SITM

As of 2026-10-20 · partly unblinded: the model was committed blind, meaning worked out before we saw the market's price, but a few facts leaked through the masking (details after the block). Revision 1 (same date, section 9) corrected seven model inputs after an outside review of the first draft; the blind value was $267.75.

We rate SiTime Avoid for new money at $494.71. Our central estimate of its value is $287.18 a share. That is 0.581 times the price, well under the 0.85 line between Avoid and Hold ($420.50). The bull case ($630.47) is the only case above the price: it earns +17.0% a year from here, 5 points above the 12% hurdle, and the central case +0.2% (both with the modeled free-cash payouts); the case-weighted value ($311.73) is 37% below the price.

Terms used below. Each report values a company three ways: the central case is our best estimate, the bear case a severe but plausible bad outcome, and the bull case the matching good outcome. EBIT is operating profit; NOPAT is operating profit after tax. The exit price is our estimate of the share price on 2031-12-31. The long fade values the company by letting its growth and its return on new capital fall gradually toward normal levels over 15 years; the outside view applies what comparable companies trade at today. "Stock pay" is stock-based compensation, which we count as a cost. The cost of equity is the yearly return shareholders require, used to bring future cash back to today; beta measures how far a stock moves with the market, and Blume-adjusted beta is pulled part of the way toward 1. "The Street" means the published estimates and price targets of Wall Street analysts. SiTime reports three markets: CED (communications, enterprise and datacentre), AIA (automotive, industrial, aerospace) and MIC (mobile, IoT, consumer). TPD is the timing business bought from Renesas.

What leaked through the blind step. The masked filings still showed the 2031 notes' implied conversion price (about $1,038), the share value inside the Renesas consideration (about $700 a share on 2026-07-01), and a CEO remark at a September conference about a "$20 billion market cap" (assumptions.md lists each). None of them is a model input. The model, forecast and assumptions were committed blind at 8d851c9e5 (first made as 249b951e8, rebased onto main unchanged) before we read the price, the Street's estimates or our research note on the timing-chip sector. The SiTime deal press release that states the share-price collar was never opened unmasked (process.md). The market pass (market.py, market.txt, odds.json) changed no model input. Revision 1 rests on a filing or a method rule in every case, and on the price or the Street in none.

1. SiTime must roughly triple its share of the timing market for our central case to hold

Every electronic system needs a clock: a resonator that vibrates at a fixed frequency and a circuit that turns it into a timing signal. For 70 years the resonator has been a cut quartz crystal, made by Japanese and Taiwanese specialists (Seiko Epson, Nihon Dempa Kogyo, Kyocera, Daishinku, Murata, TXC) and sold as oscillators by them and by Microchip, Rakon, Abracon and Diodes. SiTime makes the resonator in silicon (MEMS, a vibrating structure etched on a wafer) and pairs it with its own analog CMOS die. Its pitch is that silicon is more stable under heat and vibration, smaller, programmable at the end of the line (so one die serves thousands of part numbers with short lead times), and made on semiconductor tools rather than crystal-cutting lines. The 10-K names its competitors as the quartz makers above plus Microchip, Renesas, Skyworks and Texas Instruments (whose BAW resonators are the other non-quartz approach).

The decisive number is share. Management claims the timing market is about $11B growing 5-6% a year (Q4 2025 call). SiTime's 2025 revenue of $327M was about 3% of it; with the Renesas timing business (TPD) bought on 2026-07-01, the 2025 pro forma was $534M, about 5%. Our central 2031 revenue of $2,069M would be about 14% of that market grown at 5.5% (about $15B), or about 9% of the $22-25B management claimed for 2030 at a September 2026 conference (that larger figure adds resonators and new categories; the two claims do not reconcile, and we use neither as an input). So the central case is a company that roughly triples its share in five years, most of it in datacentre timing. That is the bet.

The cost side is simple. SiTime is fabless: Bosch and Teledyne make the MEMS wafers, TSMC and UMC the CMOS dies (10-K 2025; the blind assumptions.md wrongly put Teledyne on the CMOS side, which changes no input). Non-GAAP gross margin was 66% in H1 2026 and is guided to about 68% for Q3 with TPD in; GAAP gross margin was 52-57% in 2023-25 (FMP). Stock pay is large: $103.5M in 2025, 32% of revenue, against a peer median of 6.7% (the ten peers' latest fiscal years, FMP). The model charges it as a cost every year (11.5% of revenue in 2027 falling to 9% in 2031) and holds the share count at today's 30.07M (basic shares plus the Renesas shares; the unvested awards are not added because their cost is already the stock pay), so dilution is counted once.

