Groq announced a $650 million funding round led by Disruptive and Infinitum to accelerate its cloud inference platform expansion. The company plans to grow its processing capacity to 200 megawatts by 2027, scaling a service that already handles trillions of tokens per week for millions of developers worldwide.

What Makes Groq's Inference Chip Different?

Groq has built its business around a specialized processor called the LPU, or Lateral Processing Unit, designed specifically for AI inference workloads. In December, Nvidia licensed Groq's underlying technologies and created the Nvidia Grok LPU 3, which debuted in March as part of a liquid-cooled system called the LPQ. This partnership represents a significant validation of Groq's chip architecture from the world's dominant AI hardware maker.

The LPU 3 includes several technical advantages that speed up how AI models generate responses. The chip features 500 megabytes of onboard SRAM, a high-speed memory that processes data faster than the off-chip RAM used by other AI accelerators. Additionally, the processor includes 92 data lanes, each capable of moving information at 112 gigabits per second, totaling 2.5 terabits per second of bidirectional bandwidth. For context, this means data moves between chips nearly instantaneously, reducing the delays that typically slow down AI inference.

One innovation addresses a common problem in large AI clusters: clock drift. When multiple accelerators fall out of sync with each other, data traffic slows down and AI model response times suffer. The LPU 3 includes an automatic clock-drift correction feature that prevents this bottleneck from occurring.

How Is Groq Scaling Its Cloud Platform?

• Current Scale: Groq's cloud platform operates across 13 data centers spanning multiple continents and processes trillions of tokens per week for 5 million developers.
• Growth Target: The company plans to reach 200 megawatts of processing capacity by 2027, with some new power coming from the LPQ liquid-cooled appliance that Nvidia debuted in March.
• Funding Use: The $650 million raised will be deployed directly toward expanding inference capacity and supporting the company's cloud infrastructure growth.

What Competitive Advantages Could Groq Develop?

While other cloud operators could theoretically build their own LPQ-powered inference services, Groq has an opportunity to differentiate itself by expanding beyond raw infrastructure. The company could extend its platform with higher-level services such as managed databases, following a strategy already employed by competitors like CoreWeave Holdings. This approach would allow Groq to capture more value from its customer relationships and create stickier products that go beyond commodity compute.

The funding round was led by growth investment firm Disruptive and hedge fund Infinitum, signaling strong investor confidence in Groq's ability to execute on its expansion plans. The timing comes seven months after Nvidia's $20 billion chip licensing deal with Groq, which included hiring several key Groq employees, including its founding chief executive.

For developers and enterprises relying on AI inference, Groq's expansion could mean faster response times and more reliable service availability. The company's focus on optimizing inference specifically, rather than training large models, addresses a growing market need as organizations deploy AI applications at scale and seek to reduce latency in production environments.