Qualcomm Acquires Modular: A Strategic Advancement in AI

Qualcomm and Modular: A Significant Acquisition in the AI Landscape

Last week was marked by two significant developments in the field of artificial intelligence, illustrating how tech companies are seeking to master the rising costs and complexities associated with AI computing. Qualcomm, the San Diego-based giant, announced the acquisition of Modular, a software startup located in Palo Alto, California. Modular specializes in simplifying the execution of AI models across various types of computing chips, a skill that is becoming increasingly crucial as AI diversifies.

In parallel, reports revealed that SambaNova, a chip-focused startup, is finalizing an $800 million funding round led by General Atlantic, valuing the company at $10 billion. These two transactions highlight a growing reality in the tech sector: as hardware remains scarce and expensive, the software layers that connect these chips are becoming as valuable as the silicon itself.

The Role of Dave Munichiello in These Developments

Dave Munichiello, managing partner at GV (Google Ventures), is closely monitoring these developments. He has led early investments and holds board seats at Modular and SambaNova. Munichiello brings pragmatic operational experience to tech investing, having served as a captain and paratrooper in the U.S. Army before transitioning to the private sector. He then worked as an early executive at Kiva Systems, contributing to the evolution of the warehouse automation company until its acquisition by Amazon for $775 million.

With a background in mathematics and computer science from Emory University and an MBA from Harvard Business School, Munichiello has dedicated his venture capital career to core software infrastructure, developer tools, and data systems, including early support for companies like Slack, GitLab, and Segment.

In a recent interview, he discussed the mechanics behind the Qualcomm-Modular deal, the practical realities of managing hardware scarcity, and what the current wave of consolidation means for the future of independent startups.

Qualcomm and Modular: A Software Decoupling Strategy

Qualcomm's acquisition of Modular underscores a strong desire to decouple AI software from hardware fragmentation. This strategy reflects a shift in ultimate value within the AI stack, moving from proprietary hardware architectures to developer-friendly software layers capable of operating in any computing environment.

According to Munichiello, the types of hardware required for AI are becoming increasingly heterogeneous. Initially, only Nvidia GPUs seemed necessary, followed by those from AMD and other players. Now, hardware is moving towards disaggregated inference, an approach that involves separating the different computations used for various parts of the response to a question when interacting with a model.

It seems increasingly likely that there will be three types of chips used in disaggregated inference: an AI-specific chip, a CPU, and a GPU. For a player like Qualcomm, all three components are present, so they need a software layer that connects them. Elsewhere, including at Nvidia, companies typically sell themselves with CPUs and accelerators, but there isn't really a software solution that works across all these elements.

Investing in AI Infrastructure: A Long-Term Vision

Munichiello began investing in AI infrastructure and semiconductors as early as 2016, starting with a company called Lattice, which was sold to Apple and integrated into the Siri team. Subsequently, GV invested in Determined AI, co-founded by Evan Sparks, which was sold to HPE and became an important part of its stack. HPE has actually become the computing partner for OpenAI and has worked closely with CoreWeave.

GV has also been interested in semiconductors long before the current wave, leading the Series A funding round for SambaNova. Munichiello met this company when it had only three people and a slide deck. GV led this round in December 2017, after Lip-Bu Tan led the initial investment, and Munichiello has been on the board since. This initial investment was $15 million at a valuation of $480 million.

Consolidation and the Future of Independent Startups

Consolidation in the semiconductor and cloud provider sectors raises questions about the future of independent startups. Munichiello asserts that there is still a path to independent IPOs. Cerebras has beautifully demonstrated this trajectory, and he expresses his satisfaction for Andrew Feldman and his team. There is absolutely a pathway to build large independent companies, as demand for computing is expanding rapidly.

Everyone is looking to find additional capacity by making everything more efficient. Technology often emerges with a huge boom in mass demand and high prices, and then we figure out how to make it cheaper. We are currently in this efficiency stage. The demand for inference is ubiquitous, from medicine and law to programming, customer support, and finance.

We are trying to extract every last bit of value from the chips. This involves using multiple types of chips: using cheaper CPUs when we can, GPUs when we need them, and the most expensive chips only for the most complicated parts of the process.

The Impact of Open-Source Models on the Market

The rise of open-source models is also changing this dynamic. The universe of potential buyers expands even further as open-source models become prolific. In Qualcomm's announcement, they spoke extensively about their enthusiasm for open source—not only to keep Modular open-source but also to ensure that the models are open-sourced. When this happens, instead of companies paying hundreds of millions of dollars to model providers for inference, the companies themselves will own their models and run them on their own hardware.

IPOs and the Future of Tech Companies

Munichiello remains convinced that IPOs are not entirely out of reach for early-stage tech and hardware companies. He cites SpaceX, which is heavily hardware-focused, as a success story. He believes we will see many IPOs in the next six months. He knows of at least 15 to 20 companies planning to go public, promising a very busy period ahead.

In a market where valuations are rapidly multiplying based on technical metrics like chip throughput, Munichiello explains that true traction relies on execution from quarter to quarter, meeting sales demands, and deploying physical systems for customers.

A company becomes very attractive to investors when it delivers a massive volume of technology in production environments—such as data centers for large enterprise brands and devices we all use daily.

Adapting Capital Requirements to Market Realities

GV has a history of supporting foundational technologies long before the current hype cycle around generative AI. Munichiello explains that GV's framework has adapted now that capital requirements for AI infrastructure have exploded. When a startup needs hundreds of millions just to compete at the cutting edge, GV maintains a focus on the team and the relationship without being overwhelmed by the scale of capital.

It has always been complicated to start from scratch and build a significant, generational company. GV does not engage in momentum investing. They seek foundational technologies and substantial companies that can stand on their own.

When GV met Modular, there were only Tim and Chris with an idea, and they convinced the team to accept their $23 million investment. At the time, they were nervous about valuing the company at over $80 million or $90 million, and it ultimately was valued at $155 million in that first round.

GV took a 15% stake in the company right from the start in a round that seemed very bold for that moment in the world. But Modular recruited an incredible team of compiler engineers, began to grow, and built in a field that has become the most strategic in all of AI.

They value different companies based on their specific markets. Some are incredibly capital-intensive and require billions of dollars, which means GV cannot do it alone. As an investor, they need to bring their network and a syndicate of other investors who can write checks for hundreds of millions of dollars.

Software companies can move a bit faster, make more mistakes, and pivot. In hardware, if you tape out a chip and it doesn't work, you're delayed for years and need to raise much more money. It's much more binary when it comes to the physical world. One hundred million dollars goes much further in software because you can always optimize your token usage or engineering to change direction, which is incredibly difficult to do in robotics or hardware.