2. SiTime has beaten its own revenue guide in each of the last seven quarters

SiTime has beaten its own revenue guide midpoint in each of the last seven quarters, by +6.4% to +11.7%, median +8.6% (releases, Q4 2024 to Q2 2026). Q2 2026 revenue was $157.4M, up 127% on the year: CED (communications, enterprise and datacentre) $101.2M, AIA (automotive, industrial, aerospace) $24.8M, MIC (mobile, IoT, consumer) $31.4M. The largest consumer customer (Apple, through the distributors Pernas and Quantek) was $22.8M of it.

The Q3 2026 guide is $285-295M: core SiTime $200-210M and TPD about $85M, gross margin about 68%, non-GAAP operating cost $80-85M. Our central Q3 is $314M, the guide midpoint plus the median beat (core $222M, TPD $92M). That is a measured pattern, not a judgment of management's intent; whether the guide is deliberately low is an inference we do not need. Q3 reports in early November 2026 and is the first test of this model (section 6).

The Q3 guide alone implies a non-GAAP operating margin near 40% before stock pay (68% gross margin, $82.5M of cost on $290M), above the 34% of Q2 and above management's "34% or higher" long-term target. Our central 2031 margin is 41% before stock pay and 32% after it, so the model assumes the margin SiTime already guides to, not an expansion.

3. The forecast gives a central value of $287.18 a share

The model (model.py, output in output.txt) grows four revenue lines and runs one cost base. Central path, US$ millions:

2026 is reported revenue, with TPD only from July; on a pro forma basis 2026 is $1,071M (+100% on the 2025 pro forma of $534M) and 2027 is +43%. The jump in 2027 is mostly a full year of the Q4 2026 run-rate ($1,332M a year): 2027 is only +15% on that run-rate. After 2027 the central case assumes the AI networking build slows to a digestion year in 2029 (a year of flat demand while customers work through earlier orders; CED +3%, TPD +2%), then a timing business growing 7-8%. The 2026 free cash is negative because TPD came with almost no working capital: the purchase accounting shows $2.7M of inventory and no receivables (8-K/A 2026-09-09 exhibit 99.2), so SiTime funds about $130M of receivables and inventory for it in H2 2026. Working capital is held at 26% of revenue, the ratio the filings measure (section 9, finding 7).

Valuation, central: free cash is counted at face as it is earned, the $1,350M of 2031 notes are valued today as debt (below), and the exit at 2031 year end is the method's mean of two views:

• Long fade, $306 a share. 14.8x year-six NOPAT, with growth fading from 10% to 3% over 15 years at an 11.52% cost of capital and returns falling to the peers' 29.8% median.
• Outside view, $586 a share. The method's outside view is the ten peers' median of 35.0x EV (enterprise value) over GAAP operating income (SiTime has one valid year of its own). We changed one thing: the peers' multiple is on GAAP EBIT after their own acquired-intangible amortization, while this model's EBIT is before SiTime's $200M of 2031 amortization from the Renesas and Aura deals. Each peer's ten-year median is restated by adding that peer's own amortization of intangibles, from its 10-K, back to each year's GAAP operating income; the median of the eight restated medians is 26.5x, which is applied to each case's own pre-amortization 2031 EBIT. The per-peer figures and their sources are in section 9 (finding 1). This is the only change from the default, recorded in Forecast.changes.

The two views are 92% apart, more than the 30% the method flags. They disagree because they price different things. The fade asks what 10% growth that slows to 3% is worth at an 11.5% cost of capital: not much more than 15x earnings, since growth near the cost of capital adds little unless returns stay high for decades (the value-driver formula at a constant 10% gives 43.8x, which shows how much rests on that slope). The outside view asks what buyers paid for analog and mixed-signal peers in 2016-2025, a decade of low rates and of Monolithic Power (MPWR) and MACOM (MTSI) trading near 60x. We see no reason to prefer either for a company whose 2031 position is this uncertain, so we keep the method's mean. On the fade alone fair value is about $208; on the outside view alone about $367.

Bridge to today (-$338.7M in all, -$11.26 a share): post-close cash of $321.9M beyond the Renesas payment and a $100M operating minimum; the 2031 notes, -$952.3M; the Aura earnout liability (-$81.4M); a $20M equity stake; and $353.1M for the tax the federal NOL pool ($344.3M) and the Section 197 deduction on the whole $4,006M Renesas price will save above the model's 15% normal rate (the NOL pool is the net operating loss carried forward; the Section 197 deduction lets acquired intangibles be written off for tax over 15 years). The tax shield is $11.74 a share. The notes are a bond with a conversion option, so they are valued as debt: $1,350M repaid at par on 2031-06-15, discounted at the risk-free rate (5.29%) plus 2.50%, the top of the secured revolver's SOFR + 1.75% to 2.50% grid (8-K 2026-07-01; the notes are unsecured, so 2.50% is a floor), which is 7.79% and $952.3M, not the cost of equity (which gave $813.1M). The conversion option net of the capped call (a hedge SiTime bought that offsets part of the dilution if the notes convert) is a named claim we do not value: it is a claim of the noteholders on the equity and reduces equity value, but the cap price is masked in the filings and the capped call is equity-classified, so no fair value of either is reported, and the notes' own Level 2 value (US$1,499.6M at 2026-06-30, 111% of par) comes from the stock's price. At our values it is out of the money in every case (central exit $446, bull $989, against a conversion price near $1,038). The notes' Level 2 value less their value as debt at 7.79% on that date ($931M) leaves roughly $570M, about $19 a share, as the market's whole price for the option before the capped call's offset (inference). Its sign is negative and it is not in the value.

