Stellate memo

Stellate is an edge-caching layer for SaaS application databases, founded in late 2023 by ex-Vercel and ex-Cloudflare engineers who built and shipped cache invalidation systems at production scale at their prior companies. The company is at $1.2M ARR with 25% month-over-month growth across 47 paying SaaS customers, claimed 138% net revenue retention, and verifiable production deployments at Notion, Linear, and Ramp. Sequoia is leading the $5M Series A at a $45M post-money valuation; Hemrock would participate to fill out the round at the current allocation of $1.25M for a 2.8% target ownership stake. The thesis is that edge data infrastructure is the next layer to consolidate after edge compute (Cloudflare Workers, Vercel Edge Functions), and the founding team has both the technical depth and the proximity to the buyer (SaaS infra engineering leads) to win it.^[1][2][3][4]

The single largest unresolved issue is a 60% discrepancy between the company-stated burn rate ($400K/mo in the deck) and the financial model's implied two-year average ($650K/mo derived from the $18M projected 2027 ARR and stated hiring plan). Either burn ramps aggressively to fund the engineering hires required to support the projected logo growth, or the revenue plan is materially more conservative than the deck headline. A second-order concern: 138% NRR is at the top of the comparable range — PlanetScale and Upstash report 120-130% at similar scale — and is the assumption most load-bearing for the year-3 projection. Both items must be addressed in the partner conversation before a commitment can be made.^[1][2]

[Partner to complete]

Stellate provides a database query caching layer that runs at edge points-of-presence and sits between the application server and the origin database. SaaS engineering teams drop in the Stellate SDK without modifying their existing database access layer, and the system automatically caches read queries against a low-latency edge store with TTL-based invalidation and a write-through path that pushes mutations back to the origin database. The company's stated value proposition is a 30-40% reduction in monthly Postgres infrastructure spend plus measurable latency improvement for read-heavy SaaS workloads. The current customer base is 47 paying SaaS companies, anchored by Notion, Linear, and Ramp as flagship logos with public mentions of Stellate in their engineering blogs.^[1][2]

The edge data infrastructure category is roughly a $5B addressable spend today (third-party estimates from Battery and Bessemer market maps), growing at 35-40% annually as more SaaS applications adopt globally distributed architectures. The category has produced at least two billion-dollar outcomes in the last 36 months — PlanetScale (acquired-talked at $1B+ implied valuation), Cloudflare R2 (public Cloudflare segment). Direct competitors include PlanetScale (edge-replicated database with caching), Upstash (edge Redis as a service), and Tigris (globally distributed object storage with caching primitives). Stellate did not name any of these in the deck — they positioned Cloudflare as a primitive provider rather than a competitor, which is defensible but also reveals that they're thinking of themselves as a vertical product on top of a platform. Tailwinds: AI-driven query growth, increasing latency sensitivity at the application layer, and continued consolidation toward edge-first architectures.^[1][2]

Three-person founding team: Alex Chen (CEO), Priya Mehta (CTO), and Devon Rhodes (VP Engineering). Alex Chen, age 32, spent four years at Vercel from 2020-2024 on the Edge Network team, joining as a senior engineer and exiting as a tech lead; prior to Vercel he was an early engineer at Hashicorp on the Consul team for two years. Priya Mehta, age 34, was a staff engineer at Cloudflare from 2019-2024, working on the Workers KV product and previously the cache-invalidation subsystem of Cloudflare's CDN; before Cloudflare she spent three years at Akamai. Devon Rhodes, age 29, was a senior engineer at Vercel alongside Alex from 2022-2024, focused specifically on the cache-coherency layer of the Edge Network; prior to Vercel he was at Snowflake on the compute-storage separation work. All three hold BS Computer Science degrees — Alex from Stanford, Priya from CMU, Devon from Waterloo.^[1][2]

Founder-market fit is unusually direct: Alex's last project at Vercel was an internal cache-invalidation tool that Vercel decided not to productize, and Stellate is effectively the externalized version of that work with a different business model. Priya owned cache invalidation at Cloudflare during the period when Cloudflare scaled Workers KV from beta to GA — she has shipped this exact category of system at production scale serving trillions of requests per day. Devon's work on Vercel's cache-coherency layer overlaps precisely with the consistency guarantees Stellate's SDK provides today. Across the three of them, this is a team that has built and operated cache invalidation systems at three of the most demanding shops in the industry (Akamai, Cloudflare, Vercel) for a combined ~10 years before starting the company.^[1][2]

Public output is substantial for a team this size. Alex co-authored Vercel's widely-referenced 2023 engineering blog post "How we invalidate the edge cache" which has been cited in 14+ infra-engineering talks at QCon and Strange Loop. Priya was a co-author on the Cloudflare Workers KV consistency model write-up (2022) and gave a keynote on edge consistency at All Things Distributed 2023. Devon maintains an active Github with non-trivial contributions to the open-source `tigerbeetle` distributed database project. None of the three are first-time founders, but none have founded a venture-backed company before either.^[1][2]

Partner Q&A surfaced two notable reference data points. From the Vercel reference call (verified via former colleague): Alex's product-engineering judgment was specifically called out as the differentiator — "Alex is the kind of engineer who would push back on a feature spec, build a prototype to prove the alternative, and ship the better version on time. Most engineers do at most two of those three." From the founder-market-fit Q&A: when asked what he'd do if Stellate failed, Alex answered "I'd go back to building cache invalidation systems somewhere else, probably as an early engineer. This is the only company I'd start" — which reads as a founder whose identity is bound to the problem domain rather than to the act of company-building.^[1][2]

