Scale AI - The $7.3Bn Worth AWS for AI
As the AI finds more use cases, Scale is up there with Nvidia & Open AI (one of Scale's customers) to massively benefit from it, let's look at what makes Scale such an important startup in this sector
Editor’s Note: Thank you to 47 new people who decided to subscribe to our newsletter, including execs from Amex, Vercel, and Instacart. Term Sheet Digest is a weekly newsletter for the Attack Capital community. If you are in the mood to read about private company stocks that are heating up in the secondary market and the biggest news around Silicon Valley, you are in the right place.
Founded by Alexandr Wang & Lucy Guo, Scale wants to build a foundational infrastructure of AI/ML, just like AWS built the infrastructure of the internet.
Scale was founded in 2016, the company is currently valued at $7.3Bn after raising $603Mn in total from blue chip investors including Y Combinator, Accel, OpenAI’s CTO Greg Brockman, Thrive, Tiger Global, Founders Fund.
Alexandr Wang realized that most of his classmates at MIT didn’t like building AI tools. The reason, large amounts of data that are needed to work on the tools need a lot of cleaning, sorting, and organization.
Alexandr realized this problem is universal, from small startups to enterprises, and students on the college campus face the same problem.
While interning at Quora, he met Lucy Guo, who was also thinking about the same problem, and they decided to start Scale, a data labeling and annotation platform for anyone building Large Language Models.
Scale built its product proposition around the data engineering part of the AI lifecycle. Companies need data annotation and labeling of “ground truth” data. The truth means correctly labeling data in an expected format, a simple task such as tagging a picture of a table as a “table” and helping to differentiate a “table” from a “chair” in an image.
Scale Data Engine for Building AI
The company started with data annotation, but since then, it has become a one-stop-shop for building LLMs. It offers data labeling solutions for 3D, Image, mapping, text, and audio. The company also found Crème de la Crème customers who believe in its vision - Open AI, Harvard Medical School, Toyota Research Institute, and Nuro, the autonomous delivery vehicle manufacturer are some of them.
How Nuru uses Scale to make their Autonomous Vehicles much Safer on Roads
Nuru realized it needed to make their recognition system more robust to continue delivering commercial goods through their autonomous vehicle without compromising on saftey, they needed help with identifying:-
Pedestrians in unusual poses.
Animal cases—horses, large birds, small non-pet animals.
Occluded and backlit pedestrians.
Infrequently encountered vehicles such as excavators
Some of these cases were recognized by the Nuru’s internal tools, but not at the scale for Nuru to get confident in the ability of its model. ML models only deliver the highest accuracy when they can handle edge cases that might be challenging, uncommon, or even dangerous, for Nuru the 50-60 cases that their model was identifying wasn’t enough.
Nuru with the help of Scale’s Nucleus, a tool which allows for data exploration, debugging of bad labels, comparing accuracy metrics of different versions of ML models, and finding failure cases solved this issue.
Their ML model was able to identify 10X+ critical edge cases than they were able to find with their internal tools.
Scale’s other Offerings
Some other offering as part of their data engine focuses on delivering an all-in-one solution for building LLMs. Beginning with annotation and labeling, Scale also offers Reinforcement Learning from Human Feedback (RHFL), which is needed to fine-tune models.
The company also offers a suite for dataset management, which takes care of testing, model evaluation, and model comparison.
Scale’s Solution for Applying AI
Every company is now experimenting with Generative AI solutions. But, to build AI tools, there are a lot of roadblocks:-
Customization Requirements: Enterprises have proprietary data and unique needs that require extensive fine-tuning of base foundation models and prompt engineering.
Observability and Reliability: Broadly trained base models are black boxes that often hallucinate, responding to users with false, harmful, or unsafe results requiring continuous evaluation and monitoring by experts.
Security and Safety Risks: Cloud models can leak your proprietary data, IP, PII, and model interaction history and pose other security and safety risks, requiring infrastructure hosted within the enterprise VPC and role-based access controls (RBAC).
Scale lets Enterprises Compare, test, and deploy foundation models from OpenAI, Anthropic, Google, and more. The company also fine-tunes base models with its enterprise data and Scale's data engine.
Enterprises can Build, compare, and securely deploy generative model applications with Scale’s developer platform, Spellbook. They can also use Scale’s turnkey applications, Chat, Forge, and E-commerce AI, or build custom apps to accelerate their businesses.
Scale’s current customer base shows how the company has slotted itself into the lifecycle of AI applications. To build an AI product, you need a foundational model. Scale helps companies who are building foundational models like OpenAI, Co:here and fine them.
On the other hand, there are enterprises like Microsoft and Toyota and startups like Instacart Brex that need those foundational models to build their own AI applications. Scale found a way to build tools for them as well and gains from an increase in demand from both sides.
Scale AI & Ukraine War
When the invasion of Ukraine started, most of the humanitarian groups relied on first-person data to find places that had been damaged, displacing the habitants that might need help.
In just a month since the Russian Invasion, Scale developed Automated Damage Identification Service, a machine learning tool that algorithm developers can use to rapidly train AI to automatically detect new damage to buildings and structures.
Scale’s tool also offers geotagged links to attack reports, helping humanitarian groups target affected areas. Scale offered this service to Ukrainian and NATO operations for free.
Scale’s tool has analyzed more than 770 square miles of Ukraine and identified over 370,000 structures, all of them concentrated in the cities of Kyiv, Kharkiv, and Dnipro. That data can then be used by government officials and NATO partners to understand the scope of the destruction and allocate resources accordingly.
Secondary Markets
Scale is one of the first companies that started building pure-play AI solutions. The company is backed by some very prominent investors - Y Combinator, Accel, Tiger Global, and Coatue Management, to name a few. The company has raised a total of $603Mn.
Scale, since the AI hype, is finally becoming one of the most important companies, especially due to its focus on infrastructure and not just applications. There is no doubt that its secondary stock is in high demand, but most of the late-stage investors do not want to sell its stocks due to the AI hype. The company is expected to show even more growth in the coming years, and it will likely be at least the top three companies that will reshape the AI infrastructure of the future.
Top Stories
Walmart has spent $3.5 billion this year to acquire shares of Flipkart, the Indian E-commerce platform, illustrating just how aggressively it’s betting on India at a time when its chief global rival Amazon is scaling back on its expenditures in the South Asian market.
Intel and Tower ink major foundry deal, $300M investment after Intel cancels its $5.4B Tower acquisition; Intel was planning to acquire Tower Semiconductor, but the deal received major pushback from regulators.
Arm’s high-end IPO share price of $51 would put the British chip maker’s valuation north of $52 billion. The world’s largest hardware makers have adopted its architecture. Arm currently counts Apple, Google, and Samsung among its clients, along with leading component manufacturers Nvidia and AMD.
Please share this newsletter with your friends, family, and colleagues if you feel that it’ll add value to their lives - probably the best compliment we can get :)