Groq Nvidia lawsuit puts $20 billion deal under shareholder scrutiny

Groq faces a shareholder lawsuit over its $20 billion Nvidia deal, despite presenting the transaction as a non-exclusive technology license rather than a sale. The Groq Nvidia lawsuit alleges that directors transferred the startup’s core technology and engineering organization while denying some shareholders a fair process.

Former Groq engineers Benjamin Serebrin and Joshua Rubin brought the proposed class action against former CEO Jonathan Ross and Groq’s former directors. The complaint was filed under seal in Delaware’s Court of Chancery on September 29, 2026, then unsealed on October 5.

The plaintiffs say Nvidia paid $17 billion for a technology license available to Groq’s backers. They also allege Nvidia created a separate $3 billion stock pool for employees who moved to the chipmaker.

Those claims remain allegations, and the court has not decided whether the transaction violated Delaware law. Groq calls the lawsuit meritless and says the agreement delivered exceptional value to the company, investors, and employees.

The dispute matters beyond one startup’s capitalization table. License-and-hire transactions have become an important alternative to conventional acquisitions in artificial intelligence. They let a buyer obtain technology and talent without necessarily purchasing corporate equity, customer contracts, or every operating asset.

That distinction now faces a direct legal test. The central conflict is between the deal’s contractual form and its practical result: Groq remained independent, but Nvidia gained the technology and team responsible for its defining product.

The Groq Nvidia lawsuit targets the deal’s unusual structure

The plaintiffs argue that the transaction functioned like a company sale even though Nvidia did not purchase Groq itself.

Groq announced its agreement with Nvidia on December 24, 2025. The company described it as a non-exclusive license covering Groq’s inference technology, which runs trained AI models and produces responses.

The announcement also said Ross, Groq President Sunny Madra, and other team members would join Nvidia. Groq said it would remain independent under a new chief executive and that GroqCloud would continue operating.

Those statements established two simultaneous outcomes. The corporation survived, but its founder, senior leadership, engineers, and strategically important technology moved into Nvidia’s orbit.

The lawsuit challenges how Groq’s directors handled that split. According to the complaint, the board transferred assets essential to Groq’s identity while preserving a separate company controlled by certain continuing investors.

The plaintiffs claim this process favored directors, affiliated investment funds, senior executives, and employees selected to join Nvidia. They allege that other shareholders received an undervalued cash exit without an appropriate vote or an independent effort to test the transaction’s price.

A court case summary quotes the complaint as accusing the board of handing Nvidia Groq’s technology and its builders. The filing characterizes the surviving business as a diminished company after the transfer.

Nvidia is not named as a defendant. The lawsuit instead focuses on the fiduciary duties of Ross and Groq’s former directors, including whether they treated shareholders fairly.

This distinction is important. The court does not need to decide that Nvidia behaved improperly before examining whether Groq’s board approved a conflicted transaction.

Groq rejects that account. A company spokesperson said the transaction delivered exceptional value and promised a vigorous defense.

The original deal announcement supports several facts that neither side appears to dispute. Groq licensed inference technology, key leaders joined Nvidia, and Groq continued as an independent company.

The disagreement concerns what those facts mean under corporate law. Groq sees a license, employee movement, and continued independence. The plaintiffs see a functional sale that separated the company’s value from shareholders who helped finance its development.

That conflict makes the lawsuit more consequential than a routine valuation complaint. It asks whether a board can divide a startup’s technology, workforce, and remaining corporation without applying protections associated with a conventional sale.

Nvidia’s accounting shows where it believed the value resided

Nvidia’s own financial statements assign far more value to Groq’s workforce and future development than to the existing technology asset.

Nvidia reported $17 billion of consideration for the license arrangement. Its fiscal 2026 annual report says $13 billion was paid when the transaction closed, with another $4 billion payable within one year.

The filing recorded a $2.5 billion developed-technology intangible asset. Nvidia valued that asset through a cost-to-recreate method and assigned it a five-year useful life.

It separately recorded $14.4 billion of goodwill, which is an accounting asset representing value that cannot be assigned to separately recognized tangible or intangible assets. Nvidia said this goodwill primarily reflected the workforce and future development of the licensed technology.

That breakdown provides unusually clear evidence about the transaction’s commercial logic. Roughly 85 percent of the recorded license-related assets consisted of goodwill, not the recognized value of the completed technology.

