A Beer, a Garage, and a Billion-Dollar Problem
The idea that would become a $1.64 billion company started over a beer in San Diego in 2018.
Seshu Madhavapeddy and Surya Ganti had just completed work on a fingerprint sensor at Qualcomm, where Madhavapeddy served as VP/GM of the in-display fingerprint sensor business. Ganti, a former senior research scientist at GE — one of the world’s largest jet engine manufacturers — floated an unconventional idea: what if the principles used to cool aircraft engines could be miniaturized to cool computer chips?
By July 2018, the pair set up shop in Ganti’s garage in Los Altos, California, and began building Frore Systems — a name derived from “frozen” and “frosty.” Eight years and $340 million in funding later, that garage experiment has produced the world’s first solid-state active cooling chip and a liquid cooling system that leading GPU manufacturers are integrating into their AI data center architectures.
The Problem: AI is Overheating
The timing of Frore’s founding was prescient, even if the founders could not have predicted the specific catalyst that would make cooling a billion-dollar market.
AI training and inference workloads have pushed chip power consumption to levels that existing thermal solutions struggle to manage. NVIDIA’s latest GPU generations consume hundreds of watts per chip. Rack-level power densities in AI data centers are reaching 100+ kilowatts — far beyond what traditional air cooling was designed to handle. Estimates suggest that cooling accounts for 30-40% of total data center energy consumption.
The industry’s response has been a frantic search for better thermal solutions. Traditional fans are noisy, bulky, and increasingly inadequate. Standard liquid cooling systems — while more effective — involve complex plumbing, leak risks, and maintenance overhead. Industry analysts project the global data center cooling market to reach the $20-40 billion range by the end of the decade, depending on the scope of technologies included.
Frore is attacking this problem from two angles.
AirJet: The Solid-State Revolution
Frore’s first product, AirJet, is the world’s first solid-state active air-cooling chip. Unlike conventional fans, which use rotating blades to push air, AirJet uses vibrating membranes — technology adapted from Ganti’s jet engine research at GE — to create high-velocity air currents directly over chip surfaces.
The result is a cooling solution that is silent, contains no moving parts that can break, is dustproof and water-resistant, and fits into form factors as thin as a few millimeters. For edge devices — laptops, tablets, industrial IoT systems — AirJet enables AI workloads that would otherwise cause thermal throttling.
AirJet was the product that first demonstrated Frore’s technology. It attracted early attention from NVIDIA, whose CEO Jensen Huang received a demonstration approximately two years before the Series D and suggested the team develop liquid-cooling options for data centers.
LiquidJet: Scaling to Data Centers
That suggestion became Frore’s second major product line. LiquidJet is a multi-stage 3D short-loop jetchannel liquid cooling coldplate designed specifically for AI data center GPUs.
The technical innovation lies in the manufacturing process. Rather than traditional machining, Frore builds LiquidJet using semiconductor manufacturing techniques adapted to metal wafers, creating 3D microstructures that optimize coolant flow at the chip level. The result is higher GPU performance, improved Power Usage Effectiveness (PUE), and reduced Total Cost of Ownership (TCO) compared to conventional liquid cooling coldplates.
Frore has demonstrated LiquidJet compatibility with NVIDIA’s Blackwell Ultra GPUs and has showcased a prototype coldplate capable of handling up to 1,950 watts of thermal design power (TDP) for NVIDIA’s upcoming Rubin GPU generation. A newer LiquidJet design targets up to 4,400 watts for the even more demanding Feynman generation beyond that.
The third product, LiquidJet Nexus, integrates multiple LiquidJet units into a unified system that eliminates hoses, connectors, and manifolds — enabling half-U compute trays and dramatically simplifying data center plumbing.
Advertisement
The Funding Arc
Frore’s journey to unicorn status followed a classic deep-tech trajectory: long development periods punctuated by accelerating investor interest as the market caught up to the technology.
- Seed through Series B: Early funding supported R&D and the development of AirJet for edge devices.
- Series C ($80-90 million, 2024): Accelerated manufacturing and began the pivot toward data center applications.
- Series D ($143 million, March 2026): Led by MVP Ventures with participation from Fidelity Management & Research, Top Tier, Mayfield Fund, Clear Ventures, Addition, Qualcomm Ventures, StepStone Group, and Alumni Ventures. This round brought total funding to $340 million and valuation to $1.64 billion.