[Partner to complete — character assessment, founder-market fit judgment, and overall team score]

Company-stated traction is $1.2M ARR, 25% MoM growth, 138% NRR, and 81% gross margin, with the ARR figure broken out as 47 paying customers averaging ~$2.1K MRR per logo. Notion and Linear are verified production users — both reference Stellate in their public engineering blog posts and reference checks with engineering leadership at both companies confirmed live deployment serving production traffic. Ramp is verified via the company's public customer page but reference outreach was not completed in time for this draft. The 138% NRR claim could not be independently verified within the diligence window; comparable peer companies PlanetScale and Upstash report 120-130% NRR at similar scale per their public S-1-style filings and analyst notes, so 138% is plausible but at the top end of the comparable band and would need direct review of cohort data to confirm.^[1][2][3][4]

Subscription pricing structured as a base platform fee per workspace plus usage-based overage on cached query volume above the included monthly quota. Gross margin is claimed at 81%, which is consistent with peer infrastructure businesses operating at similar scale (PlanetScale reported 78% at their last public disclosure). The financial model projects $18M ARR by end of 2027 — a 15x increase from the current $1.2M base in 24 months, driven by a combination of new-logo acquisition (compounding at the stated 25% MoM rate) and expansion within existing customers (138% NRR). Reaching the projection requires both growth rates to hold simultaneously at scale, which has not been demonstrated for the category. Sensitivity analysis is the most important missing piece of the model — partner should request a downside scenario at 100% NRR and 12% MoM growth.^[1][2]

Direct competitors include PlanetScale (edge-replicated database with caching, $1B+ implied valuation, recent acquisition-talked status), Upstash (edge Redis as a service, last private round at $400M), and Tigris (globally distributed object storage with caching primitives, Series B). Stellate did not name any of these in the pitch deck and positioned Cloudflare as a platform primitive rather than a competitor. The omission is worth flagging — most competitive maps generated by partners on cold review will include these three immediately. The defensible position Stellate articulated is integration ergonomics: customers report adopting Stellate in under a day without refactoring their database access layer, where competitors typically require a several-week migration or a wholesale rewrite. That switching cost is asymmetric (high for customers leaving Stellate, low for customers joining), which creates a meaningful one-time switching cost moat but not a structural data or network-effect advantage.^[1][2]

Three open questions before commitment. First, the burn-rate contradiction: the deck shows $400K/mo, the model implies $650K/mo on a two-year average — partner should ask Alex directly which is correct and where the delta sits. Second, revenue concentration: the top-10 customer concentration is not disclosed in the data room, and for a 47-customer business it could materially affect the durability of the 138% NRR claim if a single large customer is driving the headline number. Third, Bain Capital Ventures (seed lead) is not taking their pro-rata in the Series A; this could be a benign signal (full fund, internal IC dynamics) or a real concern about thesis, and the partner should specifically ask Alex why Bain stepped down. Beyond the three open questions, the structural risks are concentration of the founding team in a single technical area (all three are cache infrastructure engineers — no obvious GTM or sales leadership on the founding team yet) and the moat being primarily a switching cost rather than a structural advantage.^[1][2][3]

The core product is a single npm package (`@stellate/cache`) plus a managed control plane. Developers wrap their existing database client and Stellate transparently caches qualifying read queries at the nearest edge POP — currently 18 regions on Cloudflare's network. The control-plane UI shows cache hit rate, p50/p95/p99 latency, and per-query cost savings; the data-room screenshots show real metrics from a customer with a 73% hit rate on a Postgres workload that previously cost $42K/month. Two reference customers (Notion and Linear) confirmed via public engineering blogs that the integration was deployed in under a day. The defensibility argument is integration ergonomics — once customers adopt the SDK pattern, the switching cost to a competitor (PlanetScale, Upstash) requires re-architecting the database access layer, which is a one-time engineering cost that compounds as the customer's data grows.^[1][2]

Base-case scenario assumes Stellate executes the stated $18M ARR plan by end of 2027 and continues compounding into a $50-80M ARR business by year 7 (2030), which would be a credible Series C/D outcome. Recent comparable exits in edge infrastructure: PlanetScale at ~$1B implied (M&A talks, 2024); Vercel at $3.25B (last private round, 2024 — broader infra but adjacent); Cloudflare R2 segment estimated $200M+ revenue contribution. At $50-80M ARR, a strategic acquirer at 10-15x ARR multiple implies a $500M-$1.2B exit. At our entry valuation of $45M post and a $1.25M check for 2.8% ownership, the base case returns ~3-7x gross on our position over a 5-7 year horizon.^[1][2]

Upside scenario: Stellate becomes the default edge data layer for the next generation of SaaS, reaching $150-250M ARR by year 7 with continued 50%+ growth, supporting an IPO or strategic acquisition at $3-5B. Comparable: PlanetScale at peak private valuation, MongoDB Atlas at the equivalent revenue scale. At those outcomes our position returns 15-30x gross. Downside scenario: growth stalls below 100% NRR after the current customer base saturates, the year-3 projection comes in at $5-7M ARR rather than $18M, the round following this one is flat or down, and the company is either acquired for talent at $50-100M or shuts down. At the talent-acquisition floor, our position returns roughly 1-2x; at shutdown, the position is a write-off. Partner judgment fields below capture the actual underwriting decision.^[1][2]

[Partner to complete — base-case target multiple, upside thesis, downside floor, target ownership]