Key Facts
- Modular is a Palo Alto-based AI infrastructure company, founded in 2022 by Chris Lattner — creator of LLVM and Swift — and Tim Davis, that reached a $1.6 billion valuation in September 2025 after raising $250 million to challenge Nvidia's CUDA software dominance.
- Who founded Modular: Chris Lattner, the original architect of the LLVM compiler infrastructure and the Swift programming language, and Tim Davis, both former Google colleagues.
- How much has it raised: $380 million total across three rounds: a $30 million seed in 2022, a $100 million Series B in 2023 at a $600 million valuation, and a $250 million Series C in September 2025 at a $1.6 billion valuation.
- What is Mojo: A new programming language, built by Lattner, combining Python's ease of use with the performance of low-level languages like C++, designed specifically to let developers write AI code that runs efficiently across CPUs, GPUs, and other accelerators.
- What is MAX: Modular's AI inference platform, letting developers deploy generative AI models across Nvidia, AMD, and other hardware without rewriting code for each chip.
The Man Who Already Reshaped Software Twice, Now Aiming for a Third
Modular's founding story carries a different kind of credibility than most companies in the CrackTheDeck AI Unicorns cohort, because its co-founder has already fundamentally reshaped the software industry — twice. Chris Lattner is widely known as the original architect of the LLVM compiler infrastructure, a foundational piece of technology that underpins an enormous share of modern software compilation, and as the creator of Apple's Swift programming language, which replaced Objective-C as the primary language for building iOS and macOS applications. Lattner also held engineering leadership roles at Apple, Tesla, and Google before co-founding Modular in 2022 alongside Tim Davis, a former Google employee he met while working there.
That background matters directly to how Modular has positioned itself: rather than a team of machine learning researchers trying to solve an infrastructure problem from the outside, Modular is led by someone whose entire career has been built around designing the low-level compiler and language infrastructure that other software is built on top of — arguably the exact skill set needed to tackle the specific problem the company set out to solve.
The Problem: AI Software Is Fragmented, and Nvidia's CUDA Owns the Fix
Modular's founding motivation, according to EE Times' 2023 coverage of the company's product launch, stemmed directly from Lattner's "deep frustration about the fragmentation in AI software infrastructure". That fragmentation problem is specific and well understood within the AI engineering community: different AI chips — Nvidia GPUs, AMD GPUs, Intel and AMD CPUs, custom accelerators like TPUs and ASICs — each require different low-level software to run AI models efficiently, and Nvidia's CUDA platform has become the de facto standard specifically because it offers the most mature, best-optimized software layer for its own hardware, creating a powerful lock-in effect: developers write code optimized for CUDA, which only runs well on Nvidia chips, which reinforces Nvidia's pricing power and market dominance regardless of whether a competing chip might otherwise offer better raw performance or value.
Reuters' coverage of Modular's September 2025 funding round frames the company's ambition explicitly around breaking that lock-in: Modular "has developed a platform that enables developers to execute their AI applications on various computer chips without the need to rewrite code for each specific chip," and the funding was raised specifically "as it aims to challenge Nvidia's dominance". The Information's earlier reporting, from when the company was raising its Series A at a far more modest $600 million valuation, described Modular even more directly as "an AI startup challenging Nvidia" — software "endeavoring to disrupt Nvidia's stronghold on high-demand artificial intelligence processors" by simplifying "the process for AI developers to train and execute their machine-learning models on chips produced by alternative firms such as AMD, Intel, and Google".
Mojo: A New Programming Language Built for the AI Era
Modular's most technically ambitious product is Mojo, a programming language Lattner designed specifically to bridge what Wikipedia's technical entry describes as "the gap between Python's ease of use and the fast performance required for cutting-edge AI applications". Mojo builds on the Multi-Level Intermediate Representation (MLIR) compiler framework — a newer, more flexible compiler technology than the lower-level LLVM framework used by many other languages including Swift, C++, and Rust — specifically because MLIR allows Mojo to target not just CPUs but GPUs, TPUs, ASICs, and other accelerators through a single unified language.
In an extended technical conversation on the Latent Space podcast, Lattner elaborated on his motivation, describing his effort to give "developers precise control over how AI workloads hit silicon" while still preserving Python's approachable syntax and ergonomics — attempting to solve what he describes as the fundamental tension between programmer productivity and the extremely detailed hardware-level control needed to extract maximum performance from modern AI accelerators. Modular open-sourced the Mojo standard library in March 2024, inviting community contributions under the Apache 2.0 license, and published a formal roadmap toward a 1.0 release in December 2025, ultimately releasing Mojo 1.0.0 beta 1 in May 2026 alongside a dedicated language website.
MAX: The Platform Where Mojo Actually Gets Deployed
Mojo doesn't operate in isolation — it's one component of Modular's broader MAX (Modular Accelerated Execution) platform, first launched at the company's inaugural developer conference in December 2023. EE Times' coverage at launch described MAX as a suite of three integrated modules: the MAX Engine, a unified library for lowering AI models from frameworks like TensorFlow or PyTorch down to whatever specific hardware is running them; the Mojo language itself, used to extend the engine with custom operations; and a MAX Serving module, a unified framework for deploying models into cloud production environments. At that same launch, Modular announced partnerships with both AWS, to bring the MAX platform to Amazon's production cloud services, and Nvidia itself, to enable support for Nvidia GPUs and CUDA code alongside the platform's existing CPU support for Intel, AMD, and AWS's own Arm-based Graviton processors — a notable detail, since it means Modular's strategy isn't purely adversarial toward Nvidia; the company positions itself as a hardware-agnostic layer that works well on Nvidia's chips too, rather than an explicitly anti-Nvidia alternative that sacrifices performance on the dominant hardware platform in the process.
