{"slug": "jonathan-ross-turned-a-3-week-phone-call-into-a-20-billion-nvidia-deal-here-is", "title": "Jonathan Ross Turned a 3-Week Phone Call Into a $20 Billion NVIDIA Deal. Here Is How He Did It.", "summary": "Groq founder Jonathan Ross revealed that a $20 billion deal with NVIDIA was conceived just three weeks before closing, after Jensen Huang converted a GPU purchase order into a platform partnership. The deal, one of the largest in chip history, was enabled by Groq's LPU architecture integrated with NVIDIA GPUs over three to four months of prior engineering. Ross also criticized West Coast venture capitalists for missing the opportunity due to herd behavior, forcing Groq to raise from East Coast funds.", "body_md": "# Jonathan Ross Turned a 3-Week Phone Call Into a $20 Billion NVIDIA Deal. Here Is How He Did It.\n\n### The Groq founder on near-bankruptcy, Grok Bonds, and why the fastest chip in the world almost never got built.\n\n3 weeks before the money hit Groq’s bank account, the $20 billion NVIDIA deal did not exist.\n\nThat is not a rounding error. That is the entire lesson.\n\nJonathan Ross built the TPU at Google, then spent a decade building Groq’s LPU while running out of money, getting rejected by nearly every West Coast VC, and asking his own employees to take pay cuts to survive.\n\nI watched the full 70-minute interview so you do not have to.\n\n#### Here are the ten takeaways that matter:\n\n*Together with Attio:*\n\n“When I think of revenue, I think of Attio.”\n\nShreman Shrestha, Head of Business\n\nGranola’s revenue motion runs on it:\n\n▫️ ** Zero missed leads** and 10x faster access to customer context\n\n▫️ [Lead triage](https://attio.com/?utm_source=ai_corner&utm_medium=newsletter_sponsorship&utm_campaign=ai_corner-Y26) 83% faster\n\n▫️ **5 hours saved a week** with automated updates\n\n### 1. The $20 billion deal that was a rumor three weeks before it closed\n\nEvery big platform deal looks inevitable after it closes. This one looked like nothing three weeks before it did.\n\n“The most interesting part about it is the call where the idea was first floated was about three weeks before money was in the bank.”\n\nGroq went to NVIDIA to *buy* [100,000 GPUs](https://www.the-ai-corner.com/p/ai-inference-engineering-playbook-2026?r=1krivi) for its own data centers. [Jensen Huang](https://www.the-ai-corner.com/p/founder-mental-models-ai-agent-claude-chatgpt-openclaw-2026?r=1krivi) looked at what they had built and converted a purchase order into a platform decision: make the combined GPU-LPU architecture available to every NVIDIA customer. That became [one of the biggest deals in chip history](https://www.the-ai-corner.com/p/cerebras-series-a-deck-teardown-ipo-2026?r=1krivi).\n\n**Why it matters:** the deal was a fast reaction to a working prototype, rather than a negotiation. Ross had integrated GPUs and LPUs for three to four months before that call. The speed only existed because the engineering already existed.\n\n### 2. Why two chips beat one (the 18-wheeler and the van)\n\n“If you were building out a logistics network for the United States, and I told you you could have either 18-wheelers or vans for last-mile delivery, which one would you pick? The answer is both.”\n\nEvery LLM token is a chain of [matrix multiplications](https://www.the-ai-corner.com/p/llm-token-cost-optimization-playbook-2026?r=1krivi). Some steps are compute-bound: the GPU’s home turf, the attention mechanism. Others are memory-throughput-bound: the LPU’s, applying the trained weights. Most competitors split pre-fill and generation across chips. Wrong cut. The correct split lives inside the [decoder layer](https://www.the-ai-corner.com/p/ai-coding-tools-complete-guide-2026?r=1krivi), task by task.\n\n**Why it matters:** there is no single perfect chip, because there is no single bottleneck. You are probably making the same mistake in your own [stack](https://www.the-ai-corner.com/p/give-your-agent-its-own-computer-2026?r=1krivi): one tool for two different problems.