The Nvidia annual filing also states that the company hired certain Groq employees. It says Nvidia did not purchase Groq equity, existing products, or customer contracts.

Those exclusions support Nvidia and Groq’s position that the transaction was not an ordinary acquisition. Yet the goodwill calculation strengthens the plaintiffs’ argument that Nvidia primarily wanted an engineering organization capable of continuing the technology’s development.

The difference between those interpretations will shape the case. Accounting classifications do not determine whether directors complied with their fiduciary duties. They can still illuminate what Nvidia believed it was buying.

The plaintiffs allege Nvidia hired nearly all of Groq’s engineers, possibly as many as 200. That number comes from the complaint and has not been established by a court.

They further allege that employees who transferred became eligible for a separate pool of Nvidia restricted stock. According to the lawsuit, those benefits created incentives that were not shared equally with remaining shareholders.

The reported structure therefore had three economic layers. Nvidia obtained the license, selected employees moved to Nvidia, and Groq continued with a cloud-focused operating business.

Each layer can be legitimate on its own. The legal pressure comes from how the layers interacted and whether interested directors influenced the allocation of value between them.

Nvidia’s filing also acknowledges the transaction’s operational uncertainty. It warns that incorporating Groq technology into Nvidia’s architectures will require significant engineering work.

The company says integration may face delays, fail to produce desired results, or struggle to win customer and ecosystem adoption. Nvidia also warns that it might not recover the associated cost.

That disclosure complicates any simple claim that Nvidia bought a finished, guaranteed advantage. The company paid for a technical direction and a team, but it still carries substantial execution risk.

The lawsuit is therefore not evidence that the licensed technology will succeed. It is a dispute over whether Groq’s directors negotiated and allocated the transaction’s potential value fairly.

The central reversal is that Groq now plans to buy its technology back as infrastructure

Groq survived as an independent company, yet its cloud business reportedly plans to deploy Nvidia systems built from technology Groq originally developed.

The arrangement’s most striking reversal appears after the transfer. Groq shifted toward operating AI inference infrastructure rather than continuing as an independent chip designer.

Nvidia integrated the licensed architecture into Groq 3 LPX, a specialized inference system designed to handle AI decoding. Decoding is the stage when a trained model generates output one token at a time.

Groq’s Language Processing Unit, or LPU, was designed around predictable execution and large amounts of on-chip SRAM. SRAM is fast memory placed close to processing logic, reducing dependence on external memory for certain workloads.

Nvidia has positioned Groq 3 LPX as a decode companion for its Vera Rubin systems. In that role, Nvidia GPUs handle broader AI computing tasks while the LPU accelerates token generation.

According to technical deal reporting, Groq plans to deploy racks containing Nvidia’s Groq 3 LPX hardware. Dell is expected to support those deployments.

That means Groq’s remaining cloud company could become a customer or deployment partner for infrastructure developed from its former chip technology. Nvidia now controls the product roadmap, while Groq retains experience operating LPU systems at scale.

The structure offers Groq a plausible second act. Designing and manufacturing chips requires sustained capital, supply-chain commitments, software investment, and access to customers willing to adopt a new architecture.

A cloud operator can instead purchase systems from established vendors and sell inference capacity as a service. That approach reduces some semiconductor-development risks, although it also removes direct control over hardware design.

Groq has been raising capital for this cloud-focused business. In May, the company was reportedly seeking up to $650 million from existing investors as it moved away from hardware development.

The company later disclosed a $350 million funding round at a $3.5 billion valuation. That figure is especially relevant because the lawsuit alleges departing shareholders were bought out at an unfairly low value.

A fundraising account described the new company as an AI inference cloud business backed by some existing investors. It also reported that earlier investors had already received distributions from the Nvidia agreement.

The later financing does not independently prove that the earlier shareholder payout was unfair. A surviving company can receive a different valuation after its strategy, capitalization, and ownership have changed.

However, the financing gives the plaintiffs a concrete comparison point. They can argue that some investors retained access to Groq’s next phase while other shareholders were cashed out before that value became visible.

Groq’s future dependence on Nvidia equipment adds another layer. The surviving company is not merely pursuing an unrelated opportunity. It reportedly plans to operate systems derived from the technology that drove its earlier valuation.

