What’s next for Nvidia stock in 2026
Vuk Zdinjak
Sun, December 28, 2025 at 2:50 PM EST
11 min read
In this article:
To understand what is in the cards for Nvidia in 2026, we need to go back and take a look at the most important moves the company made in 2025 and estimate how they will develop in 2026.
Nvidia (NVDA) ended the year with a deal that has left many investors and analysts surprised, and even a little bit confused. Contributing to the confusion was the fact that when CNBC broke the news, it reported that Nvidia would acquire Groq for approximately $20 billion, but once the official announcement from Groq came out, it turned out that the deal was a non-exclusive licensing deal and talent grab rather than a company acquisition.
Prediction Market powered by
Here are key questions Bank of America analyst Vivek Arya raised about the deal, in a research note shared with TheStreet:
-
What does the “non-exclusive licensing agreement” referred to by Groq imply?
-
Could Nvidia have developed this technology on its own?
-
Can Groq cloud, still an independent company, undercut Nvidia’s LPU-basedservice with lower pricing?
Despite having these questions and calling the deal surprising, Arya also said the deal is strategic and complementary. He reiterated a buy rating and the price target for Nvidia stock of $275.
To understand the Groq deal, we need to delve into what Groq technology is about and what the dominant strategies in the tech industry have morphed into.
What is Groq?
Groq’s main business is GroqCloud, an artificial intelligence inference platform. AI inference is the process of generating a response from the AI model that has already been trained.
Groq offers developers a way to run AI models on the company’s hardware and obtain responses very quickly for a competitive price. The reason a relatively small startup can compete with big players and offer competitive pricing for AI inference is its hardware.
The company’s inference platform uses application-specific integrated circuit (ASIC) chips, which it calls the Language Processing Unit (LPU), developed and optimized specifically for LLM inference.
GPUs can be used for many different calculations including gaming, 3D rendering, crypto mining, AI training, and AI inference, but Groq’s LPU chips have only one purpose — AI inference.
This means they have razor-sharp focus, and that makes them many times as efficient at that particular task.
What is Nvidia getting from the deal with Groq?
When Gemini 3 launched, Google touted that it had been trained 100% on its Tensor Processing Units (TPUs), and of course, it is doing inference on TPUs, too. You may have guessed correctly that TPUs are also ASIC chips.
Following the news about Gemini Nvidia’s post on X (formerly Twitter):
"We’re delighted by Google’s success — they’ve made great advances in AI and we continue to supply to Google. NVIDIA is a generation ahead of the industry — it’s the only platform that runs every AI model and does it everywhere computing is done. NVIDIA offers greater performance, versatility, and fungibility than ASICs, which are designed for specific AI frameworks or functions."
Related: Bank of America updates Palantir stock forecast after private meeting
The fact that Nvidia felt it needed to address Gemini in its post showed that the company is worried about the competitive power of well-designed ASIC chips, and now we have proof.
Groq’s announcement about the deal with Nvidia says: “As part of this agreement, Jonathan Ross, Groq’s founder, Sunny Madra, Groq’s president, and other members of the Groq team will join Nvidia to help advance and scale the licensed technology.”
Can you guess what Jonathan Ross’s job was at Google? He was one of the designers of Google’s first generation of TPUs, of course. Nvidia’s decision to license Groq’s LPU tech stack and to “acqui-hire” its talent team is a quiet admission that ASIC chips represent the future of AI.
What does the non-exclusive licensing part of the deal with Groq mean for Nvidia?
A non-exclusive licensing deal was the only way to avoid government scrutiny. The approach here is a combination of Apple and Meta strategies. Apple manufactures custom ARM chips and has a non-exclusive licensing agreement with ARM.
But what makes Apple chips great is the talent that only Apple can attract; so far, competitor ARM chips have been unable to catch up.
Nvidia has secured top talent in this transaction by mimicking Meta’s move, which was an investment in Scale AI. The whole deal with Scale AI turned out to be a lot more about getting Alexandr Wang to lead Meta’s Superintelligence unit than about investing in Scale AI.
This is the new dominant strategy in the tech space where talent is more valuable than whole companies.
Assuming that Nvidia's contract with Groq doesn’t have some special quirks, non-exclusive licensing should mean that other companies can license the LPU designs and build similar LPUs. Nvidia is content, as it isn’t getting the talent, and it's betting they won’t be building anything impressive with just the license.
The second of Arya’s questions, whether Nvidia could have built LPUs on its own, seems like a superfluous one. Even if the company could have developed such chips (assuming no patent issues), it could not have done that in a desirable timeframe.
This underscores my thesis that Nvidia started to worry about TPUs a little late.
Wider implications of the Groq-Nvidia licensing deal
To answer the third of Arya’s questions, we need to first determine Nvidia’s game plan for LPUs. In an email to employees that was obtained by CNBC, Nvidia CEO Jensen Huang wrote the following.
“We plan to integrate Groq’s low-latency processors into the NVIDIA AI factory architecture, extending the platform to serve an even broader range of AI inference and real-time workloads.”
Huang has promoted the idea of AI factories for some time, and he appears increasingly focused on it. This new LPU plan has finally made the whole thing click for me, and the sudden shift to inference is extremely interesting and revealing.
After all the hype of chasing AGI or Superintelligence, the market is shifting toward inference. You’d think that as long as we haven’t reached that amazing life-changing technology, the training capabilities would be of paramount importance.
