AI chip startup Groq is racing to build out a global network of inference data centers, with CEO Jonathan Ross confirming plans to add more than a dozen new sites in 2027 on top of the 13 the company already operates across North America, Europe, the Middle East and Asia-Pacific as of mid-2026. The expansion marks one of the most aggressive infrastructure pushes yet by a challenger to Nvidia’s dominance in AI computing, this time focused not on training massive models but on serving them quickly and cheaply once they are built.
A Different Bet Than the GPU Giants
Groq makes what it calls Language Processing Units, or LPUs, chips purpose-built for AI inference rather than the general-purpose GPUs that Nvidia sells by the millions. Ross has argued repeatedly that Groq’s architecture can produce and deploy chips faster while consuming meaningfully less power than GPU-based systems, a pitch that has resonated with customers in markets where electricity and land are constrained. The company says its GroqCloud platform now serves more than five million developers, a figure it has touted as evidence that inference speed, not just raw training capacity, is becoming a competitive battleground in its own right.
Middle East and Asia Take Center Stage
Much of Groq’s near-term expansion is aimed at Saudi Arabia and India. Ross told CNBC in late 2025 that Saudi Arabia is positioned to become an AI data center hub given its energy surplus, and Groq has since said it secured a $1.5 billion commitment tied to a new GroqCloud data center in Dammam. In Asia, Ross has pointed to strong momentum in India specifically, crediting Groq’s lower power draw and faster deployment timelines for traction there. The company already has a footprint in the US, Canada, Europe and the Middle East, and executives say the next phase of growth will push deeper into the Asia-Pacific region.
Fresh Capital and a Rising Valuation
The expansion follows a reported $650 million raise in May 2026 that pushed Groq’s valuation past $6.9 billion, according to people familiar with the fundraising cited by TechCrunch. Investors have been drawn to the inference-focused pitch at a moment when AI labs and enterprises are spending heavily to run already-trained models in production, a category of spending some analysts argue is growing faster than spending on training new models from scratch. Groq’s stated target is to scale toward 200 megawatts of capacity by the end of 2027, a figure that would still leave it far smaller than hyperscale operators but large enough, the company argues, to matter in the inference market specifically.
Skeptics Point to a Crowded Field
Not everyone is convinced Groq’s approach will scale. Rivals including Cerebras and SambaNova are pursuing similar inference-optimized architectures, and Nvidia itself has moved to blunt the threat from specialized chip challengers by deepening commercial ties with some of them rather than competing head-on in every niche. Critics of the inference-chip narrative note that Nvidia’s own newer chips have closed much of the efficiency gap that once favored dedicated inference hardware, and that building a global data center footprint is capital-intensive even for a company flush with fresh funding. Some industry analysts also caution that demand for AI inference, while growing, remains difficult to forecast precisely, and that overbuilding capacity in markets like Saudi Arabia and India carries real financial risk if usage does not materialize as projected.
The Case for Staying the Course
Groq’s supporters counter that the company has already demonstrated it can sign large commercial commitments, pointing to the Saudi deal and its developer base as evidence of real demand rather than speculative bet-making. They argue that as AI applications shift from experimental chatbots to embedded features running billions of queries a day, the economics of inference speed and power efficiency will matter more, not less. Ross has framed the buildout as a land grab of sorts, arguing that establishing infrastructure early in markets like India and the Gulf states gives Groq a durable advantage before larger, slower-moving competitors can catch up.
What Comes Next
The coming year will test whether Groq’s expansion plans translate into sustained revenue growth or strain the company’s balance sheet. With more than a dozen additional data centers planned for 2027, Groq will need to keep signing large commercial and sovereign deals at a pace similar to the Saudi commitment to justify the buildout. Investors and rivals alike will be watching whether Groq’s developer base continues to grow at its current rate, and whether the broader inference market proves large enough to support multiple well-funded specialized chip makers alongside Nvidia’s continued dominance in AI computing overall.