\n\n### 3. The West Coast VCs who missed NVIDIA's biggest deal ever\n\n“Typical West Coast VCs are more like lemmings. Typical East Coast VCs all think that they’re smarter than each other.”\n\nOn the West Coast, one pass triggers every other pass. Groq’s early [cap table](https://www.thevccorner.com/p/safe-conversion-calculator-cap-table-tool-2026?r=1krivi) included a few firms that had lost standing, and every subsequent investor passed by association rather than analysis. The company ended up raising from [East Coast crossover funds](https://www.thevccorner.com/p/mckinsey-pyramid-principle-claude-investor-documents-2026?r=1krivi), the capital least influenced by consensus.\n\n**Why it matters:** the firms that do [independent underwriting](https://www.thevccorner.com/p/what-top-vcs-look-for-2026-founder-playbook?r=1krivi) caught the deal that herd capital missed entirely. If your process depends on who else is in the round, you own that miss.\n\n### 4. Grok Bonds: salary cuts instead of layoffs, and 80% said yes\n\nThree weeks of [runway](https://www.thevccorner.com/p/self-improving-fundraising-system-claude-14-steps-2026?r=1krivi) left. A layoff list already drafted. Ross rejected it outright.\n\n“We had it all hands, and we put up World War Two looking pictures of war bonds, and we called it Grok Bonds.”\n\nGroq was pre-product, building a [compiler](https://www.the-ai-corner.com/p/loop-engineering-coding-agents-2026?r=1krivi) that had never existed. Cutting that team meant killing the company. So Ross offered the opposite: [salary for equity](https://www.thevccorner.com/p/rippling-series-f-investment-memo-parker-conrad-2024?r=1krivi), framed as shared sacrifice. **80% opted in.** Half dropped to legal minimum, some from six figures to $50K. Attrition after the announcement: under 10%, beating the pre-crisis baseline. Runway stretched two more months.\n\n**Why it matters:** Ross frames it as putting everyone’s hands on the steering wheel. Passengers panic on a bad road. Drivers stay calm.\n\n### 5. Return on luck: Ross said no twice before running the math himself\n\n“The third time, I’m going to do it myself. I did the arithmetic and determined the performance. Everyone disagreed with me that it was possible. In the end, we ended up hitting exactly those performance numbers.”\n\nOpportunity one: [GitHub’s CEO](https://www.the-ai-corner.com/p/ai-code-review-checklist-2026-failure-modes-prompts?r=1krivi) called asking if Groq’s chips could run [LLM code completion](https://www.the-ai-corner.com/p/how-to-build-your-first-ai-agent-2026?r=1krivi). His team said it would fail, and he let them talk him out of it. Opportunity two went the same way. On the third, he stopped asking and [ran the numbers](https://www.thevccorner.com/p/acquisition-readiness-model?r=1krivi) himself. Hit them exactly.\n\n**Why it matters:** Jim Collins’s Return on Luck thesis says the best companies capture more of the breaks they get, rather than getting more breaks. Ross’s team reasoned from what GPUs could do instead of what the problem required. That is a pattern, and it lives on your team too.\n\n### 6. Reality quotient: play the dominant game, over playing the same game better\n\n“Reality quotient, in its most extreme form, is the ability to choose the dominant game that’s being played. MySpace was focused on number of accounts signed up. Facebook focused on monthly active users. It was the dominant game.”\n\nRoss separates raw intelligence from noticing which metric decides the outcome. Groq’s dominant game: **25 million tokens per second** of deployed capacity. Chip design, software, power costs, data centers, every function could translate its work into [that one number](https://www.the-ai-corner.com/p/ai-agent-memory-context-as-topology-playbook-2026?r=1krivi).\n\n**Why it matters:** picking the wrong [dominant game](https://www.thevccorner.com/p/ai-skills-complete-playbook-templates-prompts-2026?r=1krivi) means winning a contest that decides nothing. MySpace optimized signups. Signups lost.