Qualcomm Ventures’ continued participation is notable. While Frore was not an official Qualcomm spinoff, the company’s founding team and early technology emerged directly from Qualcomm’s R&D ecosystem. The corporate venture arm’s ongoing investment signals conviction that Frore’s technology is best-in-class.
Fidelity’s participation — a crossover investor more typically associated with late-stage and public market positions — suggests that Frore may be on a path toward public markets.
The “Thermal Stack” Thesis
In announcing the Series D, Frore introduced the concept of the “Thermal Stack” — positioning cooling not as a commodity component but as foundational infrastructure for the AI era, analogous to the software stack or the network stack.
This is a meaningful reframing. Data center operators have historically treated cooling as a cost center — something to be minimized rather than optimized. Frore’s argument is that as AI chip power consumption continues to increase (NVIDIA’s GPU roadmap shows no plateau), thermal management becomes a performance bottleneck that directly limits AI capability.
If cooling determines how fast a GPU can run, how densely chips can be packed, and how much energy a data center consumes, then it is not an accessory — it is a core layer of the infrastructure stack. This thesis is what investors are underwriting at a $1.64 billion valuation.
Competition and Context
Frore is not the only company pursuing advanced cooling for AI data centers. CoolIT Systems, Vertiv, and Asetek have established liquid cooling businesses. Jetcool Technologies and Iceotope are pursuing immersion cooling. NVIDIA itself has invested in cooling partnerships across its supply chain.
What distinguishes Frore is its semiconductor manufacturing approach — using wafer-level fabrication techniques to create cooling systems rather than conventional plumbing. This gives the company a manufacturing advantage at scale and positions its products as components rather than systems — making them easier to integrate into existing data center architectures.
The company is headquartered in Silicon Valley with manufacturing operations in Taiwan — the same geography that houses TSMC, NVIDIA’s primary chip fabricator, and the densest semiconductor supply chain on earth.
What to Watch
Frore’s unicorn valuation prices in significant expectations. The company must demonstrate that LiquidJet can win design slots with major GPU and server OEMs, that it can scale manufacturing to meet data center demand, and that its thermal performance advantage holds as competing solutions improve.
The company’s trajectory from garage experiment to unicorn in eight years is a reminder that the AI boom extends far beyond model builders and cloud providers. The physical infrastructure layer — chips, cooling, power, networking — is producing its own generation of high-value companies. Frore’s bet is that you cannot have AI without managing heat, and that managing heat is a harder problem than most investors realize.
Frequently Asked Questions
Why is cooling such a critical issue for AI data centers specifically?
AI workloads consume far more power per chip than traditional computing. NVIDIA’s latest GPUs draw hundreds of watts individually, and rack densities in AI data centers reach 100+ kilowatts — several times higher than conventional server racks. Traditional air cooling cannot remove heat fast enough at these densities, causing chips to throttle performance or risk damage. Cooling now accounts for 30-40% of total data center energy consumption, making it both a performance bottleneck and a major cost driver.
How does Frore’s approach differ from traditional liquid cooling?
Traditional liquid cooling uses machined metal plates with simple channels to flow coolant over chips. Frore uses semiconductor manufacturing techniques — the same wafer-level fabrication methods used to build chips — to create 3D microstructures within its LiquidJet coldplates. This enables more precise coolant flow at the chip level, handling up to 4,400 watts of thermal load. The approach also produces components (not systems), making integration into existing data center architectures simpler.
What does Frore’s success signal about the broader AI investment landscape?
Frore’s $1.64 billion valuation demonstrates that the AI boom extends far beyond software companies building models and applications. The physical infrastructure layer — chips, cooling, power, networking — is generating its own unicorns. This “picks and shovels” layer tends to produce more durable businesses than the application layer because infrastructure demand persists regardless of which AI models or applications win. For investors and entrepreneurs, the signal is clear: the AI value chain includes hardware and infrastructure opportunities that most market participants overlook.
Sources & Further Reading
- Another Deep Tech Chip Startup Becomes a Unicorn: Frore Hits $1.64B — TechCrunch
- Frore Systems Becomes $1.64B Unicorn — Frore Systems Press Release
- Fidelity, Qualcomm Back Frore’s $143M Liquid Cooling Bet for NVIDIA GPUs — TechFundingNews
- Taking the Heat: How Frore Systems’ New Cooling Chip Unleashes Your Device’s Power — Mayfield
- Frore Systems Demos LiquidJet Coldplate With Up To 1950W TDP Cooling on NVIDIA’s Rubin GPUs — WCCFTech