That reversal sharpens the lawsuit’s primary conflict. The deal preserved Groq in legal form, but it changed the company from an independent chip challenger into a cloud operator within Nvidia’s expanding infrastructure network.

The case pressures AI boards using license-and-hire deals

A ruling against Groq’s former directors would make it harder to treat technology, employees, and corporate ownership as unrelated pieces of one transaction.

Technology companies have increasingly used license-and-hire arrangements when a complete acquisition presents regulatory, integration, or valuation problems. The buyer licenses intellectual property and recruits key employees while the original corporation remains independent.

These transactions can preserve products and customer relationships that the buyer does not want. They can also give founders and engineers a faster path into a larger company.

For regulators, however, the structure creates difficult questions. A transaction can alter competitive conditions even when the buyer acquires no voting shares.

The Groq agreement attracted attention because Nvidia already occupies a dominant position in AI computing. Groq represented a distinct approach focused on inference, one of the fastest-growing portions of AI infrastructure demand.

Nvidia’s decision to license Groq’s architecture and hire its creators reduced the distance between a potential competitor and the market leader. It also gave Nvidia another specialized processor to pair with its GPU systems.

The Department of Justice has reportedly examined the arrangement for potential antitrust concerns. That inquiry is separate from the Delaware shareholder case and has not produced a public finding of wrongdoing.

The two proceedings nevertheless examine related economic facts from different directions. Antitrust authorities focus on competition, while the Delaware court focuses on directors’ obligations to shareholders.

For boards, the shareholder case presents the more immediate governance warning. Calling a transaction a license does not necessarily resolve how directors should evaluate it.

If a buyer obtains the technology, people, and development capacity responsible for most of a company’s value, directors may need a process resembling a company sale. That can include independent negotiations, valuation work, conflict controls, and shareholder approval.

The plaintiffs reportedly rely on Delaware General Corporation Law Section 271, which governs sales of all or substantially all corporate assets. They contend that the Nvidia transaction crossed that threshold in substance.

Yet the complaint acknowledges that no Delaware decision has directly resolved this exact question. That uncertainty gives Groq’s former board a meaningful defense.

The company retained legal ownership, customer relationships, cloud operations, and a continuing business plan. Nvidia’s filing expressly states that it bought no equity, customer contracts, or existing products.

A court could conclude that these retained elements prevent the license from qualifying as a sale of substantially all assets. It might also assess the fiduciary claims without adopting the plaintiffs’ broader legal theory.

The outcome will depend on details beyond the public announcements. The court will likely examine board deliberations, financial advice, conflicts, negotiations, employee incentives, and the value assigned to the surviving business.

That evidence will determine whether directors made a defensible business judgment or designed the transaction to benefit insiders and continuing investors.

The pressure therefore falls on both startup boards and major technology buyers. Future deals may need clearer separation between license consideration, employee compensation, and shareholder payouts.

Boards may also need independent committees when directors or affiliated funds participate differently from other investors. A transaction becomes harder to defend when the decision-makers receive benefits unavailable to the shareholders they represent.

What the allegations still do not establish

The complaint presents a serious conflict theory, but it does not yet prove that Groq’s shareholders received an unfair price or that the transaction was legally a merger.

A proposed class action tells the plaintiffs’ version of events. It is written to survive judicial scrutiny and obtain relief, not to provide a neutral account.

The defendants have not yet presented a complete factual response in public. Groq’s statement denies the claims but does not address every allegation concerning board conflicts, employee incentives, or valuations.

Several unresolved questions will determine the strength of the case. The first is how much of Groq’s operating value remained after the Nvidia agreement.

GroqCloud continued serving customers, and the company retained its contracts and products. It also held expertise in deploying inference infrastructure, even after many chip engineers reportedly departed.

The second question concerns the $3 billion stock pool. The plaintiffs characterize it as a separate benefit for employees who followed the technology to Nvidia.

Employee compensation is not automatically shareholder consideration. Nvidia could argue that restricted stock compensated transferred employees for future service, retention obligations, and the risks of joining a new organization.

The court will need to evaluate whether that stock pool effectively diverted acquisition value from Groq shareholders. Vesting conditions, individual allocations, negotiations, and eligibility rules will all matter.

The third question is taxation. The plaintiffs say Groq’s receipt of license income created tax costs that a conventional stock transaction might have avoided.

