NVIDIA and OpenAI Place $20B AI Chip Bets Each – Common Keyword: 'Inference'

Two $20 billion bets in the 2026 AI chip market that are too similar to be coincidental are drawing attention. NVIDIA secured Groq's IP and key talent, while OpenAI locked in Cerebras chips, each for $20 billion. The common keyword is one: Inference.

---

NVIDIA (NVDA) announced an acqui-hire and IP licensing deal worth approximately $20 billion (about 27.6 trillion won) with AI chip startup Groq on December 24, 2025. OpenAI signed a $20 billion Master Relationship Agreement with AI chip startup Cerebras on April 17, as reported by The Information and officially confirmed in Cerebras' S-1 filing with the SEC the same day. The amounts are virtually identical. NVIDIA through IP acquisition, OpenAI through purchase. Different methods, but both addressing the same problem.

---

The Problem is Inference

NVIDIA's GPU dominance in the AI training market reaches 90%. However, the AI cycle is now shifting its center of gravity from training to inference. The problem is that GPUs may not be the optimal solution for inference.

• Groq's LPU (Language Processing Unit): Faster performance than GPUs for low-latency language model inference
• Cerebras' Wafer Scale Engine (WSE): Competitive advantage in inference speed
• Both NVIDIA and OpenAI started from recognizing the reality that 'GPUs alone cannot handle inference demand'

---

NVIDIA's $20 Billion – Acqui-hire, Not Corporate Acquisition

NVIDIA's Groq deal is commonly called an acquisition, but it's technically different. Jensen Huang explicitly stated shortly after announcing the deal, 'We are not acquiring Groq as a company.' The actual structure is non-exclusive IP licensing + key talent acqui-hire. Groq founder and CEO Jonathan Ross joins NVIDIA, while GroqCloud service was excluded from the deal. Groq appointed new CEO Simon Edwards (former CFO) and continues operating as an independent entity.

The $20 billion NVIDIA paid represents approximately a 3x premium over Groq's last valuation of $6.9 billion. Market consensus interprets this as 'a strategic bet to fill the inference domain where their own solutions are lacking.'

'We are not acquiring Groq as a company'

Jensen Huang / NVIDIA CEO

This deal also raised antitrust concerns. In March this year, Senators Elizabeth Warren and Richard Blumenthal sent a letter to NVIDIA questioning whether this deal constitutes a reverse merger structure circumventing Hart-Scott-Rodino Act (HSR) notification requirements and urged the DOJ and FTC to review it. No official enforcement action has been taken to date.

---

OpenAI's $20 Billion – Escaping NVIDIA Dependency

OpenAI's Cerebras contract details were confirmed through the S-1 filing. OpenAI will receive a total of 750MW of inference computing capacity from Cerebras at 250MW annually from 2026-2028, with an option to purchase an additional 1.25GW. The contract reportedly includes OpenAI's equity stake acquisition in Cerebras.

Cerebras officially filed its S-1 with the SEC on April 17, targeting a NASDAQ listing (ticker: CBRS) by mid-May. Financial metrics disclosed for the first time through the S-1 exceeded market expectations.

• 2025 Revenue: $510 million (+76% YoY)
• GAAP Net Income: $87.9 million (turned profitable from prior year loss)
• Series H Round (February 2026) Valuation: $23 billion (about 33 trillion won)
• IPO Target Valuation: Approximately $35 billion (about 51 trillion won), offering size about $2 billion

To dilute the UAE revenue concentration issue (86%) that was the biggest risk in the previous S-1, they also secured additional partnership agreements with AWS. The OpenAI + AWS combination strengthened the supply diversification story. Lead underwriters are Morgan Stanley (lead), Citigroup, Barclays, and UBS.

---

Direction Both Bets Are Pointing

NVIDIA spending $20 billion to embrace Groq's IP and talent, and OpenAI purchasing $20 billion worth of Cerebras chips simultaneously demonstrate that the AI chip ecosystem is being restructured from GPU monopoly to a structure where inference-dedicated chips coexist.

• NVIDIA: Filling inference market weaknesses through acqui-hire and IP licensing – Groq maintains independent entity status
• OpenAI: Reducing single supplier (NVIDIA) dependency risk through supply chain diversification
• Cerebras: Securing position as core supplier to major AI companies as independent chip designer, initiating IPO with S-1 filing

These two $20 billion deals represent the entry fees each paid in the battle for AI inference era supremacy.

---

InteliView Editorial Team | 2026.04.26