Modular's own current product materials describe MAX as GenAI-native serving infrastructure capable of deploying models including DeepSeek, Gemma, and Qwen through a high-speed, OpenAI-compatible endpoint, running on Nvidia or AMD hardware across any cloud provider, while explicitly noting that "MAX doesn't depend on PyTorch, CUDA, or ROCm, so there's nothing to bundle, patch, or keep in sync" — a direct pitch to enterprise infrastructure teams tired of managing the complex, often brittle dependency chains that come with stitching together multiple AI frameworks and hardware-specific software layers.
Funding Trajectory: From a $30 Million Seed to a $1.6 Billion Series C
Modular's valuation growth traces a steady, multi-year climb rather than the explosive same-year sprints seen elsewhere in this series. The company raised a $30 million seed round in June 2022, the same year it was founded. A $100 million Series B followed in August 2023, valuing the company at approximately $600 million. Two years later, in September 2025, Modular closed a $250 million Series C led by the U.S. Innovative Technology Fund, with continued participation from existing backers DFJ Growth, GV (Google Ventures), General Catalyst, and Greylock, at a $1.6 billion post-money valuation — nearly tripling the company's worth in roughly two years, and bringing total funding since founding to $380 million.
That roughly three-year path from founding to unicorn-tier valuation is faster than deeply patient infrastructure builds like d-Matrix's six years, but considerably slower than the same-year sprints seen among companies like Unconventional AI — consistent with the genuine multi-year technical maturation required to build both a new programming language and a production-grade AI inference platform simultaneously, work that inherently can't be compressed the way a narrower application-layer product sometimes can.
Revenue and Commercial Traction
Independent revenue tracking from Latka places Modular's 2025 revenue at $49.7 million in annual recurring revenue, describing "consistent revenue growth" since the company's 2022 launch. Sacra's analysis details the underlying business model: Modular operates as a B2B infrastructure software company with a consumption-based revenue structure, where "customers pay based on compute resources consumed and inference requests processed through the MAX engine, similar to cloud computing pricing but optimized for AI workloads" — a pricing model that ties Modular's revenue growth directly to how much actual AI inference workload its enterprise customers run through the platform, rather than to flat software licensing fees.
Modular's own customer-facing materials cite a partnership described by an unnamed source as helping "supercharge" the company's "mission for our millions of AWS customers, helping them bring the newest GenAI innovations and traditional AI use cases" to production — reinforcing the AWS relationship first established at the company's 2023 launch as a continuing, central go-to-market channel.
Why Breaking CUDA's Lock-In Is Worth a Billion-Dollar Bet
Modular's core strategic thesis rests on a specific structural observation about the AI infrastructure market: Nvidia's dominance isn't purely a function of having the best silicon — it's substantially reinforced by CUDA, the deeply entrenched software layer that makes switching away from Nvidia hardware technically painful even when a competing chip might offer a better price-to-performance ratio for a specific workload. If Modular's MAX platform and Mojo language succeed in giving developers genuinely comparable performance across Nvidia, AMD, and other hardware without requiring separate, hardware-specific code for each, it would meaningfully weaken the software-driven lock-in that currently protects Nvidia's pricing power — a prospect valuable enough to attract not just venture capital but continued strategic interest from cloud providers like AWS and chip competitors seeking exactly that kind of software-layer disruption to Nvidia's dominant position.
Lattner has been candid in public interviews about just how difficult that challenge remains. In a June 2025 podcast appearance, he discussed at length "the CUDA monopoly" and the specific technical work required to match "NVIDIA performance with AMD" — acknowledging directly that matching, not just approximating, Nvidia's performance on rival hardware is the genuinely hard technical bar Modular has set for itself, rather than simply providing a lower-performance, more portable alternative that trades away the raw speed enterprises actually need for production AI workloads.
FAQ
Who founded Modular? Chris Lattner, the original architect of the LLVM compiler infrastructure and the Swift programming language, and Tim Davis, both former Google colleagues.
How much has it raised? $380 million total across three rounds: a $30 million seed in 2022, a $100 million Series B in 2023 at a $600 million valuation, and a $250 million Series C in September 2025 at a $1.6 billion valuation.
What is Mojo? A new programming language, built by Lattner, combining Python's ease of use with the performance of low-level languages like C++, designed specifically to let developers write AI code that runs efficiently across CPUs, GPUs, and other accelerators.
What is MAX? Modular's AI inference platform, letting developers deploy generative AI models across Nvidia, AMD, and other hardware without rewriting code for each chip.
What is Modular's business model? A consumption-based model, similar to cloud computing pricing, where customers pay based on compute resources consumed and inference requests processed through the MAX engine.
Source: CrackTheDeck Research.