\n\n### 7. The people spec: hire by screening negatives\n\n“The biggest flip in my hiring was when I went from looking for positives, which is what you do when you’re trying to grow talent, to looking for negatives, which is what you do when you’re trying to select talent.”\n\nGroq keeps a versioned [people spec](https://www.the-ai-corner.com/p/ai-engineer-roadmap-production-projects-2026?r=1krivi), like a product spec. Every positive trait has a negative twin: the opposite of return on luck is squandering it; the opposite of poetic design is a product where the user hunts for where to click.\n\n**Why it matters:** [growing](https://www.thevccorner.com/p/growth-accounting-virality-model?r=1krivi) an existing employee means showing a positive path. Hiring means vetting the negatives they carry in. Confuse the two motions and smart, talented people cause organizational damage. One toxic trait spreads to the whole team once inside.\n\n### 8. Stop having one-on-ones: the NVIDIA lesson on politics\n\n“There is no circumstance at NVIDIA where Jensen has one-on-ones with people and tells them one thing.”\n\nRoss learned it the hard way: private conversations produced different understandings of the same decision, and employees compared notes into conflict. A room hearing the same words at the same time skips that problem entirely. His add-on rule: when someone complains about a colleague by email, copy the colleague.\n\n**Why it matters:** politics grows in the gaps between private conversations. Close the gaps and it has nowhere to live. Trust compounds into retention, [Jensen’s whole operating system](https://www.the-ai-corner.com/p/anthropic-30b-arr-passed-openai-revenue-2026?r=1krivi) in one habit.\n\n### 9. \"I intend to\": the two-word fix\n\n“If you express intentional leadership, you say, I intend to do this. People don’t tend to offer their opinion, but if it’s very wrong, and there’s a reason, they will push back.”\n\nRoss took it from Captain David Marquet’s turnaround of the USS Santa Fe, the fleet’s worst-rated submarine. Marquet swapped “should I dive the boat” for “I intend to dive the boat,” and a silently compliant crew started catching genuine errors, including an open hatch. The sub went from worst to first.\n\n**Why it matters:** “should I” invites reflexive pessimism. “I intend to” invites correction only when something is genuinely wrong. Ross lost two return-on-luck opportunities to asked opinions, then applied this to the third. The fix cost two words, the cheapest [leadership upgrade](https://www.the-ai-corner.com/p/one-person-startup-operating-system-2026?r=1krivi) available.\n\n### 10. Manufactured discontent: the fuel that keeps burning\n\n“At the moment, I’m discontent with the lack of compute in the world... Every single person who dies from cancer, every single person who becomes old and infirm and dies, there could come a point where AI comes up with ways to slow aging. All of that I feel is kind of on my shoulders.”\n\nRoss describes a room of people worth hundreds of millions comparing their discontent. The entrepreneurs were restless about something specific: a project, a product, a number. Contentment with the status quo produces zero new companies. Named, quantified discontent does, the same way Edwin Land counted headlight-glare deaths before Polaroid.\n\n**Why it matters:** wealth satisfaction and drive satisfaction are separate variables. Every top performer Ross has met is discontent about something *specific*, rather than everything.\n\n## The Groq Playbook\n\nRoss’s thesis in one line: **speed compounds, in chips, in decisions, and in leadership language.** [The founders who see it first](https://www.thevccorner.com/p/yc-summer-2026-requests-for-startups-ideas?r=1krivi) capture the return-on-luck advantage before anyone else notices the window.