Nvidia’s largest developments this year so far include an ongoing software acquisition spree, the emergence of larger ambitions in the PC market, the reveal of Blackwell Ultra products and future AI computing platforms, and its $4 trillion market capitalization milestone.
More than halfway through 2025, Nvidia has already made and been the subject of several major developments—all related to its dominance of the AI computing market.
One of those developments, the rise of efficient AI models like DeepSeek-R1, has brought into question how long Nvidia can enjoy the fast revenue growth it has experienced over the last few years due to high demand driven by generative AI development.
[Related: How Dell, Lenovo And Supermicro Are Adapting To Nvidia’s Fast AI Chip Transitions]
However, the Santa Clara, Calif.-based company has defied concerns that models like DeepSeek-R1 will undercut demand for its GPUs, reflected by Nvidia becoming the first company to hit $4 trillion market capitalization several months later.
At the company’s GTC 2025 event in March, Nvidia CEO Jensen Huang said that reasoning models like DeepSeek-R1 and the agentic AI workloads they power will create significantly greater demand for the company’s products instead of deflating interest.
That’s because of how reasoning models have significantly increased the number of tokens—or words and other kinds of characters—used for queries as well as answers when compared with traditional large language models, according to the Nvidia founder.
“The amount of computation we need at this point as a result of agentic AI, as a result of reasoning, is easily 100 times more than we thought we needed this time last year,” Huang said during his keynote at the event.
What follows are the 10 biggest Nvidia news stories of 2025 so far, ranging from Nvidia’s ongoing software acquisition spree and the emergence of bigger ambitions in the PC market, to the upcoming Blackwell Ultra products and future AI computing platforms.
10. Nvidia Continues Software Acquisition Spree
Nvidia continued its software acquisition spree in 2025 by scooping up three startups in the first half of the year: Gretel, Lepton AI and CentML.
The company acquired Gretel, a synthetic data AI startup, on March 19, in a deal that valued it at over $320 million, according to Wired. The San Diego, Calif.-based startup had raised roughly $65 million from investors and had struck big partnerships with the likes of Amazon Web Services, Google Cloud and Microsoft.
Gretel’s platform leveraged advanced generative models to create artificial data that retains the statistical properties of real-world datasets while ensuring data privacy. It supported various data types—such as structured tabular data, time-series data and unstructured text—which allow customers to share, analyze and develop AI models without exposing sensitive or any proprietary IP, according to the startup.
Nvidia bought Lepton AI, a Chinese startup that rents out GPU capacity to customers from various cloud providers, the next month, The Information reported at the time.
Then in May, the company announced Nvidia DGX Cloud Lepton, which it called “an AI platform with a compute marketplace” that connects AI developers “with tens of thousands of GPUs, available from a global network of cloud providers.”
Most recently, Nvidia acquired CentML, a Canadian startup that developed software to optimize how AI models run on computer chips, The Logic reported in June.
Last year, the AI infrastructure giant acquired six software startups, including AI infrastructure management startup Run:ai and smaller firms such as Deci and Shoreline.io.
9. Nvidia Reveals New Kind Of PC Chip Ahead Of Bigger Plans
Nvidia in January revealed a mini desktop PC powered by a smaller version of its Grace Blackwell Superchip, foreshadowing what could become a larger expansion by the company in the system-on-chip market for personal computers.
Originally given the code name “Project Digits,” the PC was later named DGX Spark at the company’s GTC 2025 event in March, where Nvidia said that it would release the device later this year alongside OEM partners who plan to release their own versions.
Intended for AI developers, the company said DGX Spark will feature its GB10 Grace Blackwell Superchip to deliver up to 1,000 trillion operations per second of AI computation for the fine-tuning and inferencing of reasoning models. The GB10 was designed in partnership with Taiwanese chip designer MediaTek.
The small-form-factor PC will also feature 128 GB of unified coherent system memory, which will support AI models with up to 200 billion parameters.
To support models with up to 405 billion parameters, two DGX Spark systems can be connected with a cable through their Connect-X networking ports.
Asus, Dell Technologies, HP Inc. and Lenovo plan to release their own versions of DGX Spark under different names, such as the Dell Pro Max with GB10.
