Deep Research

Alphabet (GOOGL)

4 min read

Snapshot

  • Ticker: GOOGL (Nasdaq, Class A)
  • Bucket: Compute & AI Megacaps
  • Q1 2026 position: 555,023 shares, $159.6M, 3.19% of 13F — roughly the #8 equity position
  • HQ: Mountain View, California
  • What it does: Search and advertising monopoly-adjacent franchise plus YouTube, Google Cloud, Gemini frontier models, and the only vertically integrated TPU silicon stack among hyperscalers.

Business Overview

Alphabet’s FY2025 revenue was 350B, with net income of 70B run rate with Q4 2025 growth of 48%.

Q1 2026 accelerated further: total revenue ~460B — roughly half expected to convert to revenue within 24 months. Gemini 3’s launch landed as a clear technical win, and the company reported 350M paid subscriptions with its strongest-ever quarter for consumer AI plans. The bill is commensurate: Q1 capex of 180–190B, and management flagged FY2027 capex “meaningfully” above that.

Alphabet’s structural distinction is full-stack vertical integration — models (Gemini/DeepMind), custom silicon (TPU, now ~v7/v8 generation), its own datacenters, and distribution through Search, Android, and YouTube. That integration is both the bull case (cost-per-token advantage) and, lately, the contested claim (see below).

Financial Trajectory

PeriodRevenueYoYNotes
FY2024$350.0B+14%Net income $100.1B
FY2025$402.8B+15%Net income 70B run rate
Q1 2026~$110B+22%Cloud +63%; backlog >35.7B; FY26 capex guide $180–190B

Why Atreides Owns It

Alphabet is the more conflicted of Atreides’ two hyperscaler longs, and the position history shows it. On Invest Like the Best EP.451 (Dec 2025) — recorded into the Gemini 3 launch — Baker argued Gemini 3 confirmed that scaling laws were intact and that Google’s TPU stack plus its data and distribution made it the company best positioned to win if pre-training scaling continued: vertical integration meant Google could serve frontier-quality tokens at structurally lower cost. The Q2 2025 five-fold share-count rebuild (189K → 951K shares) preceded that public framing.

By May 2026 the tone had shifted. In the “Watts and Wafers” interview (ILTB EP.473) and at Sohn New York 2026, Baker argued Google had “lost its accelerator edge”: TPU v8 made deliberately conservative design choices while Nvidia and Amazon’s Trainium went aggressive, and Google still lacks an operational switched scale-up network of the kind powering Nvidia and Trainium clusters. His pointed line — Google invented MLPerf but doesn’t submit TPU results to its own benchmark — captures the skepticism. Crucially he paired it with “I would never short Google or Broadcom”: scale, data, and distribution advantages survive even if the silicon edge doesn’t.

The position reads accordingly: a meaningful but not maximum-conviction megacap long (~3% of 13F, roughly stable since Q4 2025), sized well below Amazon despite Alphabet’s larger AI-revenue base — consistent with a fund that thinks the TPU cost advantage is eroding at the margin while the Gemini/Search franchise remains formidable.

Position History

QuarterTypeShares/NotionalValue% of 13F
Q4 2024Common443,621$83,977,4551.84%
Q1 2025Common189,077$29,238,8670.89%
Q2 2025Common950,859$167,569,8824.65%
Q3 2025Common550,260$133,768,2062.61%
Q4 2025Common514,339$160,988,1071.97%
Q1 2026Common555,023$159,602,4143.19%

A genuine whipsaw: cut ~57% into Q1 2025, rebuilt 5x in Q2 2025 (ahead of the Gemini 3 cycle and Baker’s December scaling-laws commentary), then trimmed ~42% in Q3 2025 and held roughly flat since. The share count has been broadly stable for three quarters while the fund’s narrative on Google’s silicon advantage deteriorated — suggesting the residual position is about the model/data/distribution franchise, not the TPU edge.

Risks

  • Accelerator-edge erosion (Baker’s own May 2026 critique): if TPU’s cost-per-token advantage fades versus Nvidia and Trainium, the vertical-integration premium compresses.
  • Search disruption: AI chat interfaces (including Google’s own AI Mode) can cannibalize the highest-margin ad real estate ever created; monetization parity is unproven.
  • Capex supercycle: $180–190B in 2026 and more in 2027 — returns on that capital depend on inference demand that is still contract-backlog, not realized revenue.
  • Antitrust: US remedies in the Search and ad-tech cases, plus EU DMA enforcement, could force distribution or structural changes.
  • Position-sizing signal: Atreides has already shown willingness to cut this name >50% in a quarter; it is a trade-sized holding, not a core compounding bet like Astera or Unity.

Sources

Backlinks

Graph View