More Nvidia:
The problem is that LLMs have peaked, and although the shift to inference and “AI factories” is Huang’s stealthy pivot, LPUs are just one part of the puzzle. Nvidia recently announced the Nvidia Nemotron 3 family of open models, data, and libraries. These models are the key component of the latest pivot, which is AI factories and sovereign AI.
Data ownership, privacy, and model fine-tuning are some of the reasons any company or organization that can afford to have a sovereign AI would want it. This is why open-source, and at least open-weight, models are the future, just like ASIC chips.
We can see a slow, ongoing shift toward this, as hundreds of academic papers presented at NeurIPS, the premier AI conference, used Qwen, as reported by Wired.
“A lot of scientists are using Qwen because it’s the best open-weight model,” Andy Konwinski, cofounder of the Laude Institute, a nonprofit established to advocate for open U.S. models, told Wired.
Huang’s plan seems to be a complete sovereign AI solution that offers the fastest inference for the lowest power consumption offered by LPUs, combined with GPUs for training, and Nemotron as a starter software platform.
Arya also wrote this in his note: “We envision future NVDA platforms where GPU and LPU co-exist in a rack, connected seamlessly with NVDA’s NVLInk networking fabric.”
I will adamantly say that this idea is incorrect.
Related: Bank of America resets Micron stock price target, rating
LPUs have a completely different memory model, based on so-called SRAM memory, which is very expensive and very fast. According to Groq, its LPUs connect directly via a plesiosynchronous protocol, aligning hundreds of chips to act as a single core.
Groq calls its chip-to-chip interconnect technology RealScale. LPUs have one other key difference compared to GPUs: they are deterministic. These architectural differences mean that LPU and GPU chips can’t work together to run the same software (perform inference), and placing them in the same racks would only cause problems and complicate matters.
Every LPU has very little memory; a vast number of LPUs are needed to run big LLM models. This will be the deciding factor on how many racks of LPUs will be needed to run the model.
It is certainly possible for Nvidia to develop its LPUs to be a lot different compared to Groq’s, to allow for the mixing with GPUs in the same racks, but in that case, they would take a lot more time to develop. I believe that for Huang’s plan for AI factories, speed of development takes priority.
In any case, a launch of Nvidia’s LPUs in 2026 is highly unlikely, considering that chip design takes at least a year. The Groq deal and inference pivot tell us we need to watch what happens to OpenAI very closely.
Nvidia deal with OpenAI remains in question
On December 19, Reuters reported that SoftBank Group is racing to close a $22.5 billion funding commitment to OpenAI. Considering that SoftBank’s pledge was to invest that money by the end of the year, they are cutting it pretty close.
Waiting until the last moment to follow through makes the company look unsure if it is a good investment.
Nvidia’s agreement with OpenAI to invest up to $100 billion in the startup is still not finalized, according to a Reuters report from December 2. OpenAI doesn’t expect to be cash flow positive until 2030, according to Forbes.
It's easy to see why Nvidia isn’t rushing to finalize the deal with OpenAI. The best-case scenario for OpenAI is that Nvidia is waiting for it to have an IPO first, while the worst-case scenario, of course, is that there is no deal.
OpenAI failing to secure more investments will have a domino effect that will hurt Oracle, Nvidia, and Microsoft the most. Nvidia’s AI factories strategy is a good way for the company to protect itself from dependence on OpenAI as a customer.
What can we expect from Nvidia's partnership with Intel?
According to the leaks, Intel Serpent Lake is the first chip that will feature an integrated Nvidia GPU, which won’t launch before 2027. Even that is optimistic, and 2028 is more likely, as reported by PC GAMER.
Nvidia’s revenue mix and estimates for 2026
The latest Bank of America research note that includes estimates for Nvidia is from November. Arya and his team estimate that Nvidia's revenue for fiscal year 2026 will be $212.83 billion, and non-GAAP EPS will be $4.66. Nvidia missed consensus estimates in Q3 for its revenue from gaming by 4%. There have been rumors that Nvidia is looking to cut gaming GPU production by up to 40% in 2026 due to VRAM supply issues, as reported by PC GAMER.
We can expect that due to the memory industry going all-in on AI, skyrocketing RAM prices will have a side effect of fewer gaming PCs being sold and built, so the gaming revenue could easily miss the consensus again.
In the automotive segment, a similar situation is evident, as Nvidia missed consensus estimates for Q3 by 6%. The company's guidance for Q4 is significantly lower than consensus estimates of $700 million, at $592 million.
The company’s guidance for the pro visualization segment for Q4 is optimistic at $760 million and higher than the consensus $643 million. OEM, including the crypto segment outlook for Q4, is close to consensus at $174 million compared to $172 million.
Non data-center revenue segments look tiny compared to Nvidia's outlook of $51.2 billion and consensus estimates of $57 billion for Q4. As the company focuses more on its highest-margin products, revenue from non-data center segments will continue to shrink.
The Vera Rubin line launch will be the defining moment of 2026, because if the chips bring the promised performance and efficiency uplift, it will destroy any doubts regarding Nvidia’s supremacy.
The next year will be the year of Nvidia, especially if the rumor that Google failed to secure HBM shipments for its TPUs, as reported by Android Headlines, proves true.
I wouldn’t be surprised if this rumor is true, as Huang is always several steps ahead of the competition, except for the moment when he underestimated Google TPUs.
Related: Veteran analyst has blunt message on Intel stock
This story was originally published by TheStreet on Dec 28, 2025, where it first appeared in the Investing section. Add TheStreet as a Preferred Source by clicking here.
View Comments