AI Scorecard: 60% Accuracy for the Week of August 10th

By Yogurt · 2026-08-15 · AI Scorecard

Our weekly review of Yogurt's AI-driven earnings predictions. This week the engine reviewed 15 earnings reports and correctly called 9 outcomes — including COHR, CSCO, AMAT, NU, and CoreWeave — for a 60% accuracy rate. Full breakdown inside.

The Yogurt AI engine reviewed fifteen earnings reports for the week of August 10-14, 2026 - the most AI-concentrated earnings slate of the entire Q2 season. The engine correctly called nine outcomes for a 60% accuracy rate, an improvement over last week's 53%. The model landed clean calls on the highest-conviction AI infrastructure names but was tripped up by mixed-signal reports and one dramatic earnings implosion. Here is the complete breakdown.

This Week's Results (August 10-14, 2026)

Below is the full table of predictions versus actual reported outcomes:

The AI Infrastructure Report Card: Mostly Delivered

At the start of the week, the central question was whether the AI buildout would prove real and broad. The answer: mostly yes, with nuance.

Coherent Corp ($COHR) was the highest-conviction call of the week, and it delivered. EPS of $1.74 versus $1.65 estimated, revenue of $2.05 billion up 34% year-over-year. 800G optical transceiver demand is structural, not cyclical. The model saw this clearly.

Cisco Systems ($CSCO) reported record revenue of $17.3 billion, non-GAAP EPS of $1.22 (vs $1.19 estimate), and disclosed $9.3 billion in AI infrastructure orders for FY2026 before raising FY2027 guidance. The stock dropped despite the beat, illustrating how elevated expectations have become in AI infrastructure. Still a correct prediction.

Applied Materials ($AMAT) posted record Q3 revenue of $9.12 billion with EPS of $3.50 (vs $3.39 estimate) and guided for $10.25 billion in Q4 - a 30%+ equipment growth trajectory validating multi-year AI chip manufacturing expansion at TSMC, Samsung, and Intel.

CoreWeave ($CRWV) proved the AI GPU cloud is a genuine business: EPS of -$1.03 versus -$1.49 estimate (30.87% beat), with full-year revenue guidance of $12.4-$13.2 billion. The AI cloud pure-play thesis is intact.

Lumentum ($LITE) and Nu Holdings ($NU) both exceeded on the strength of secular demand. Nu crossed $1 billion in quarterly net income for the first time in its history - a milestone for Latin American digital banking.

Where the Model Stumbled

Cerebras Systems ($CBRS) was the week's most dramatic miss. The stock fell 14-16% after reporting. The core issue: two different revenue numbers - total revenue of $180M versus core/adjusted revenue of $210M - with GAAP EPS a massive miss versus consensus. The model had 73% confidence in a beat; clearly too high for a company still navigating complex accounting for government AI contracts.

Super Micro Computer ($SMCI) defied the prediction in the opposite direction. The model expected near-consensus results; instead, SMCI delivered EPS of $1.62 versus $0.96 expected - a 69% beat - while missing on revenue ($11.1B vs $11.6B). Gross margins nearly doubled to 17.5%. The model's "MEET" call correctly identified SMCI was unlikely to miss badly, but failed to anticipate the dramatic margin improvement that drove an EPS blowout.

Sea Limited ($SE) and Brookfield Corp ($BN) also went against predictions. Sea's revenue grew 48% year-over-year to $7.8 billion - well above expectations - but higher costs caused EPS to miss at $0.70 versus $0.83 estimated. Brookfield's GAAP EPS of $0.14 missed the $0.65 consensus substantially, though distributable earnings were up 15% per share year-over-year.

Week Summary: 60% Accuracy - What It Means

Nine correct calls out of fifteen represents a step forward from last week's 53%. The pattern was instructive: high-conviction structural calls on AI infrastructure landed cleanly (COHR, CSCO, AMAT, CRWV, LITE), while mixed-signal, evolving-business calls (CBRS, SMCI, SE, BN) tripped the model.

The lesson from CBRS is particularly valuable: companies with complex revenue accounting (pass-through contracts, customer warrant amortization, government AI contract structures) are harder to score on a simple beat/miss basis than companies with clean, recurring revenue. Yogurt's confidence calibration will apply a haircut to high-complexity accounting situations going forward.

The cumulative accuracy across all weekly scorecards stands at 78 of 127 predictions correct - a 61% long-run rate. The Schwartz watches all. May your portfolio be with you this week.