Bear ($42.07): a networking bust (CED -22% in 2028 and -10% in 2029), loss of the largest consumer customer's socket in 2029 (MIC -40%), gross margin back to 60-61% as quartz makers and customers press on price, stock pay stuck at 12-13%, growth after 2031 of 6%. 2031 revenue $1,056M and EBIT of $122M; its free cash from H2 2026 to 2031 (about $500M) does not cover the $1,350M of notes, which is most of why the bear is worth so little. The 2031 notes are repaid at par in every case. The bear's outside view is 26.5x on its pre-amortization EBIT of $122M ($107 a share), although its GAAP EBIT, after $200M of amortization, would be negative; the restated multiple is on pre-amortization EBIT by construction, so that is the same footing, but a buyer of a business that earns less than its amortization may pay less, and the bear is generous by that much. Bull ($630): CED +70% in 2027 and 12-25% a year after, TPD +70% as the transition supply opens up, gross margin 69%, stock pay 8%. 2031 revenue $3,155M at a 43% margin. No case reaches the notes' conversion price, so none converts.

Sensitivities on the central $287 (each alone, output.txt):

The single most important operating assumption is the gross margin SiTime holds once datacentre timing is a billion-dollar line and the quartz makers answer. The beta is borrowed from the peers (SiTime's own needs its price history); it is worth as much as the margin. The exit multiple is worth more than either: each 1.0x is about $6.

4. About 70% of revenue rides on AI datacentre spending, and Bosch is the largest single risk

The owner of this research asked the questions below; each answer follows its question.

Is MEMS displacing quartz, and how fast? The evidence is SiTime's own growth, not a market statistic. SiTime's revenue went from $116M (2020) to $327M (2025), through a 2023 drop of 49% when customers cut inventory (FMP). Management claims about 40% share of timing in 400G optical modules and close to double that at 1.6T (Goldman Sachs conference, September 2026), and an optical module opportunity of about $0.5B for SiTime within a couple of years (claim). The 10-K describes design wins only qualitatively; SiTime does not publish a win rate, a win count or a funnel value, so the rate of design wins is not sourced here. What the filings do show: a design win takes a year or more to produce revenue, and once a part is designed in, customers seldom switch until the next product generation (10-K risk factors). That makes share gains slow to reverse, in both directions. Quartz remains the cheap default in most consumer and industrial sockets, and the quartz makers' own revenue and margins are not on FMP, so the share-shift view rests on SiTime's claims and its growth.

How much is AI datacentre timing? CED was $172M in 2025 and $177M in H1 2026 alone. About 75% of TPD's revenue is also datacentre and communications clocks (management description of the business it bought, 2026 calls; not a reported split). So roughly 70% of SiTime's revenue now rides on datacentre and networking spending: optical modules (MEMS oscillators replacing quartz in 800G and 1.6T transceivers), switches and NICs, and PCIe and server clocks (TPD's network-synchronization and jitter-attenuator parts). The central case grows CED 45% in 2027 and then slows to 3% in 2029; the accelerator spend path in our semis-ai sector note (AI-chip spending) is +72%, +8%, 0%, +4%, +6% for 2027-31, so our path is a little below the spend in 2027 (share gains do not fully offset a slowing build) and above it after (share gains continue).

Customer concentration. The top three distributors were 59% of 2025 revenue (56% in 2024) and 66% of Q2 2026; the ten largest end customers were 65% of 2025 (10-K 2025, 10-Q). Apple was 17% of 2025, down from 22% in 2024 and 21% in 2023. The bear case removes Apple's socket in 2029 (MIC -40%), because one product design can be lost in a single redesign (10-K risk factors). TPD adds its own concentration: WT Micro 51.6%, Macnica 18.1% and Avnet 13.2% of its 2025 sales (8-K 2026-05-19).

Capacity and Bosch. SiTime's MEMS wafers come primarily from Bosch, with Teledyne the second fab (10-K 2025). The Bosch supply runs under an agreement through February 2027 that renews automatically (10-K 2025); the CEO expects a renewal soon (claim, Q2 2026 call). SiTime's licences to Bosch MEMS patents expire 2025-2029. A non-renewal would not stop supply at once, and Teledyne already runs SiTime MEMS, but whether it could carry the volume alone is not found in any source (our timing-sector note, open question 3), so Bosch is the largest single operating risk and we do not price it as a scenario. No renewal 8-K had been filed by 2026-10-20 (sector note). TPD's products are made by Renesas under a transition services agreement (TSA); management says TPD growth next year (60-80%, claim) is limited by that supply. The central case takes +45% for TPD in 2027.

