Alphabet, Q4 2024.

A Fortune 100 announcement window. The decision to spend $75 billion on AI infrastructure was defensible. The framing was the own-goal.

Market cap destroyed

$215B

in the first 90 minutes after the release

T=0

2025-02-04

16:15 PT, after-hours earnings release

01 / The decision

A $75 billion number without a story around it.

Alphabet walked into Q4 2024 earnings with a capex announcement that was going to define the company's AI posture for the next three years. The underlying strategic decision was defensible. Investor day materials from the prior months had trailed a significant infrastructure build. Microsoft had already committed to roughly $80 billion of 2025 capex and Meta was weeks away from announcing $60 to $65 billion. Spending on AI infrastructure was the consensus posture across every frontier-model operator at that scale. The question in front of the Alphabet CFO and the Investor Relations team was not whether to announce the number. It was how to frame it.

That is the archetype. A locked decision. A framing window. A short time horizon. Every major announcement window carries the same shape — earnings, M&A, strategic reset, guidance revision, workforce event. The financial consequences of the framing decision are frequently larger than the financial consequences of the underlying decision itself, and a framing decision made by committee over one weekend is almost never rehearsed against the population that will actually react to it.

On 4 February 2025 at 16:15 PT, Alphabet released Q4 2024 earnings and committed to $75 billion of 2025 AI capex. The release also disclosed a Google Cloud revenue miss against consensus. The stock fell as much as 9% in after-hours trading and closed down around 7% the following regular session. Roughly $215 billion of market capitalization disappeared in the first 90 minutes. That same week, Meta announced a comparable $60 to 65 billion capex commitment with a different framing — a data centre the size of Manhattan, 1.3 million GPUs, concrete product milestones — and the stock rose around 1.7% on the day. Same capex scale. Different framing. Opposite market reaction.

T=0 / The cutoff

3 February 2025, internal cutoff.

The rehearsal was run on documents dated on or before 3 February 2025 — the internal cutoff a Fortune 100 Investor Relations team would realistically work to the day before a Tuesday after-hours earnings release. In working memory at T=0: the capex number itself, prior-quarter guidance, the Google Cloud revenue line, Microsoft's comparable commitment, and the market's framing of the frontier-model race.

Held out of the rehearsal and never uploaded to Glasshouse: Alphabet's own 4 February release, the after-hours tape, the tier-1 analyst reaction, the comparative read to Meta on 5 February, and every piece of reporting written after the 1615 PT release. Gemma 3's training cutoff is August 2024; the Alphabet, Microsoft, and Meta 2025 capex announcements are all post-cutoff. The model has no memorized reaction to retrieve.

02 / The rehearsal

What Glasshouse produced.

The rehearsal surfaced the narrative split between credible AI infrastructure investment and uncontrolled spending with vague returns inside the first eight rounds of the simulation. By round twelve, the second framing was dominant across the F100 buy-side and tier-1 equity-research personas. The rehearsal also placed the Meta comparison correctly as the rising alternative-framing signal, and identified the Google Cloud revenue line as the detonator — the specific disclosure that converted a neutral-to-negative announcement into a structurally-damaging one. 1

The simulation's narrative-ranking output predicted the 72-hour after-hours reaction window — the interval in which the framing narrative locks in before the next day's regular session can move the stock. The personas driving the dominant uncontrolled spending narrative were the sell-side equity research and long-only institutional tiers, not the retail tier. The retail tier in the simulation largely adopted the framing they received from the top two tiers rather than forming an independent one. 2

The specific framing recommendation coming out of the rehearsal was to lead with concrete infrastructure milestones and the unit economics of the committed spend — the Meta approach, not the Alphabet approach — and to separate the Google Cloud revenue disclosure from the capex number by a session. 3

03 / What actually happened

Rehearsal alongside record.

Glasshouse rehearsal · T=0 2025-02-03

Predicted narrative ranking

01 Uncontrolled spending with vague returns — dominant by round 12
02 Google Cloud revenue miss — detonator, tier-1 coverage
03 Meta-comparison stakeholder framing — rising alternative
04 Credible AI infrastructure investment — minority view by round 12

The rehearsal placed the framing trajectory inside the 72-hour after-hours window and named the detonator before the release landed.

Real-world record · 4 to 7 February 2025

Documented reaction

01 Uncontrolled spending framing dominated tier-1 equity research notes in the 72 hours after release
02 Google Cloud revenue miss named as the detonator in Bloomberg and Reuters coverage
03 Meta's 5 February announcement reframed as the positive comparison — concrete-milestone framing, +1.7% on the day
04 Alphabet stock down as much as 9% after-hours, ~7% in the following regular session

Primary reporting from Bloomberg, Reuters, Financial Times, and CNBC coverage dated 4 to 7 February 2025.

The variance between the Alphabet and Meta reactions is the commercial anchor. The rehearsal cost is a rounding error on that variance.

04 / Methodology

Four rules. No exceptions.

Rule / 01

T=0 is explicit.

T=0 for this case is 3 February 2025, the day before the Alphabet release. The seed contains only documents dated on or before that day.

Rule / 02

T=0 is post-cutoff.

Gemma 3's training cutoff is August 2024. The 4 February release and every piece of post-release reporting are six months post-cutoff. The model has no memorized reaction to retrieve from training data.

Rule / 03

Financial anchors are real.

The $215 billion market-cap number and the Meta +1.7% comparison come from primary reporting in Bloomberg, Reuters, Financial Times, and CNBC. Every number on this page is verifiable.

Rule / 04

Never tuned to pass this case.

Glasshouse is tuned on aggregate eval results across many scenarios. This case is in the public gallery because it already passes the rubric as a byproduct of aggregate tuning, not because Glasshouse was reverse-engineered around it.

05 / Related rehearsals