Yesterday my post on Nvidia's $20B Groq acquisition blew up. I got the regulatory arbitrage angle right. I got the employee compensation angle completely wrong.
I was wrong.
Everyone gets paid at the $20B valuation: VCs, employees, founders, the whole cap table. 85% upfront, the rest by end of 2026. No one got left behind.
Good news, right?
Not exactly. Nvidia bought a patent arsenal that will likely weaponize through NPEs (Non-Practicing Entities) to create a "scorched-earth zone" around SRAM-based inference. Nvidia gets protected. Everyone else gets sued.
And GroqCloud? The deal confirms it's now a shell company running at 10% employee capacity with no IP, no technical leadership and 2.5 million developers wondering who's actually running the service now.
Let's get into it.
The Actual Deal Structure
So the $20B breaks down as follows:
Groq shareholders (VCs, employees, everyone):
- 85% paid up-front
- 10% paid mid-2026
- 5% paid end of 2026
- All at the $20B valuation
90% of Groq employees joining Nvidia (~550 people):
- Vested shares: paid in cash immediately
- Unvested shares: paid at $20B valuation in Nvidia stock (on a vesting schedule)
- For around 50 employees the entire stock package is being accelerated and paid in cash (who are these people? Early employees? Key engineers? Nvidia's not saying)
Let's pause on that 90% number. That's roughly 550 people moving to Nvidia. This isn't "hiring a few key engineers" as most of us suspected. For context: Instagram was 13 people for $1B, WhatsApp was 55 people for $19B. This might be the largest acquihire in tech history by headcount. At $20B for 550 people, that's $36M per person. Nvidia isn't messing around.
The remaining 10% of employees staying at GroqCloud:
- Vested shares paid in cash
- "Economic participation in the ongoing company" (more on this in a second)
- Anyone at the company less than one year gets their vesting cliff dropped
So Chamath's Social Capital still makes ~$2.0B (some speculate up to $4.0B) but good news everyone else makes money too. No one got left behind.
Why $20B?: The Patent Scorched-Earth Strategy
I covered the basics in my previous post. Nvidia gets the team, the implementation knowledge of making SRAM-first inference work at commercial scale and neutralizes a competitor threatening GPU dominance in low-latency inference.
I completely missed a very important angle:
Groq's patents will likely end up in an NPE (Non-Practicing Entity) portfolio. Strategic patent trolls is the technical term I believe.
Here's how that works: GroqCloud still exists as a shell company with patents. Those patents eventually get sold or transferred to an NPE that specializes in patent litigation. The NPE doesn't build products they simply sue people for patent infringement.
Now imagine Google, Amazon, or any other company tries to build SRAM-based inference chips. The NPE (armed with Groq's patents) sues them. The target company has two options:
- Pay a (potentially/very likely) exorbitant licensing fee
- Abandon the SRAM inference approach entirely
Nvidia, of course, already HAS non-exclusive licensing rights. They're protected. Everyone else faces a patent minefield.
This creates a "scorched-earth zone" around Nvidia's position.
Either way, GroqCloud as an independent entity isn't the point. It's a side effect. The remaining 10% are there to keep customer contracts alive while Nvidia decides what to do with the implementation knowledge they just acquired and while the patent portfolio becomes a competitive moat.
The GroqCloud Problem
Wait a second though. Somethings off. Nvidia licensed all the IP, 90% of employees are leaving FOR NVIDIA (with all the juicy knowledge in their brains), what remains at GroqCloud?
Let's do the math on what's left:
- People: 10% of the original workforce (maybe 50-70 people based on sources).
- IP: Nish. Nvidia has non-exclusive licensing rights to everything.
- Leadership: CEO Ross and President Madra are at Nvidia. New CEO is Simon Edwards, who was the CFO.
- Technical team: If 90% left, most of the engineering team is gone.
- Product: GroqCloud was the cloud inference business built on proprietary LPU architecture.
The remaining GroqCloud entity has no IP, no technical leadership, 10% of the workforce, and a CFO running it (with all due respect).
What business are they in?
GroqCloud's Big Boy Customers
In February 2025, Saudi Arabia committed $1.5 billion to expand Groq's Dammam data center. Groq built the region's largest inference cluster serving "nearly four billion people regionally adjacent to the KSA" according to Ross during the Series E announcement.
GroqCloud operates 13 facilities across the US, Canada, Europe, and the Middle East. Someone has to run these, fulfill the Saudi contracts, and serve existing customers.
We can assume that is what the remaining 10% are doing. Operating infrastructure for existing contracts while Nvidia figures out the transition plan.
But without IP rights and without the engineering team that built the LPU compiler and architecture, GroqCloud can't innovate. Yes they can keep the lights on, but they can't compete. They're a managed services company running someone else's (Nvidia's) technology.
And those employees who stayed were promised "economic participation in the ongoing company." Participation in what, exactly? LOL? A cloud provider with no competitive moat, no technical team, and no IP? That equity is worth whatever Nvidia decides it's worth when they eventually wind down or absorb GroqCloud entirely.
But Wait, It's Not Just Saudi Arabia
The Saudi deal is the most visible. Right before the acquisition, Groq was serving 2.5 million developers and a lot of Fortune 100 companies had accounts on the platform. They projected $500M in revenue for 2025.
Here's what GroqCloud had going on:
Meta/Llama Partnership (April 2025): Groq became the official inference provider for Meta's Llama API, serving 1.4 million developers and Fortune 500 companies. They got exclusive rights to launch Llama 4 in the Middle East.
IBM Partnership (October 2025): IBM and Groq announced a strategic partnership to integrate GroqCloud into watsonx Orchestrate. Two months before the acquisition.
HUMAIN (Saudi Arabia): HUMAIN runs their entire HUMAIN ONE operating system on Groq infrastructure.
Enterprise Hardware: Aramco Digital ordered hundreds of GroqRacks (each rack has 64-576+ LPU chips). GroqRacks are deployed at hyperscalers, sovereign clouds and regulated industries.
Now imagine you're one of these customers. It's December 2025. You just integrated GroqCloud into your stack, or you're planning your 2026 roadmap based on Groq's LPU architecture.
Then you find out:
- 90% of the engineering team is leaving for Nvidia
- The CEO and President are leaving for Nvidia
- All the IP is licensed to Nvidia
- Your point of contact is now... a CFO and 10% of the original workforce?
What team is left to help you? When you file a support ticket who answers? (Don't say an ai agent.) When you need a feature for your enterprise deployment, who builds it? (AGAIN, DO NOT SAY AN AI AGENT.) When the Llama API needs an update to support Llama 5, who does the integration work? (...)
If you're IBM and you just bet on GroqCloud integration in October, what's your Q1 2026 plan? If you're one of 2.5 million developers building on the GroqCloud API, do you start looking at alternatives?
GroqCloud can keep the servers running. They can maintain existing deployments. But can they innovate? Can they compete? Can they even keep up with the pace of change in AI when the team that built everything is gone?
To Conclude (Again)
Everyone gets paid :D! The implementation IP goes to Nvidia :/. GroqCloud becomes a zombie that serves existing customers until it doesn't :(.
This was a quick follow-up based on new information. If you have questions or especially corrections, shout me again ossamachaib.cs@gmail.com.