Taiwan Semiconductor Manufacturing (TSM) - Strong foundry sales face a demanding earnings test

Original research record

TSMC reported September revenue of NT$511.86 billion, up 54.6% from a year earlier but down 0.6% from August. Its results call is scheduled for 15 October at 14:00 SGT. We connect the OTR data's foundry-demand work with the separate memory-scarcity work. Demand is visible at more than one production stage. This supports an earnings-led recovery if management confirms capacity use and pricing. Our call concerns the response to the coming outlook, not an assumption that already-published sales are a surprise.

We separate the date of the underlying evidence from today's new synthesis. We look for the link from demand to deliveries, cash flow and the share price. The actionable test is whether new company or market evidence supports that link during this short window; a repeated headline does not add support.

A weaker forward outlook, falling advanced-node utilisation or unexpected margin pressure would break our case. We would also reassess if rates rise sharply after US inflation data.

We favour bullish over the stated 72-hour nominal window, with only modest confidence. The outcome check compares the later reference against US$453.31 and requires a dated result.

Bayesian: we use a cautious subjective judgement; no fitted likelihood ratio or calculated posterior is claimed. De Finetti: the semiconductor and power stories share spending drivers, so these calls are correlated and their sources cannot be counted as independent bets. Kolmogorov: we define a neutral band of ±0.25% around the research reference before publication. Bullish above +0.25%: 57%; bearish below −0.25%: 33%; neutral inside the band: 10%. These total 100%. The separate directional HIT/MISS check uses the sign of the price change, with an unchanged price recorded separately.

OTR Database Data - 12 October Mega Research 13:56 and 14:00 editions; Research Intelligence 14:14; behavioural and learning reports 14:30-14:44. Repeated source stories are one evidence cluster, not independent confirmations. The 13:56 edition separates memory, accelerator, power and networking economics. The 14:00 edition requires issuer-specific confirmation instead of treating sector exposure as a completed call. We add the company evidence listed below, and keep announced plans separate from reported results.

The reviewed 30-asset behavioural extract does not provide an asset-specific analogue row for this instrument. We do not substitute another company's graph. Relevant sector behaviour is context only; full intraday structure and options positioning were not independently reproduced.

Expectations are already high. A small monthly sales decline does not establish accelerating growth. Customer concentration, export rules and power costs could weaken margins even while revenue rises.

This is a new forecast at 12 October 2026, 15:11 SGT. We have not established an earlier directly comparable, locked forecast for this exact horizon, so we do not claim a historical direction change. Where the behavioural score differs, we preserve the disagreement rather than rewriting it.

The broad narrative is that AI spending benefits the whole supply chain. The OTR data shows why that is too simple: each business faces different delivery, pricing, power and funding limits. We use the company's actual stage in that chain and distinguish a signed agreement from an event, forecast or technical article.

The latest OTR learning work does not establish a stable out-of-sample edge and reports abstention across its tested rule set. A formula confidence score is not a realised success rate. No verified matching historical win rate is available for this particular call; we therefore use a low subjective estimate and will preserve the eventual result.

How likely is this call? · Three frameworks

Source-linked subjective estimate. Historical comparisons and missing data are disclosed below.

Bayesian · what changed the estimate?

we use a cautious subjective judgement; no fitted likelihood ratio or calculated posterior is claimed.

de Finetti · are the comparisons fair?

the semiconductor and power stories share spending drivers, so these calls are correlated and their sources cannot be counted as independent bets.

Kolmogorov · do the numbers agree?

we define a neutral band of ±0.25% around the research reference before publication. Bullish above +0.25%: 57%; bearish below −0.25%: 33%; neutral inside the band: 10%. These total 100%. The separate directional HIT/MISS check uses the sign of the price change, with an unchanged price recorded separately.

Evidence behind this view

OTR Database Data - 12 October Mega Research 13:56 and 14:00 editions; Research Intelligence 14:14; behavioural and learning reports 14:30-14:44. Repeated source stories are one evidence cluster, not independent confirmations. The 13:56 edition separates memory, accelerator, power and networking economics. The 14:00 edition requires issuer-specific confirmation instead of treating sector exposure as a completed call. We add the company evidence listed below, and keep announced plans separate from reported results.

What past price moves show

The reviewed 30-asset behavioural extract does not provide an asset-specific analogue row for this instrument. We do not substitute another company's graph. Relevant sector behaviour is context only; full intraday structure and options positioning were not independently reproduced.

Evidence against this view

Expectations are already high. A small monthly sales decline does not establish accelerating growth. Customer concentration, export rules and power costs could weaken margins even while revenue rises.

Why the direction changed

This is a new forecast at 12 October 2026, 15:11 SGT. We have not established an earlier directly comparable, locked forecast for this exact horizon, so we do not claim a historical direction change. Where the behavioural score differs, we preserve the disagreement rather than rewriting it.

What people say vs the facts

The broad narrative is that AI spending benefits the whole supply chain. The OTR data shows why that is too simple: each business faces different delivery, pricing, power and funding limits. We use the company's actual stage in that chain and distinguish a signed agreement from an event, forecast or technical article.

How past forecasts performed

The latest OTR learning work does not establish a stable out-of-sample edge and reports abstention across its tested rule set. A formula confidence score is not a realised success rate. No verified matching historical win rate is available for this particular call; we therefore use a low subjective estimate and will preserve the eventual result.