impact of negative free cash flow for the rest of the year, we believe this is the right strategy to position the company for the next era."

That guidance has already shown up: Q2'26 FCF was −$1.09B.

Musk, on the size of the bet:

> "I think Optimus will be our biggest product. Not just Tesla's biggest product ever, but probably the biggest product ever."

And on why Tesla is building a fab at all — this is the most striking framing on the call:

> "Terrafab is not some sort of mechanism to generate leverage over our chip suppliers. It's just literally we don't see a path to having enough, sufficient quantity of AI chips down the road as we scale production to high levels... we just anticipate hitting the wall if we don't make chips ourselves."

Also notable and rarely quoted — Musk actively deflating the Optimus ramp:

> "I don't know what the production rate of Optimus will be this year. It is impossible to predict these things... it'll move as fast as the least lucky, slowest, dumbest part in the entire 10,000."

And an admission with a real cost attached, on Hardware 3 cars: HW3 "simply does not have the capability to achieve unsupervised FSD," with Tesla offering "a discounted trade in" plus computer and camera replacement requiring purpose-built "micro factories."

Operational constraints management flagged: battery pack capacity is "our biggest limiter"; tariffs plus "sustained high interest rates"; interest-rate subvention costs recognised upfront; energy margin compression from competition and China-sourced cells; and $230M of one-time warranty true-down benefit inside the 19.2% ex-credit auto margin. FSD paid base was "nearly 1.3 million paid customers globally," with growth mostly from subscriptions — upfront purchases up only 7%.

---

7. Alternative data — the divergences are the story

News (last 3 weeks, TSLA). The Cybercab event on 03 Sep is being read two ways at once: the stock rose 5.4% into it on 03 Sep and is −6.0% today. Headlines flag Gary Black calling the launch "largely a bust," WSJ's Tim Higgins calling the promised "storm" a "drizzle" — and, materially more important than sentiment, NHTSA opened a probe into whether ~1,000 Cybercabs were properly certified (04 Sep). The Texas Cybercab fleet was 45 vehicles as of 01 Sep; Gene Munster expects 300 in Austin. Tesla sold 86,166 China-made vehicles in August. Cybertruck prices were raised >7% on 25 Aug "despite sluggish sales."

Analyst ratings. The dispersion on TSLA is extraordinary: StoneX Buy $475 (03 Sep), Stifel Buy cut $508→$491 (03 Aug), GLJ Research Sell $24.86 (18 Aug). That is a 19x spread between the highest and lowest target on a $1.4T company. NVDA saw ~20 reiterations after 27 Aug earnings, essentially all Buy, with Raymond James $352→$515 and Bernstein $315→$400. AMD's post-earnings targets ran $465 (Morgan Stanley, Hold) to $700 (Cantor, Wells Fargo, Rosenblatt) — note JP Morgan and Morgan Stanley both raised targets sharply while keeping Hold.

Insiders. NVDA director Mark Stevens sold $268.0M on 02 Sep, $14.0M on 01 Sep and $128.9M on 31 Aug — ~$411M in three sessions, right after earnings, on top of ~$407M in June. His holding fell from 6.90M to 3.96M shares since 02 Jun. EVP Timothy Teter sold $6.5M (31 Aug). AMD saw broad C-suite selling: Papermaster $13.6M (20 Aug) + $3.6M + $3.3M + $3.2M, Su $7.2M (18 Aug) after $57.6M in June, Hu $7.1M, Norrod $7.9M, Grasby $7.7M. TSLA insider selling was trivially small by comparison ($1.0M by Taneja); Musk converted 303,960,630 stock-appreciation rights on 16 Jun. Much of this looks schedule-driven, but the Stevens cluster is unusually concentrated immediately post-print.

Retail activity (FINRA-derived). TSLA retail sentiment has been negative on all 19 sessions from 10 Aug to 03 Sep (range −2 to −6) while the stock rose ~7% off the 10 Aug close. Retail was net-selling the rally. NVDA flipped from −3 (17 Aug) to +6 (25–26 Aug) around earnings and has cooled to +1. AMD retail participation is tiny (0.57%–1.52% of volume) and sentiment consistently negative.

Short volume. TSLA short volume rose from 5.8M (18 Aug) to 18.6M (31 Aug) and 16.2M (03 Sep) — the two heaviest days of the window landed on the two biggest up days (+5.5% and +5.4%). NVDA short volume peaked at 37.9M/35.3M on 27–28 Aug (earnings reaction) and has since halved to ~21M.

Dark pool (ATS, weekly). Data runs through the week of 27 Jul: TSLA 104.0M, NVDA 319.1M, AMD 78.3M. TSLA's off-exchange volume swung from 76.3M (13 Jul) to 136.7M (20 Jul) — the fortnight around the 22 Jul earnings miss. Nothing more recent is available.

13F (Q2 filings, filed Jul–Aug). Net institutional flow skewed to buying in all three, though these are largely index/custody-scale prints: TSLA saw 24 buys vs 6 sells above $500M, including BlackRock +6.0M shares ($2.5B) and Sixth Street +19.3M ($8.1B); against that, Viking Global cut to 547.8K shares and ExodusPoint exited (−4.9M / $2.1B). NVDA's tape was genuinely two-sided — Sanders Capital +15.4M ($3.1B), Van Eck +19.5M ($3.9B), FMR +32.2M ($6.4B) versus Capital World −21.1M ($4.2B), CalPERS −22.3M ($4.5B), Barclays −22.3M ($4.5B), Wellington −11.8M, Citadel −6.8M, Point72 down to 437.8K. AMD: T. Rowe +7.1M ($4.1B), Franklin +4.8M, Whale Rock +1.4M, Dragoneer new 944K; ARK sold 1.3M and D.E. Shaw sold 4.0M.