Multiple reports over the past year have pointed to the possibility of Nvidia planning a solo line of Arm-based chips using its own CPU and GPU designs for Windows PCs. Reports have also suggested that Nvidia could be working with MediaTek on a system-on-chip that incorporates the former’s GPU and the latter’s CPU.
Nvidia has yet to confirm any plans for a system-on-chip for Windows PCs.
8. Nvidia Makes Integrated Optics Push With Networking
Nvidia in March revealed plans to release new Spectrum-X Ethernet and Quantum-X InfiniBand networking switches that use silicon photonics to lower energy consumption and, as a result, enable larger-scale GPU clusters.
At its GTC 2025 event, the company said the Quantum-X Photonics InfiniBand switches will launch later this year while the Spectrum-X Photonics Ethernet switches will arrive next year from “leading infrastructure and system vendors.”
By integrating photonics on the switch silicon, the new switches reduce the number of lasers by four times compared to traditional pluggable switches that use optical transceivers, according to Nvidia. This results in a reduction in energy consumption by 3.5 times and an improvement in signal integrity by 63 times.
Gilad Shainer, senior vice president of networking at Nvidia, said this also results in 10 times greater resilience for the network because of the improved signal integrity as well as the fewer components required.
In addition, getting rid of the need for optical transceivers will result in 30 percent quicker data center buildouts, according to Shainer.
The significant reduction in energy consumption enabled by silicon photonics means that data centers can support three times more GPUs than those relying on traditional pluggable optics at the same power envelope, the executive said.
“We can bring more GPUs under the same power envelope, essentially enabling further scale and increasing compute density,” Shainer added.
7. Competition Rachets Up From Rivals While A Couple Stumble
Nvidia continued to face growing competition this year from various companies, though at least a couple of them stumbled in their efforts to introduce new AI chips.
One of the strongest signs of competition came from AMD, which in June revealed Instinct GPUs and rack-scale AI systems that the company said will make it increasingly competitive against Nvidia.
At its Advancing AI event last month, the chip designer revealed several details for the Instinct MI350 series GPUs and their corresponding rack-scale systems coming later this year as well as the MI400 processors that will power double-wide racks to go against Nvidia’s Vera Rubin platform next year.
During her keynote, AMD CEO Lisa Su (pictured) said seven of the 10 largest AI companies now use Instinct GPUs, including OpenAI, Meta and xAI. She also cited large-scale cloud and internal deployments of Instinct with Microsoft, Oracle and others.
“With the MI350 series, we’re delivering the largest generational performance leap in the history of instinct, and we’re already deep in development of MI400 for 2026 that is really designed from the grounds up as a rack-level solution,” she said.
Su also revealed that OpenAI, a major Nvidia customer, is a “very early design partner” for AMD’s MI450 GPU, which OpenAI CEO Sam Altman call “extremely” exciting.
This came a few months after Google revealed in April its seventh-generation TPU, Ironwood, which it said it designed to improve performance and scalability for inferencing.
Last December, Amazon Web Services launched its Trainium2 AI chip, saying that it typically offers 30 percent to 40 percent better price performance than other GPU-powered instances that were available at the time.
Microsoft, on the other hand, has reportedly been facing challenges with its homegrown AI chip design efforts, according to a June report by The Information.
The report said Microsoft’s Maia 100 accelerator chip wasn’t designed for generative AI but rather image processing, resulting in the company only using it to train staff. Mass production for a next-generation AI chip code-named Braga, on the other hand, was delayed by at least half a year by Microsoft due to feature requests by OpenAI.
In a statement to CRN, a Microsoft spokesperson said, “Microsoft has been very consistent in our approach to multi-generational systems design efforts, spanning silicon, systems, software, and facility integration. We remain committed to developing optimized systems based on Microsoft’s knowledge about our internal and customer workloads, as well as working with our close silicon partners to implement systems that represent the best of what the industry can offer.”
Intel has also faced hurdles in finding momentum for its AI accelerators after the company said last year that it wouldn’t meet its $500 million revenue goal for Gaudi chips.
In late January, the company announced that it had cancelled its next-generation Falcon Shores chip that was due later this year so that it can focus on developing a “system-level solution at rack scale” with a successor chip it’s calling Jaguar Shores.”
