Groq taps Aramco for massive AI compute hub

Silicon Valley chip startup Groq just pulled off a major coup in the desert.

While Gulf sovereign funds have spent the past year struggling to get their hands on Nvidia GPUs due to strict U.S. export controls, Groq found another way in: partnering with oil heavyweight Aramco to build what it claims will be the world’s largest AI inference center in Saudi Arabia.

Aramco Digital will fund the project with an investment "in the order of nine figures," while Groq operates the facility. The data center is slated to go live by the end of this year, starting with 19,000 Language Processing Units (LPUs) and potentially scaling up to 200,000 units over time.

Speed, cheap power, and 4 billion users

Groq, founded in 2016 and recently valued at $2.8 billion following a $640 million funding round led by BlackRock, has built its reputation entirely on blistering inference speeds. Its LPUs run open-source language models at hundreds of tokens per second.

According to Groq CEO Jonathan Ross, the Saudi facility will take advantage of ultra-cheap domestic energy, ample land, and strategic geography: within a 100-millisecond ping, the hub can reach roughly 4 billion people across the Middle East, Africa, and India.

The facility targets processing billions of tokens daily by late 2024, expanding to hundreds of billions daily by 2025 to support millions of regional developers.

Ross also made a point to address regulatory headwinds, noting Groq has been completely transparent with the U.S. Commerce Department about its Riyadh deployment and does not anticipate export hurdles, partly because Groq does not sell to Chinese entities.

Why it matters

We think this partnership is a masterstroke in regulatory and competitive positioning.

For regional developers across the Middle East and South Asia, local access to ultra-fast inference could dramatically lower the barrier to deploying real-time conversational AI, without pinging distant servers in North America or Europe. If you are building voice bots or high-throughput agents in the region, Groq's low-latency footprint could soon be worth testing immediately upon rollout.

For the AI chip market, Groq is outflanking Nvidia by changing the playing field. Trying to beat Nvidia at frontier model training is practically impossible right now, and Washington's export bans make shipping high-end GPUs to the Gulf an uphill battle. But inference on specialized chips is a different ballgame. By focusing purely on inference and letting Aramco shoulder the massive nine-figure capital expenditure, Groq is securing massive physical distribution without burning through its own balance sheet.

The risk, however, lies in software flexibility and hardware scaling. Groq's architecture relies heavily on fast SRAM, which excels at high-speed token generation for models like Llama but requires specialized compilation. Scaling smoothly from 19,000 chips to 200,000 is a monumental hardware hurdle for a startup. Whether Aramco's bet can attract sustainable commercial workloads beyond regional vanity metrics remains the real test.