Stock-based pay and dilution. Stock pay was $103.5M in 2025 and $61.8M in H1 2026, with $198M not yet recognized (10-Q 2026-06-30). SiTime also spent $87.3M of cash in H1 2026 withholding tax on vesting shares, which keeps the count down at a cash cost. New shares: the June 2025 follow-on (about $388M net), a small at-the-market sale, and the 3.56M shares to Renesas. We charge stock pay as a cost (central 9% of revenue by 2031, still above the peer median) and hold the count at 30.07M; 12% instead of 9% costs $27 a share. The 1.297M unvested stock awards (RSU, PRSU and MYPSU types) and the 0.110M inducement RSUs (awards made to attract new hires) are not in the count: their $198M of unrecognized cost is part of the stock pay we charge, so adding the shares too would charge them twice (section 9, finding 6).

Acquisitions. Two, from the filings. Aura Semiconductor's timing assets in December 2023 ($259.2M fair-value consideration, part of it a sales-based earnout through 2028, $81.4M still owed at fair value on 2026-06-30). Renesas's timing business, closed 2026-07-01: $1.5B of cash and 3.56M shares, about $4.0B in all, funded by $1.35B of 0% convertible notes due 2031 and cash on hand; a $900M bridge loan was arranged and not drawn, and a $200M revolver is undrawn. Purchase accounting put $2.12B on acquired intangibles and $1.88B on goodwill. TPD made $207.7M in 2025 (flat on 2024) and $153M in H1 2026, so the deal is a bet that SiTime can sell Renesas's clocks into the AI build faster than Renesas did; H1 2026 says it has started.

5. The peers are mostly far larger companies, so they say little about timing

The peers that compete in timing are mostly far larger companies where timing is a small line, so their statements say little about timing itself. The ten in the exit-multiple comparison (chosen by scripts/research-anchor-data.py) are analog and mixed-signal companies; their latest fiscal year (FMP):

SiTime's central 2031 (65% non-GAAP gross margin, 32% operating margin after stock pay before amortization) sits near Rambus and Texas Instruments on margins, with revenue of a Rambus-sized company. The quartz makers named in the 10-K are not on FMP's statements, so their margins, capacity and pricing could not be compared; that is the largest gap in this report.

6. The price needs a much higher exit multiple or a business nearly twice our size

At $494.71 (FMP quote, 2026-10-20T17:20:42Z) SiTime's equity is worth about $14.9B on our 30.07M fully diluted shares. Our central $287.18 is 0.581 times that, and the case-weighted $311.73 is 0.630 times. What the price needs (market.py, everything else central):

• On our operations, a much higher exit. An outside view of 59.6x EV over pre-amortization EBIT on 2031 EBIT of $665M, which is 85x on GAAP EBIT, against the peers' 2016-2025 median of 35x on GAAP and 26.5x restated. Put another way: an exit price near $811 on 2031-12-31 against our $446, or 37x 2031 EBIT for the whole equity.
• At our 26.5x, a much bigger business. 2031 revenue of about $3.2B and EBIT of about $1.18B, 1.8x our central, which takes every segment 12.5 points a year faster than central in 2028-31. That is close to our bull case (2031 revenue $3.16B and EBIT $1.35B, reached with a 69% gross margin and stock pay of 8%), which is why the bull's $630.47 is the only case above the price.
• No single driver does it (alpha research odds): within each bracket the best end gives $382 (CED +120% in 2027), $478 (the 2028-31 CED path 22 points higher) or $385 (a 75% gross margin). The price needs several to go right together and an exit multiple above the peers' history.
• The odds in our scenarios are under 1% (0 of 5,000 draws; odds.json range 0.0% to 0.08%; a mixture of the base world at 75% and the digestion world at 25%). Simulated median $260, 10th percentile $180, 90th percentile $324. That P is our distributions, not a calibrated probability, and it is the one figure quoted for the odds everywhere in this report. Plainly: the six simulated drivers, inside the ranges we set before the price, cannot reach it; the bull case, which sets roughly ten inputs at once (section 7), does. The distributions centre on our view, so P measures how far the price sits outside our ranges, not the chance the market is wrong.

Cost of equity. We used the ten peers' Blume-adjusted median beta (1.48, cost of equity 11.52%). SiTime's own FMP profile beta is 2.885, Blume-adjusted 2.263, a cost of equity of 14.79%; on it fair value is $224.01 (-$63.17). SiTime's own beta reflects a stock that has moved several times the market in both directions (FMP's 52-week range is $243.68 to $901.81); we keep the peers' because a business beta should not carry one stock's recent swings, and say so: the own-beta build would widen the gap.