Congress. Small and directionally muddled. AMD saw 10 purchases vs 2 sales since April (mostly Cisneros, Gottheimer, Khanna) with reported gains ranging +68% to −25%. NVDA was net-sold, including Senator Whitehouse's two partial sales ($100–250K in May, $15–50K in June). TSLA: Khanna traded it six times in four months, alternating direction, with every disclosed position marked at a loss (−2% to −18%).

CNBC/TV sentiment. NVDA is the consensus long — 47 mentions since 01 Aug, of which only 3 were bearish/negative (Dan Nathan 18 Aug, Jenny Harrington 21 Aug). Bryn Talkington said she "bought more Tesla" on 31 Aug. AMD got two mentions: Cramer bullish 05 Aug, Cramer "Trim" on 31 Aug. Uniform TV bullishness on NVDA is itself a crowding datapoint, not a confirmation.

---

8. What the options market says about next week — it prices more than the seasonal edge

TSLA at 353.72, real-time OPRA chain:

Expiry	ATM ($350) call IV	ATM put IV	Implied move
09 Sep (5 DTE)	34.3%	35.3%	±4.0% (±$14.3)
18 Sep (14 DTE)	39.5%	39.7%	±7.7% (±$27.2)
16 Oct (42 DTE)	40.5%	40.2%	±13.5%

The seasonal "edge" of +4.5% is roughly one week's implied move. Options are pricing a ±4% range into 09 Sep, so the historical average week-37 gain is fully inside the noise band. That does not invalidate the pattern, but anyone treating +4.5% as an expectancy is comparing a 15-sample average to a distribution the market already prices as symmetric.

Positioning detail: the 18 Sep monthly has a call wall at $450 (21.0K OI) and $400 (21.8K), and a put wall at $300 (21.7K) and $340 (20.7K). Today's flow is put-heavy at the money — the 04 Sep $350 put traded 256,000 contracts versus 162,600 on the $360 call. Front-week skew is modest (5 DTE: 320-strike put 44.5% vs 380-strike call 44.5%), so there is no visible panic bid for downside beyond the ATM.

---

MOST NOTABLE / SURPRISING STATS

• NVDA's supply and capacity purchase commitments hit $279B, up 6.1x YoY from $45.8B and up from $95.2B two quarters ago — the hardest number in this data set on the scale of the compute buildout.
• NVDA's trailing P/E is 26.1, the lowest of the last seven quarters, while revenue grew 105.9% and operating margin hit 66.2%. Multiple compression amid triple-digit growth.
• AMD now trades at a higher P/S (20.6x) than NVDA (16.5x) with one-quarter of NVDA's operating margin, and at 89x EV/EBITDA versus NVDA's 25x.
• TSLA's regulatory-credit revenue collapsed to $146M from $439M a year ago (−66.7%) — the highest-margin line in the company shrank by two-thirds while auto sales grew 27%.
• TSLA Q2'26 operating margin was 1.41% and FCF was −$1.09B, with capex/revenue at 20.5% — management pre-committed to "over 25 billion of capex" and "negative free cash flow for the rest of the year."
• TSLA's energy gross margin fell from 39.7% to 20.6% in one quarter as a >$250M one-off tariff recognition rolled off.
• NVDA director Mark Stevens sold ~$411M of stock across 31 Aug–02 Sep, cutting his holding from 6.90M to 3.96M shares since June.
• TSLA retail sentiment was negative every single session from 10 Aug to 03 Sep while the stock gained ~7% — retail sold the entire pre-Cybercab rally.
• AMD's Data Center segment swung from a −$155M operating loss to +$2.1B of operating income in four quarters, on revenue up 109%.
• Analyst targets on TSLA span $24.86 (GLJ, Sell) to $491 (Stifel, Buy) — a 19x dispersion on a $1.4T market cap.

---

CONTRADICTIONS OR WEAK POINTS IN THE BULLISH FRAMING

• "Tech leaders" is not what the relative-strength data shows. TSLA sits at the 15th percentile and AMD at the 12th percentile versus the S&P 500 on a quarterly basis; only NVDA (52nd) is above median. AMD was at the 99th percentile in mid-July — that leadership has already broken.
• The seasonal edge is smaller than the options-implied noise. ±4.0% priced into 09 Sep versus a +4.5% 15-sample historical average.
• Week 37 is not "September strength," it's a one-week blip. NVDA and AMD both revert to 27% win-rates in week 38, and AMD's September is −4.4% average / 25% positive overall.
• "Tesla leads the pack" on seasonality but trails badly on fundamentals: lowest operating margin in seven quarters, negative FCF, a 25% EPS miss last quarter, ROIC of 0.43%, and the maximum valuation in its entire industry on four separate ratios.
• The AI/robotics optionality is explicitly unquantified by management. Musk on Optimus: "It is impossible to predict these things"; on robotaxi revenue: "will not be super material this year." The capex is contractual and immediate; the revenue is not.
• Today's actual TSLA catalyst is regulatory, not sentiment. NHTSA opened a probe into whether ~1,000 Cybercabs were properly certified — a harder issue than an "underwhelming" event, and one the seasonality data cannot see.
• NVDA's TV sentiment is near-unanimous (44 of 47 CNBC mentions bullish since 01 Aug) while 13F data shows genuine two-way institutional flow (CalPERS −$4.5B, Barclays −$4.5B, Capital World −$4.2B). The consensus is more crowded than the actual positioning.
• Dark pool data is stale — the most recent week available is 27 Jul, so nothing in that series speaks to the last five weeks.

This is research, not advice, and I can't make position decisions for you. Not trading advice.