That claim requires a full comparison between realistic deal structures. A hypothetical acquisition could have introduced different regulatory risks, liabilities, integration costs, and tax consequences.

The fourth question is whether Groq had a better alternative. The plaintiffs say the board failed to test the market and account for Nvidia-specific synergies.

Defendants may respond that Nvidia offered unusual value for a non-exclusive license while leaving Groq free to operate. If no competing bidder would match those terms, the final consideration may look more favorable.

Nvidia’s integration risk also cuts against treating the $17 billion payment as the value of a completed product. Its annual report warns that commercial deployment might be delayed, unsuccessful, or poorly adopted.

Groq 3 LPX entering production gives Nvidia tangible progress, but it does not eliminate those risks. Customers still must deploy the systems, integrate software, and achieve attractive economics at scale.

Competition also remains active. Google continues to develop Tensor Processing Units, while Amazon and Microsoft operate custom AI accelerators. AMD, Cerebras, SambaNova, and other suppliers pursue different approaches to inference.

Large cloud providers increasingly combine general GPUs with specialized processors. Nvidia must make Groq technology fit that mixed environment while preserving the software advantages that support its existing platform.

An inference market analysis describes Nvidia’s push into specialized inference as a response to growing competition from custom chips. That context explains the deal’s strategic value without proving the plaintiffs’ valuation claims.

The lawsuit should therefore be read as a challenge to process and allocation, not a final judgment on either. Its allegations are specific, but the decisive evidence remains largely outside public view.

Three signals will determine what happens next

The next phase will test the board’s process, Nvidia’s technical execution, and Groq’s ability to build a viable company after licensing its chip architecture.

The first signal is the Delaware court’s treatment of the defendants’ expected dismissal arguments. The court must decide whether the complaint states viable claims under fiduciary-duty principles or Delaware’s asset-sale rules.

A decision allowing the case to proceed would increase pressure on other AI companies using license-and-hire transactions. It would also open a path toward discovery of board records, valuations, and negotiation documents.

A dismissal would weaken the plaintiffs’ theory, especially if the court finds that Groq retained enough of its business to avoid shareholder-approval requirements. The precise reasoning would matter more than the headline result.

The second signal is the progress of any federal antitrust inquiry. A public enforcement action would strengthen the view that license-and-hire deals deserve acquisition-level scrutiny.

Closure without action would not resolve the shareholder claims. It would, however, reduce regulatory uncertainty around Nvidia’s use of the licensed technology.

The third signal is commercial adoption of Groq 3 LPX. Nvidia needs to show that the system can deliver useful decoding performance inside Vera Rubin deployments.

Customer shipments, production scale, software support, and repeat orders will matter more than architectural specifications alone. Strong adoption would validate Nvidia’s decision to place so much accounting value on Groq’s workforce and future development.

Weak adoption would support Nvidia’s own warning that the agreement carries material integration risk. It could also complicate efforts to calculate damages, since the transaction’s ultimate benefit would remain uncertain.

Groq’s progress deserves equal attention. Its cloud business reportedly expects to expand data-center capacity and deploy Nvidia systems based on the licensed LPU architecture.

Customer retention, financing terms, infrastructure utilization, and expansion into new locations will show whether the surviving company has an independent economic future.

If Groq grows into a substantial inference provider, the defense can point to value preserved after the transaction. The plaintiffs can still argue that selected investors captured that value while others were forced out.

If Groq struggles, the company may look closer to the hollowed-out shell described in the complaint. Yet poor later performance would not automatically prove that directors breached their duties when they approved the deal.

For developers and enterprise buyers, the immediate issue is operational continuity. GroqCloud customers need clarity about platform support, hardware supply, model availability, and the transition toward Nvidia-backed infrastructure.

For founders and employees, the case concerns how intellectual property and future employment are valued together. Compensation tied to joining a buyer can create conflicts when the same people influence the transaction.

For investors, the lesson is to examine what remains after any technology license. Corporate survival means little if the people, product roadmap, and strategic assets have moved elsewhere.

The Groq Nvidia lawsuit will not settle every question surrounding AI license-and-hire deals. It can still define which questions boards must answer before approving one.

Watch the Delaware docket, the federal inquiry, and real Groq 3 LPX deployments. Together, those signals will reveal whether this was a fair license, a disguised sale, or a new transaction model requiring stronger safeguards.