TMTB: Citi Conference Takeaways: MDB, AMD, NOW, ROKU, ON BKNG, ON, STX, CDW, MTCH

MDB

Presenter: CEO and CFO

Key Takeaways

• Atlas re‑acceleration: Atlas growth re‑accelerated to ~29%; strength came from large enterprise workloads (U.S. and Europe) and healthier workload quality post go‑to‑market shift; SMB adds via self‑serve are up and many new customers are AI‑native startups.

• AI today: not a major revenue driver yet; inference use cases fit MongoDB’s OLTP/document model; several notable AI customers exist but are not yet material.

• Platform stance vs. Postgres: relational market consolidating on Postgres, but MongoDB argues Postgres’ tabular/single‑node heritage and non‑native JSON create scaling limits; MDB’s platform bundles search, vector, and embeddings in one place.

• Voyage embeddings: embeddings are the bridge to private data; Voyage Model 3 improves long‑document understanding and precision; sold stand‑alone and inside MongoDB to widen the funnel.

• Relational migration: building automation (schema remap, data move, AI code refactoring) to ease moves off legacy databases—driven by cost, EOL risk, regulator pressure, and the need to AI‑enable apps. More detail at Investor Day.

• Efficiency & partners: small restructuring, but continue investing; Azure and hyperscaler ties remain strong; MDB expects multiple winners in a massive market.

Q&A Highlights

• Workload is the unit of competition; workloads often grow then plateau—so consistently acquiring high‑quality, strategic workloads is key.

• Self‑serve engine is an efficient top‑of‑funnel; go‑to‑market mix will keep tilting up‑market where productivity is highest.

---

AMD

Presenter(s): Jean Hu (EVP & CFO); Matt Ramsay (Corp VP, Financial Strategy & IR).

Key Takeaways

• Quarterly cadence, not annual targets: AMD emphasized near‑term guidance over long‑range revenue forecasts; Q2 revenue was $7.7B (+32% y/y) with Q3 guided to $8.7B (+28% y/y); Data Center Q2 at $3.2B (+14% y/y) despite MI308 export limits to China. MI350 ramp drives double‑digit sequential DC growth in Q3; MI400 launches in 2026 and MI500 in 2027.

• Supply is tight but secured: Advanced-node wafer capacity and HBM remain constrained; AMD says it’s secured critical wafers/memory with TSMC and memory vendors to support 2026 ramps.

• Sovereign AI pipeline: Collaboration with HUMAIN (multi‑billion opportunity) and 40+ active sovereign engagements; expect more activity in 2026 as licensing progresses.

• AI GPU margins: Data Center GPU margins are below corporate today (slightly dilutive) as AMD prioritizes share and TCO for customers; expects gross‑margin expansion with scale.

• Customer concentration: Medium‑term concentration persists given hyperscaler capex; AMD counts 7 of top 10 AI spenders as customers; consumption will diversify via cloud and “neocloud,” but invoiced customers may stay concentrated.

• China MI308: Wrote off $800M of inventory; has a license but isn’t starting new wafers; will consider China‑compliant next‑gen only with licenses.

• ASIC vs GPU mix: AMD still sees ~20–25% of AI TAM served by ASICs; expects programmable GPU‑led systems to address the balance.

• TAM framing: Company points to >$500B AI TAM by 2028; avoids frequent TAM updates, focusing instead on annual roadmap cadence and TCO.

Q&A Highlights

• Why pull back on long‑range forecasts? It’s early in the ramp; better to provide quarter‑ahead updates and highlight roadmap cadence (MI350→MI400→MI500).

• Supply readiness for 2026? AMD cites tight but secured advanced wafers and HBM; prepared for rack‑scale deliveries.

• Tier‑2/3 training strategy: Using MI355 to onboard customers into production‑level training ahead of Helios/MI400, de‑risking “train‑to‑train.”

• Pricing/BOM: Memory content rises each gen; ASPs rise with BOM while preserving TCO; memory % of BOM not disclosed.

• “Bubble/overbuild” risk? AMD views AI as early‑stage; hyperscaler results show ROI; capacity remains constrained across compute (including CPUs).

---

ROKU

Main presenter: Dan Jedda, CFO.

Key Takeaways