Street. FMP's consensus target is $793.33 (median $875, range $370-$900); five targets in the last quarter average $748 and 22 in the last year $547.95, so the Street has been raising targets. FMP's Q3 2026 revenue estimate is $291.8M, at the top of the guide; our $314M takes the measured beat. So our near-term revenue is above the Street, and our value is far below the price: the gap is the exit, not the next year. The Street's targets imply that the AI-led growth of 2025-27 lasts well past 2027 at a multiple near today's.

The Street through our model (street.py, street.txt; flag 17). The Street's average revenue (FMP, 5-8 analysts) is $875M for 2026, $1,353M for 2027 and $1,735M for 2028: below ours in 2027 ($1,532M) and level in 2028 ($1,720M). Run through our model, the Street's revenue moves fair value by +$1.07 to $288.25: on revenue we and the Street agree. The Street's EBIT is not on one basis (34% of revenue in 2027, then 50% in 2028 from five analysts); taking its 2028 figure at face needs a 73.9% gross margin for 2028-31 and gives $372.38. So, driver by driver: revenue +$1, the Street's 2028 margin +$84, and the remaining $122 to the price is the value after 2028, the exit. The Street's $793 average target is 39.5x its own 2028 EPS of $20.06 (non-GAAP, before stock pay and amortization), against our exit of about 14x (fade) to 26.5x (outside view) EBIT on 2031 figures. The disagreement is about what the business is worth once AI networking growth slows, not about the next two years.

Automated checker flags, and how each was answered. The repository's report checker (scripts/research-check.py) raises numbered flags when a report shows a pattern that needs an answer. The checker's flag number is in brackets, so its output can be matched to this list.

• A large move from the blind value (flag 1, medium): $267.75 to $287.18 (+$19, +7.3%). The largest change is the share count (+$12.9, finding 6), then the exit multiple (+$12.0, finding 1). Both come from the review's reading of the filings and the method, not from a management claim or a Street figure found after the price. Section 9 has the bridge.
• Value 42% below the price: is a risk charged twice? (flag 3, medium). We checked each place a risk could enter twice. Stock pay is a cost and the share count is held flat, so dilution is charged once. Acquired amortization is left out of operating profit and out of the exit EBIT (the 26.5x restatement), and its tax saving is counted once, today. The networking slowdown is in the central path (a flat 2029); the digestion world sits only in the odds, not in fair value. The beta is the peers' (1.48), not SiTime's own (2.26 adjusted), which would cut $63. We found no double charge. The input the market reads differently is growth past 2027 and the exit multiple (above and market.py).
• Outside view 92% above the fade: the input behind it (flag 4, low). Growth after 2031 against the cost of capital. The fade takes 10% growth in 2032 down to 3% over 15 years at 11.52%; at that rate, growth near the cost of capital adds little, so the fade lands at 14.8x year-six profit. The outside view carries the peers' 2016-2025 multiples, set when rates were lower and MPWR and MTSI traded near 60x. At 14% growth after 2031 (the bull) the fade rises to 18.0x. Section 3 keeps the method's mean.
• P(price justified) under 1% (0 of 5,000 draws): build the world where the price is right (flag 16, high). It is close to our bull case, which is worth $630.47, above the price: CED +70% in 2027 and 12-25% a year after, the acquired business +70% then +25%, gross margin 69%, stock pay 8%, 2031 EBIT of $1.35B, growth after 2031 of 14%. market.py finds that at our 26.5x exit the price needs 2031 EBIT of $1.18B, inside that world. What would make it true: SiTime's share in 1.6T optics doubling as management claims, the quartz makers not answering on price, and AI networking spending avoiding the flat year that the record of LightCounting (an optical-networking market research firm) says follows every 2-3 growth years. We weight that world at 25% in the case-weighted value ($311.73) and it is why we do not treat the price as absurd, only as a bet on the bull.
• Fair value 0.40x the discounted Street target (flag 17, high): the Street-through-our-model paragraph above.
• The price outside our 10th-90th percentile range ($180-$324): are the ranges overconfident? (flag 18, medium). The simulation varies six drivers around the central case and has a digestion world but no mirror-image boom world, so its 90th percentile ($324) sits far below the bull case ($630), which sets about ten inputs at once (the six drivers near the tops of their ranges, plus AIA and MIC growth, operating-cost pass-through, working capital and the mature return). We did not widen any range after seeing the price, as the blind rule requires: a range widened toward the price would make the odds circular. So the figures are not in conflict: the simulation's ranges cannot reach the price (under 1%, 0 of 5,000 draws) and the bull case, whose inputs sit outside what the simulation draws jointly, does. The case spread, $42 to $630, contains the price. Process.md proposes paired worlds for future reports.
• A convertible in the share count (flag 8, high if missing). The 2031 notes are not in the share count because no case converts (the conversion price is near $1,038 against a bull exit of $989); they are valued as debt in the bridge, and the conversion option net of the capped call is a named, unvalued claim with a negative sign (section 3 and finding 5).
• A lenient input where a harsher sourced one exists (flag 6, medium). The normal tax rate is 15%; the 10-K shows a 21% federal statutory rate and no foreign-derived income benefit yet (2025). At 21% fair value is $280 (-$7). The 15% is an unsourced assumption and stays disclosed (finding 9).

