Investment Memo: Perplexity
Prepared for hypothetical Series A consideration (acknowledging the company is in reality far past this stage; the exercise is to evaluate the core bet as if at entry)
---
Thesis
Search is the single largest monetizable behavior on the internet — Google generates ~$175B+ annually from it — and it is being unbundled by LLMs for the first time in 25 years. Perplexity is the purest bet on that unbundling: a consumer-grade "answer engine" that replaces ten blue links with a cited, synthesized response. If even 2-3% of global search intent migrates to answer-first interfaces and Perplexity holds a meaningful share of that migration, this is a fund-returner. The bet is not that Perplexity out-innovates Google on models; it's that an incumbent with a $175B cash cow cannot cannibalize itself fast enough, and that a focused challenger can own the "research-grade query" segment before the giants reorganize.
Product & wedge
The product is a conversational search interface that retrieves live web results, synthesizes an answer with inline citations, and supports follow-up questions. The wedge is trust and speed on complex, multi-source queries — the 15-20% of searches where ten links genuinely underserve users (research, comparison shopping, technical questions). Citations are the key design choice: they mitigate hallucination anxiety and differentiate from raw ChatGPT. The company has layered on Pro subscriptions ($20/mo, model choice, file upload), an Enterprise Pro tier, an API (Sonar), shopping/commerce features, and the Comet browser — signaling ambition to own the full query-to-action loop rather than remain a destination site.
Notably, Perplexity is model-agnostic — it orchestrates GPT, Claude, and its own fine-tuned models. This is either smart capital efficiency or a structural lack of moat, depending on your priors.
Market & competition
TAM is effectively the search advertising + knowledge-work software market: hundreds of billions. Competition is the most brutal imaginable:
- Google: AI Overviews and Gemini shipped directly into the search page, with distribution to billions and the deepest index on earth.
- OpenAI: ChatGPT with search is the largest consumer AI product in history; every Perplexity feature is a candidate for absorption.
- Microsoft/Bing Copilot: distribution via Windows and Edge.
- Anthropic, Meta AI: adjacent conversational assistants with search capabilities.
- You.com, Brave Search, Kagi: smaller direct analogues.
Perplexity's position: better-executed product for power searchers, but no proprietary model, no proprietary index at Google's scale, and no owned distribution channel — hence the browser and mobile-carrier/device preload deals (e.g., Motorola, Deutsche Telekom partnerships) to buy distribution.
Traction & business signal (public information only)
- Reported ~15M+ MAU (2024 figures); company has cited hundreds of millions of queries per month, growing fast. Exact current DAU/retention: unknown.
- Reported ARR crossed ~$50M in 2024, with later reports suggesting ~$100M ARR run-rate; precise figure and margin structure: unknown, and inference costs likely make gross margins thin.
- Raised successive rounds at rapidly escalating valuations (reports of $9B in late 2024, subsequent reports of $14B–18B+ discussions), backed by IVP, NEA, NVIDIA, Jeff Bezos, SoftBank.
- Churn, paid conversion rate, enterprise seat counts, and CAC: unknown publicly.
- Notable negative signal: multiple plagiarism/scraping disputes (Forbes, News Corp/Dow Jones lawsuit, Wired allegations) — a legal and reputational overhang on the core retrieval mechanism.
Risks (the three that kill the deal)
1. No structural moat against feature absorption. Perplexity sits on rented models and rented distribution. ChatGPT search and Google AI Overviews are the same product with 100–1000x the users. The honest counterargument — "focus and product velocity" — has historically lost to distribution in search (ask Neeva, whose superior product died and whose team now runs Snowflake's AI). Perplexity must convert product taste into a durable asset (index, browser install base, habit) before the giants' products reach parity. The window is 18–36 months.
2. Unit economics of answering are worse than the economics of linking. Google's genius was making users do the last mile while ads paid for the index. Perplexity pays inference cost per query and monetizes a fraction of users via subscription; its ad product is nascent and unproven. If cost-per-query doesn't fall faster than query volume grows, growth is a treadmill. Margin data: unknown — that opacity is itself a yellow flag at these valuations.
3. Legal exposure at the foundation. The product's value is synthesized publisher content. Active litigation (News Corp) and publisher hostility could force revenue-share economics that crush margins, or injunctions that degrade the index. Google survived this fight with scale and cash; a startup may not.
Verdict
Invest at a true Series A (i.e., at early-2023-equivalent entry price), sizing for a power-law outcome: the market is enormous, the team ships faster than anyone in the category, and even the acquisition floor (Apple, Meta, and others have reportedly circled) protects the downside. Pass at current reported valuations ($14B+), where you're paying for the Google-disruption outcome with none of the moat questions resolved.
The one question I'd need answered first: What is 6-month cohort retention for free users, and how has it trended since ChatGPT search launched? Retention against a free, ubiquitous substitute is the single number that distinguishes a durable habit from a leaky bucket filled by paid distribution — and it determines whether everything else in this memo matters.