Image of MrBeast clipboard icon showing that 1.5 million users were simultaneously supported

Explore the agentic architecture that delivered MrBeast’s $1M puzzle hunt

How Salesforce delivered a platform that could handle 10 million sign-ups and 1.5 million simultaneous players — in just six weeks.

The Results

10M

registration capacity

1.5M

simultaneous users supported

tokens generated in the first 72 hours

150M

verification emails/day capacity

Super-Bowl Scale in Six Weeks

When Salesforce partnered with MrBeast for a Super Bowl commercial, the goal was ambitious: Turn a 30-second ad into a live, $1 million agentic puzzle hunt during the biggest advertising moment of the year.

The scale was unprecedented. The experience was projected to consume 1 trillion tokens in its first 72 hours — roughly the processing equivalent of reading every word on the entire English-language internet. Standard API limits of 200,000 requests per minute simply wouldn’t be enough.

Salesforce needed infrastructure capable of supporting up to 10 million registrations and 1.5 million concurrent users without disruption — and it all had to be designed, built, and deployed in just six weeks, four times faster than the typical six-month timeline.

Millions of Players, One Entry Point

Salesforce Professional Services led the ambitious execution, starting with the layer that would face the most pressure first: the entry point. They designed it as a series of protective layers that could absorb massive traffic spikes while keeping the core platform stable:

A high-performance landing layer: The static landing page was performance-tested to handle 400,000 visitors per second, with 90% of traffic served via Akamai NetStorage as static HTML. By delivering most content outside the core platform, the team kept nearly all early traffic completely off the Salesforce back end.
A controlled entry point during peak demand: An Akamai waiting room, tested to 500,000 visitors per second, acted as a pressure valve during the Super Bowl surge. It throttled inflow to 85,000 users per minute, ensuring downstream systems could scale safely without overwhelming the database.
A protective pre-login buffer: Once through the waiting room, users entered a custom AWS pre-login layer built with CloudFront, S3, API Gateway, and a web application firewall. This layer verified identity and absorbed the initial wave of authenticated traffic before requests ever reached Salesforce.
High-volume email verification: To support player sign-ins, AWS SES was scaled to deliver 150 million verification emails in 24 hours, supported by a 14-day IP warming strategy to ensure high deliverability during the surge.

These layers transformed what could have been a chaotic traffic spike into a carefully managed funnel — allowing players to move from curiosity to gameplay without disruption.

Seamless Gaming Across Repeat Logins

Once players passed the entry layer, they arrived at registration, where AWS Lambda functions routed incoming users across 78 identical Salesforce shards, each capable of processing 10,000 registrations per minute.

Volume was only half the challenge. To keep players’ progress consistent, a DynamoDB UserShardRegistry permanently assigned every player to a single shard. That way, no matter how many times someone logged in, they always returned to the same environment with their game intact. Without it, a returning player could have been routed to a different shard with no record of their progress at all.

Data was distributed across more than 100 partitions using a composite key structure to prevent database failures under millions of simultaneous writes. MuleSoft tied together Salesforce Core Platform, Experience Cloud, LLM Gateway, Slackbot, and Tableau, so players experienced one seamless game despite the complex systems running underneath.

To manage costs at this scale, the platform only updated player data when new information was recorded, cutting storage and syncing overhead by 80%. A Redis cache between Slackbot and Data 360 achieved a better-than 90% hit rate, reducing API calls from 500 per minute to just 50. What felt like a smooth, continuous experience to players was the result of an architecture built to handle massive volume and keep every journey frictionless from first login to final guess.

Real-Time Mission Control

With millions of players active at once, the team needed full visibility to ensure the infrastructure held up and to resolve issues before players noticed. Tableau was embedded directly into the self-service interface built with Experience Cloud, giving designers a real-time mission control dashboard to monitor player velocity, shard health, and queue release rates.

If players were solving puzzles too quickly, difficulty could be adjusted on the fly to keep the experience running for the intended three to four weeks. LLM request limits were pushed from 500 per minute to 10,000 through custom prompt optimization and collaboration with Salesforce product engineering.

The platform went from zero to 1.5 million concurrent users and 10 million total registrations in just six weeks, generating 1 trillion tokens in just the first 72 hours.

Beyond this one-time event, the platform’s performance sets a new standard for enterprise and B2C customers. “We were able to increase our total capacity from 200,000 requests per minute to 2 million requests per minute,” said William Dressler, Regional VP, AI and Data Architecture. “This improvement in throughput allows us to support B2C-scale customers, giving them the volume they need on the platform.”

Why MrBeast Chose Salesforce

Extensible Platform

Because Salesforce’s Agentic Enterprise architecture is designed as a flexible, extensible platform, the team was able to combine first- and third-party tools into a single, cohesive system. This flexibility allowed them to meet unprecedented scale and complexity, customizing the architecture to support millions of players without rebuilding core systems from scratch.

Salesforce Professional Services

Salesforce Professional Services made architectural decisions in real time, anticipated security threats, and built never-before-needed safeguards. The ability to blaze new trails for what the platform can do — under pressure, on an impossible timeline — was essential to bringing this ambitious idea to life.