Where we differ from the market, and what settles it. We think the market prices SiTime as if datacentre timing keeps compounding at 30%+ past 2027 and is worth a peak-cycle multiple at the end; we think 2027 is mostly the Q4 2026 run-rate, that optics will have its flat year (LightCounting's record), and that the gross margin settles near management's 65% floor as quartz makers answer. The first test is the 2027 guide and growth on the Q4 2026 run-rate: growth above 30% on that run-rate in 2027 (central +15%) would put the bull's path in play.

7. The case for a much lower value, and the case for a much higher one

Why it could be worth much less. The datacentre line could be a cycle, not a trend. SiTime's own revenue halved in 2023 after an inventory build, and optical and networking demand has busted before. Two-thirds of revenue comes through three distributors; one consumer customer is a sixth of it. The gross margin is at a peak with TPD's cheap Renesas-made parts in it; when the TSA ends SiTime pays its own costs. Stock pay at a third of revenue is a cost the non-GAAP figures leave out. Two fabs make every MEMS wafer, Bosch the main one. The bear's $42 is not a tail: it is the 2023 cycle repeated on a bigger base.

Why it could be worth much more. If MEMS displacement in optics and servers runs at the share management claims, revenue could pass $3B by 2031 at margins already guided, and a higher exit multiple than the 26.5x we use (MPWR and MTSI sit at 57x and 40x on the same restated basis) would apply. The beat record is seven for seven. The bull's $630 needs no margin beyond today's guide.

8. What would settle it: Q3 results, the Bosch renewal and the Renesas supply

• Q3 2026 results, early November 2026. Central Q3 is $314M (core $222M, TPD $92M) with a 68% gross margin. Revenue under $295M (no beat) or a gross margin under 66% moves the central toward the bear. A Q4 guide above $340M moves it toward the bull.
• Bosch renewal, before February 2027. A renewal on the same terms removes the largest unpriced risk; a non-renewal or a price change is a new case.
• TPD supply. TPD revenue in Q4 2026 and Q1 2027: under $90M a quarter means the 2027 +45% is too high (each 25 points is about $21 a share).
• 2027 gross margin as the TSA winds down. Under 63% for two quarters says the central 65% is too high (4 points is about $39).
• CED in 2028. A fall of more than 10% would mark the digestion the bear assumes.

9. Revision 1: an outside review moved the central value from $267.75 to $287.18

The review of the first draft raised eleven findings; the chief investment officer (CIO) ruled on each. This section answers each by number; the step numbers in the bridge table are the finding numbers. Every change below rests on a filing or a method rule, none on the price or the Street, and the blind commit is unchanged. The central value moves from $267.75 to $287.18 (+$19.43); the call stays Avoid.

Bridge, central, US$ per diluted share (ordered; each step on top of the last; bridge.py, bridge.txt; no residual):

The same changes taken one at a time away from the revised model (isolated, so interactions show): exit multiple +$12.55, notes -$4.63, share count +$12.83, tax basis +$1.44, working capital -$2.45. Findings 2, 3, 4, 8, 9, 10 and 11 change no fair value.

1. Exit multiple, per peer, on a pre-amortization basis (accepted; +$11.98). The flat 24.5x scaled the peers' 35.0x by SiTime's own amortization share, which is not what the peers' multiples contain. Each peer's own amortization of intangibles (the 10-K cash-flow add-back, us-gaap:AmortizationOfIntangibleAssets from the SEC XBRL facts, saved in data/exit-anchor-amortization.json by anchor_amortization.py with each year's accession number) is added back to its GAAP operating income in every year of the ten-year window, the method's EV/EBIT median is taken again per peer, and the median of those medians is applied to each case's own pre-amortization 2031 EBIT:

Silicon Labs (SLAB) and Lattice (LSCC) are out of both medians (fewer than five valid years). The filings give 26.5x, not the reviewer's 28-30x proxy. It sits between ADI (21.4x) and SMTC (31.6x), the fourth and fifth of eight, and does not depend on the two gaps: TXN has no fact for 2024-25 and MPWR none for 2019-25 (the line is no longer presented, an inference; both would sit on the same side of the median either way), and SLAB is not in the SEC facts held in the data lake. Keeping only the years that were valid on the GAAP basis gives 26.5x as well. Restating makes three loss-year GAAP rows positive (MCHP 2025, RMBS 2021, MTSI 2020); the method's own rule is applied to the restated figure. In the bear, GAAP EBIT would be negative ($122M less $200M of amortization), so the restated multiple is applied to its pre-amortization EBIT, as it is in every case; this lifts the bear's outside view and the bear's value by $2.20, and is generous for a business that earns less than its amortization. Forecast.changes records the change with the per-peer figures and sources.