\n\n▫️ * Founders:* your job is full-time change management, and rule one is making the change feel like continuity. Ladder every initiative to your one dominant game. And start a\n\n[people spec](https://www.the-ai-corner.com/p/ai-engineer-roadmap-production-projects-2026?r=1krivi)today, negatives first.\n\n▫️ * Investors:* Groq raised from crossover funds while\n\n[West Coast firms](https://www.thevccorner.com/p/the-vc-database-most-founders-spend?r=1krivi)herded past it. Build a process that survives a room where nobody else wants in, the trait\n\n[top VCs](https://www.thevccorner.com/p/what-top-vcs-look-for-2026-founder-playbook?r=1krivi)actually screen for.\n\n▫️ * Operators:* the GPU-vs-LPU framing is a lesson about\n\n[workload-specific tooling](https://www.the-ai-corner.com/p/saas-defense-playbook-ai-era-survival-guide-2026?r=1krivi). You are likely solving compute-bound and memory-bound problems with one tool. Audit where the mismatch costs you speed, starting with your\n\n[token bill](https://www.the-ai-corner.com/p/llm-token-cost-optimization-playbook-2026?r=1krivi).\n\nThe five principles to steal:\n\n**Speed is a strategy, not a tactic.** The deal moved in three weeks because the engineering was already done.**Pick one dominant game, and make it legible to everyone.** A single number beats a stack of departmental KPIs.**Hire by screening for negatives, grow by demonstrating positives.** Different skills, different processes.**Replace passive questions with intentional statements.** It is the cheapest leadership fix available to you today.**Name your discontent specifically.** Vague ambition fades. A number attached to a real cost does not.\n\nSpeed compounds before anyone notices it is compounding.\n\nThe best deals look inevitable only after they close.\n\nDiscontent, pointed at something specific, is the only fuel that does not run out.\n\n## Keep reading\n\n#### More masterclass breakdowns\n\n▫️ [OpenAI’s $122B masterclass: 10 takeaways from Sarah Friar](https://www.the-ai-corner.com/p/openai-sarah-friar-122b-masterclass-10-takeaways-2026?r=1krivi)\n\n▫️ [Brian Armstrong runs 1,200 AI agents at Coinbase](https://www.the-ai-corner.com/p/brian-armstrong-coinbase-1200-ai-agents-operating-model-2026?r=1krivi)\n\n▫️ [Demis Hassabis says AGI arrives in 2 to 5 years](https://www.the-ai-corner.com/p/demis-hassabis-agi-2-5-years-10-takeaways-2026?r=1krivi)\n\n#### Founders and capital\n\n▫️ [What top VCs look for in 2026](https://www.thevccorner.com/p/what-top-vcs-look-for-2026-founder-playbook?r=1krivi)\n\n▫️ [Inside the Cerebras Series A deck and the $56B IPO](https://www.the-ai-corner.com/p/cerebras-series-a-deck-teardown-ipo-2026?r=1krivi)\n\n#### The AI stack\n\n▫️ [The token-cost optimization playbook](https://www.the-ai-corner.com/p/llm-token-cost-optimization-playbook-2026?r=1krivi)\n\n▫️ [The AI inference engineering playbook](https://www.the-ai-corner.com/p/ai-inference-engineering-playbook-2026?r=1krivi)\n\n▫️ [AI tools and models library](https://www.the-ai-corner.com/t/ai-tools-and-models?r=1krivi)\n\nFull podcast:", "url": "https://wpnews.pro/news/jonathan-ross-turned-a-3-week-phone-call-into-a-20-billion-nvidia-deal-here-is", "canonical_source": "https://www.the-ai-corner.com/p/groq-nvidia-deal-origin-story", "published_at": "2026-07-14 15:39:07+00:00", "updated_at": "2026-07-14 15:53:11.871268+00:00", "lang": "en", "topics": ["ai-chips", "ai-startups", "ai-infrastructure", "ai-products"], "entities": ["Jonathan Ross", "Groq", "NVIDIA", "Jensen Huang", "Google", "LPU", "GPU"], "alternates": {"html": "https://wpnews.pro/news/jonathan-ross-turned-a-3-week-phone-call-into-a-20-billion-nvidia-deal-here-is", "markdown": "https://wpnews.pro/news/jonathan-ross-turned-a-3-week-phone-call-into-a-20-billion-nvidia-deal-here-is.md", "text": "https://wpnews.pro/news/jonathan-ross-turned-a-3-week-phone-call-into-a-20-billion-nvidia-deal-here-is.txt", "jsonld": "https://wpnews.pro/news/jonathan-ross-turned-a-3-week-phone-call-into-a-20-billion-nvidia-deal-here-is.jsonld"}}