a16z Partner Martin Casado Announces Partnership with TypeSafe to Advance Jev
Martin Casado stated in a public post that the software era is far from over and announced a deep collaboration with the TypeSafe team to enhance software quality, emphasizing a long-term positive outlook on developers, programming, and the Jev model.
Casado, head of infrastructure at Andreessen Horowitz (a16z), noted that this collaboration corresponds to a16z leading TypeSafe's Series A financing round, amounting to approximately $870 million, with a valuation of $7.5 billion. Casado will join the board, with Sequoia and DCVC participating in the round. TypeSafe was founded in 2024 by Diogo Almeida (former OpenAI researcher and co-creator of RLHF and ChatGPT), Sasha Sheng, and Erik Gafni, and had previously completed a $40 million seed round led by DCVC, with a valuation of $200 million.
Jev was launched on September 15, positioned as a System One model, designed for internal software decision-making: it receives unstructured states and typed questions (Choice, Score, Noul) and directly returns structured values with probabilities and confidence levels, rather than generating text. The company claims an end-to-end latency of 70 to 500 milliseconds, with input pricing at approximately $0.042 per million tokens and free output, making it 20 to 200 times faster and 40 to 400 times cheaper than leading LLMs in decision tasks.
After its launch, Jev adopted a rapid pace: processing 1 trillion tokens within three days, with daily service volume reaching trillions of tokens, and approximately 29.4% of Fortune 500 companies integrating it. Some reports indicate an annualized revenue run rate of $100 million within seven days, making it the fastest adopted model in Vercel AI Gateway's history. TypeSafe claims to have saved clients millions of dollars in production costs and plans to expand more machine-native models and enterprise features.
In the a16z announcement, Casado pointed out that true economic unlocking lies in embedding the latest model intelligence into all software in a way that aligns with traditional software economics in terms of performance and cost. Jev achieves this goal and is referred to as the "Jevons Paradox"—when intelligence becomes cheap enough, it will be invoked ubiquitously. The company emphasizes that AI should enhance software rather than merely automate the replication of existing software.
Driven by events, capital is flooding into decision-making AI infrastructure, with a16z and Sequoia leading investments signaling a shift of funds from purely generative models to composable, type-safe software primitives; developers and SaaS companies relying on reliable automation will benefit in the short term from lower latency and costs, while traditional high-cost LLMs face pricing and usage pressures in repetitive judgment scenarios.
Source: Public Information
ABAB AI Insight
Martin Casado previously promoted OpenFlow and software-defined networking at Stanford, co-founded Nicira, which was acquired by VMware for $1.26 billion in 2012, and later expanded the network and security business to an annual revenue of $600 million. After joining a16z in 2016, he focused on infrastructure investments, covering AI development tools like Cursor and Convex. He has long advocated that the fundamentals of computer science are superior to framework stacking and has publicly criticized the long-term overspending and delays in software projects.
On the capital path, a16z is injecting large-scale funds into TypeSafe, positioning Jev as an intelligent primitive that can be directly consumed by code, motivated by the goal of shifting intelligence from human dialogue interfaces to internal software calls, thereby re-anchoring software economics in an era where marginal costs approach zero. Resources are shifting from model training computing power to decision calibration, type safety, and parallel calling infrastructure.
This is similar to how SDN abstracted network control from proprietary hardware into programmable software, and it is akin to how database or operating system primitives empower the application layer; the current AI industry is transitioning from an expansion phase centered on generative text to a control and embedding phase centered on composable decision primitives, with TypeSafe attempting to occupy a standard position in the internal judgment layer of software.
Essentially, this represents a technological substitution: generative LLMs are being partially replaced by type-safe probabilistic models in structured, auditable, low-latency decision scenarios, as text output requires additional parsing and incurs high costs, while directly returning typed values eliminates the parsing layer and supports parallel calls, transforming intelligence from an expensive conversational commodity into software infrastructure that can be invoked like functions, driving the industry chain from a model-centric to a primitive and composable-centric reconstruction.
ABAB News · Cognitive Laws
When intelligence becomes cheap, it will shift from dialogue to functions.
The bottleneck of software shifts from writing code to making decisions.
Type safety is the lock that keeps illusions in a cage.