2. Five-year returns (accepted). Every printed return now comes from method.five_year_return, which includes the modeled free-cash payouts and today's claims: central +0.2% (was -3.8% on the exit price alone), bear -29.6% (was -32.2%), bull +17.0% (was +12.2%), in the header, market.py, market.txt and the text. The old "bull earns only the hurdle" sentence is replaced: the bull beats the 12% hurdle by 5 points, the central case does not. The payouts are the model's convention (free cash paid out as earned), not a forecast of buybacks.

3. The odds (accepted). One number, from odds.json, in the same words everywhere: under 1% (0 of 5,000 draws; range 0.0% to 0.08%). The header's "under 20%" band is gone. The plain statement is in flag 16 and 18: the six simulated drivers cannot reach the price at the ranges we set before it, and the bull case, with about ten inputs set together, does. No range was widened. Paired worlds (a downside and a mirror-image upside, weights set before the price) are proposed for future reports in process.md.

4. The downturn weight (accepted). The 25% weight on a real fall rather than a flat year is our judgment; the LightCounting record and SiTime's 2023 fall show such years happen, not how often a fall rather than a flat year. drivers.json says so, and the line here and in process.md that the sector note "backs the weight" is corrected. The weight is unchanged.

5. The 0% notes as debt (accepted; -$4.42). The $1,350M repayment on 2031-06-15 is no longer a negative payout discounted at the 11.52% cost of equity ($813M). It is a claim today at the risk-free rate plus a credit spread: 5.29% + 2.50% = 7.79%, $952M. The spread is the top of the secured revolver's SOFR + 1.75% to 2.50% grid (8-K 2026-07-01); the notes are unsecured, so it is a floor, and a 4.5% spread would give +$2.60. The conversion option net of the capped call is not valued (section 3): the cap price is masked, the capped call is equity-classified and has no reported value, and the only filed value, the notes' $1,499.6M Level 2 quote, comes from the stock price. Its sign is negative; the quote implies up to about $570M ($19 a share) for the gross option on 2026-06-30, before the capped call's offset.

6. The unvested-award double count (accepted; +$12.88). The 1.297M unvested awards (RSU, PRSU, MYPSU; +$11.88) and the 0.110M inducement RSUs (+$1.04) are out of the share count, now 30.071M. The alternative, taking their $198M unrecognized cost out of the expense, was rejected: stock pay is forecast as a share of revenue from recent rates, not as a schedule of grants, so the award cost cannot be cut out of it without inventing a split, while the count is a reported figure. The inducement RSUs were not in the finding; they sit in the same stock-pay line (H2 2026), so the same logic applies.

7. Tax basis and working capital (accepted; +$1.44 and -$2.45). The 8-K/A 2026-09-09 exhibit 99.2 (note 4) says the acquisition was an asset purchase for US federal income tax purposes with a tax basis stepped up to fair value, so the whole $4,006M is deductible over 15 years in every case (central 85% to 100%, bear 50% to 100%, bull unchanged); the final tax allocation is not complete. The working-capital ratio of receivables plus inventory less payables to annualised quarterly revenue was 27.0%, 23.2%, 27.1% and 26.0% over the last four quarters (FMP quarterly statements; 10-Q 2026-06-30), mean 25.8%, so the model uses 26% (bear 30%, bull 24%, the same steps from central as before) in place of the blind 22%.

8. Gross margin (note only). No better source found: management's long-term "at least 65%", the Q3 guide of about 68% with TPD, 66% non-GAAP in H1 2026 and a peer median of 57.7% are the evidence; TPD's carve-out margin is before the transition services end. The 65% stays; 4 points is -$39 / +$29.

9. Tax rate (note only). No better source found. The 10-K shows a 21% federal statutory rate and no foreign-derived income benefit in 2025; 15% needs that benefit once SiTime is profitable. At 21% fair value is $280.48 (-$6.70). The 15% stays as the weakest-sourced input.

10. Beta (note only, unchanged). The peers' Blume-adjusted 1.48 stays; SiTime's own beta would give $224.01 (-$63.17).

11. Citation and confidence (accepted). The 26.512M basic shares are from the statement of stockholders' equity at 2026-06-30 (10-Q), not the cover; assumptions.md is corrected. Confidence is now 25% for a value within ±20% of the central exit price, the definition in valuation.md (it was 30% at ±25%; the narrower band lowers it). record.json's blind commit is left for the CIO.

