The tech stack of 40 unicorn SaaS companies.

23 of 40 unicorns front their marketing site with Cloudflare. 12 of 40 use an AWS CDN (CloudFront or S3 directly). The two are within a few percentage points of each other.

The interesting subdivision is who picks which. Companies that started after 2018 and have an aggressive product-engineering culture skew Cloudflare: Notion, Linear, Vercel, Anthropic, OpenAI, Perplexity, Mercury, Ramp, Brex. Companies founded before 2015 with a mature AWS contract skew AWS CDNs: Stripe, Slack, GitHub, Dropbox, Figma, Segment, Datadog, Twilio. Akamai shows up only twice (Airbnb and Mailchimp), both of which are old enough that the Akamai contract probably pre-dates Cloudflare being a serious competitor at their tier.

3 of 40 host their marketing site on Vercel. That number is significant. Five years ago it would have been zero. Vercel has captured a quarter of the unicorn marketing-site market in roughly five years, mostly at the expense of Netlify (only 2 of 40) and self-hosted AWS marketing builds.

14 of 40 sites run a detectable bot-management vendor. Cloudflare Bot Management dominates with 12 deployments. Akamai Bot Manager appears twice. Arkose Labs and GeeTest each show up once (both at Airbnb, which stacks three vendors).

The pattern follows industry, not size. All three frontier AI labs (OpenAI, Anthropic, Perplexity) run bot management because they are the single largest targets for AI training scrapers and competitive intelligence. Fintech follows next: Mercury, Ramp, and Gusto all run it because financial-app marketing sites are heavy targets for credential-stuffing reconnaissance. The contrast group is developer tools, where GitHub, GitLab, Datadog, Dropbox, and Slack run no detectable bot management at all. Their threat profile is different and their tolerance for false-positive friction on a developer-facing marketing site is lower.

Airbnb is the outlier. Three separate bot vendors on one property (Akamai Bot Manager + Arkose Labs + GeeTest) is not architecture for the sake of architecture. It is what a fraud-engineering team builds after a measurable revenue loss has forced them to route distinct attacker populations to distinct challenge types.

Three patterns repeat across the 40 sites. First, the marketing surface has been decoupled from the product surface. The team that owns marketing infrastructure is not the same team that runs the application, and they pick different vendors. AWS for the app, Cloudflare or Vercel for the marketing site. Different cache rules, different deploy cadences, different on-call rotations.

Second, the consensus stack has narrowed. React on the frontend, Tailwind for styling, Next.js if you started after 2020 and Vite if you wanted build speed but not a framework. Six years ago the framework wars were noisy. They have ended in a quiet near-monoculture.

Third, the largest companies are removing trackers, not adding them. The 2010-2020 MarTech narrative was that more pixels meant better attribution and more revenue. The 2024-2026 reality is that the largest B2B SaaS companies have decided the tradeoff is no longer worth it on the marketing layer specifically. Whether server- side replacements catch all the lost signal is an open question. The behavior is already a fact.

The website tells you a lot. Hiring tells you what comes next. JobsPipe surfaces the backend tools, internal infrastructure, and team-side adoption that scanner-only data cannot reach.