THE MOSAIC LEDGER

TSEM

MISS :: earnings event :: tier RESOLVED

Registered claim

TSEM Q2 2026 (reports 2026-08-03) discloses 2027 contracted silicon-photonics revenue above the $1.3B secured as of Q1 2026, with explicit dollar quantification, AND prints revenue at least 2% above the frozen Street consensus snapshot.

Registered 2026-07-28.

Resolution spec

Spec registered 2026-07-28.

Verdict rule: HIT if leg A and leg B. PARTIAL if both leg A and not leg B, or both not leg A and leg B. MISS if not leg A and not leg B.

As frozen: HIT = A && B; PARTIAL = (A && !B) || (!A && B); MISS = !A && !B

Tier history

Evaluated daily since registration: 10 evaluations. The 1 that moved the tier is listed above.

Resolution receipt

Verdict: MISS. Resolved 2026-08-08.

Rule: HIT if leg A and leg B. PARTIAL if both leg A and not leg B, or both not leg A and leg B. MISS if not leg A and not leg B.

As frozen: HIT = A && B; PARTIAL = (A && !B) || (!A && B); MISS = !A && !B

Autopsy

Read the full autopsy
Resolved MISS on 2026-08-08. Both legs failed. The confirm leg came in at $460.079M against a frozen baseline of $454.68M, 1.19% above it against a 2% threshold.

The first version of this autopsy was wrong about why the core leg failed, and the correction changes what the print means. It recorded that Tower disclosed no dollar-quantified 2027 contracted silicon-photonics revenue figure, and that the company had moved its reporting from a contracted-2027 basis to an annualised-run-rate basis. Neither is true. Tower disclosed both bases on the 2026-08-04 call: chief executive Russell Ellwanger stated that the company has announced customer contracts representing approximately $1.3 billion of silicon-photonics revenue for 2027, with higher growth expected in 2028, alongside the $680M annualised second-quarter run rate and the $1B run-rate target for the fourth quarter of 2026. There was no basis change and no absence.

The leg failed on the number. It required a 2027 contracted figure above $1.3B, and Tower restated it at approximately $1.3B, flat against the disclosure from the previous quarter. Flat is not above, so the leg fails correctly, but it fails as a neutral-to-adverse datapoint rather than as a disclosure gap. That distinction is the difference between a company declining to tell you something and a company telling you the number did not move.

The analytically important fact was buried in the original resolution. The silicon-photonics run rate more than tripled year over year, from $180M to $680M, while the contracted 2027 book did not move off $1.3B. Ellwanger supplied the mechanism himself: additional capacity has been requested and committed by several lead customers, but not all of it is formally booked. The contracted figure is therefore understating demand by a booking-convention margin the company has acknowledged and not sized. That is a live watch surface rather than an anecdote about how to write a call. If other names in this cohort also present indium-phosphide and silicon-photonics demand as opportunity framing or run rate rather than as booked backlog, then the cohort is systematically underreporting contracted commitment, and any measure built on the contracted figure is measuring a convention rather than a constraint.

The print itself was strong: record revenue of roughly $460M, up 24% year over year, third quarter guided to about $520M for an implied annualised run rate above $2B, the 2028 model raised to $3.6B of revenue at 45% gross margin and 33% net margin, $290M of customer prepayments received, and a government-supported dual-track 300mm expansion in Japan. The claim that the optical architecture is transitioning was reinforced in substance and its falsifiability surface was not touched.

The corrected lesson is narrower than the one first recorded. Guarding against a change in reporting basis is not the lesson here, because no such change occurred. It is that a leg asking for a figure to grow should anticipate that a company which updates an annual contracted book quarterly will often restate it unchanged between updates, because contracted books step on contract-signing events rather than on a quarterly cadence. Where the question that matters is whether the book grew, the call needs either a horizon long enough to contain a plausible re-contracting event, or a measure of booked-plus-committed rather than booked alone.