10. Sources

SEC filings (EDGAR, fetched 2026-10-20, masked copies in data/filings/): 10-K 2025 (filed 2026-02-11); 10-Qs for 2025 Q3, 2026 Q1 and 2026 Q2 (filed 2026-08-06); earnings releases Q4 2024 to Q2 2026; 8-Ks of 2026-02-04 (deal), 2026-03-23, 2026-05-11, 2026-05-19 (TPD carve-out and pro forma), 2026-07-01 (closing, credit agreement); 8-K/A 2026-09-09 (TPD interim, pro forma with purchase accounting). FMP statement endpoints for SiTime and the ten peers, and the peers' FMP profile beta (fetched 2026-10-20). Transcripts: Q4 2025, Q1 2026 and Q2 2026 calls; Goldman Sachs conference, September 2026 (TradingView documents). Sector notes: semis-ai accelerator spend path; timing (docs/research/sectors/timing/note.md, read after the blind commit, source of the LightCounting cycle; it gives no weight for a fall rather than a flat year). Market pass: FMP quote, profile beta, price-target consensus and summary, earnings calendar (data/market/, read 2026-10-20). Market inputs: risk-free 5.29% and equity risk premium 4.20% (Damodaran, 2026-10-01, the method default). Revision 1: the peers' amortization of intangibles from SEC 10-K XBRL facts (companyfacts, read in place from the data lake, saved in data/exit-anchor-amortization.json); 8-K/A 2026-09-09 exhibit 99.2 note 4 (tax basis); 10-Q 2026-06-30 (stockholders' equity statement, note 3 notes terms and Level 2 value); credit agreement terms in the 8-K of 2026-07-01.

Fair value today      $287.18 per diluted share (cost of equity 11.52%); blind $267.75
Price                 $494.71, FMP quote, 2026-10-20T17:20:42Z (intraday; previous
                      close $495.75)  ->  fair value 41.9% below the price
Five-year exit price  $445.69 on 2031-12-31: the average of two estimates, the long fade
                      ($306) and the outside view ($586 at 26.5x EV/EBIT, enterprise
                      value over operating profit); at 10/20 years of fade $431/$459;
                      value-driver formula 43.8x against the fade's 14.8x year-six NOPAT
Five-year return      +0.2% a year from today's price on our central case, including
                      the modeled free-cash payouts and today's claims
                      (method.five_year_return, market.py); bear -29.6%, bull +17.0%
Purchase price        $280.91 to earn the 12% hurdle (the price at which a buyer would
                      earn 12% a year on our central case)
Bear / bull today     $42.07 / $630.47: a severe but plausible bad case and the matching
                      good case, each one a consistent business outcome (weighting bear,
                      central and bull 25/50/25, our weights: $311.73)
Confidence            25% (subjective, not calibrated): the chance the 2031-12-31 value
                      lands within ±20% of the central exit price
Moat                  Narrow and real: MEMS resonators (tiny vibrating structures etched
                      in silicon) on a programmable CMOS platform, 20 years of process
                      know-how, and design-in lock (customers rarely switch once a part is
                      designed in); held back by two MEMS fabs (Bosch, Teledyne) and a
                      customer base where ten buyers are two-thirds of revenue
Market expects        Far more than any one of our drivers can give: on our central
                      operations, an outside view of 60x EV/EBIT on pre-amortization
                      2031 EBIT (before amortization of acquired intangibles; 85x on GAAP
                      EBIT, against the peers' 35x on GAAP and 26.5x restated); or, at
                      our 26.5x, 2031 EBIT of about $1.18B (1.8x central), every segment
                      12.5 points a year faster in 2028-31 (market.py)
Price needs           No single headline driver reaches the price within its bracket
                      (alpha research odds): the best ends give $382 (CED, a sales segment
                      defined below, 2027 +120%), $478 (CED 2028-31 path +22
                      points) and $385 (gross margin 75%)
Odds in our scenarios Under 1% that fair value reaches the price (P(price justified), the
                      share of 5,000 simulated scenarios in which it does: 0 of 5,000,
                      odds.json; subjective, not calibrated)
Key driver            Gross margin 2028-31: 61% / 68% instead of 65% is -$39 / +$29 a
                      share. The exit multiple is worth more: the peers' unrestated
                      35.0x GAAP multiple on pre-amortization EBIT would add $53, and
                      each 1.0x of the restated multiple is about $6
Street                Consensus target $793.33, median $875, range $370-$900 (FMP,
                      read 2026-10-20); 5 targets in the last quarter averaging $748,
                      22 in the last year at $547.95, none in the last month, so targets
                      have risen; an expectations reference, not validation. Q3 2026
                      revenue estimate $291.8M against our $314M