Nvidia is facing competition from startups as well, including Axelera AI, d-Matrix, Encharge and Tenstorrent. One of those startups, Untether AI, announced in early June that it was shutting down after AMD acquired its engineering team.
6. New US Export Curbs Cause Multibillion-Dollar Write-Off
Nvidia CEO Jensen Huang in late May decried the Trump administration’s new restriction on the export of its H20 GPUs to China but said he trusts the president’s “vision” and praised the U.S. leader for boosting domestic manufacturing.
Huang made the remarks during Nvidia’s first-quarter earnings call, where the company reported that ongoing demand for its Blackwell computing products helped it grow revenue by nearly 70 percent year over year in the first quarter, offsetting a multibillion-dollar hit that was caused by the White House’s export restrictions against the H20.
“The president has a plan. He has a vision. And I trust him,” Huang said in response to an analyst questioning whether Trump’s desire to have the United States win in the AI infrastructure market will allow Nvidia to ship an alternative to the H20 into China.
As a result of the U.S. export restriction implemented against the H20 last month, Nvidia said it “incurred a $4.5 billion charge in the first quarter of fiscal 2026 associated with H20 excess inventory and purchase obligations as the demand for H20 diminished.” The company added that H20 sales in the first quarter were $4.5 billion and that it was “unable to ship an additional $2.5 billion of H20 revenue” for the period.
The export restriction also resulted in a loss of $8 billion in H20 sales for the second quarter, according to Nvidia.
On the call, Nvidia CFO Colette Kress noted that the company sold the H20 “with the approval of the previous administration.” The product was originally designed for Chinese customers by complying with export restrictions set by President Biden’s administration.
“Although our H20 has been in the market for over a year and does not have a market outside of China, the new export controls on H20 did not provide a grace period to allow us to sell through our inventory,” she said.
Expanding on remarks Collette made, Huang said Nvidia has “limited options” with respect to introducing a new AI chip product that complies with U.S. export rules.
“Obviously, the limits are quite stringent at the moment, and we have nothing to announce today, and when the time comes, we’ll engage the administration and discuss that,” he said.
5. Nvidia Touts Blackwell Ultra GPU For Reasoning Models
Nvidia in March revealed the first details of Blackwell Ultra, saying the follow-up to its fast-selling Blackwell GPU architecture is built for AI reasoning models like DeepSeek R1 while claiming that the GPU can significantly increase the revenue AI providers generate.
Announced at Nvidia’s GTC 2025 event, the Blackwell Ultra GPU, which will power new DGX SuperPods and the new GB300 NVL72 rack-scale platform, increases the maximum HBM3e high-bandwidth memory by 50 percent to 288 GB and boosts 4-bit floating point (FP4) inference performance by just as much.
Blackwell Ultra-based products from technology partners are set to debut in the second half of 2025. These partners include OEMs such as Dell Technologies, Cisco, Hewlett Packard Enterprise, Lenovo and Supermicro as well as cloud service providers like Amazon Web Services, Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure.
The GB300 NVL72 platform consists of 72 Blackwell Ultra GPUs and 36 Grace CPUs, allowing it to achieve 1.1 exaflops of FP4 dense computation, and it comes with 20 TB of high-bandwidth memory as well as 40 TB of fast memory. The platform’s NVLink bandwidth can top out at 130 TBps while networking speeds reach 14.4 TBps.
With Blackwell Ultra, Nvidia will provide two flavors of new DGX SuperPod configurations. The liquid-cooled DGX SuperPod with DGX GB300 systems consists of eight GB300 NVL72 platforms, amounting to 288 Grace CPUs, 576 Blackwell Ultra GPUs and 300 TB of fast memory that can produce 11.4 exaflops of FP4 computation.
The DGX SuperPod with DGX B300 systems, on the other hand, is pitched by Nvidia as a “scalable, air-cooled architecture” that will be offered as a new design for the company’s modular MGX server racks and enterprise data centers.
4. Nvidia Teases 576-GPU Sever Rack In Road Map Update
Nvidia provided a major road map update in March, saying that the company plans to release a rack-scale architecture for AI data centers that will connect 576 next-generation GPUs in the second half of 2027.
As the company announced last year, the company plans to follow up Blackwell Ultra with a brand-new GPU architecture in 2026 called Rubin, which will use HBM4 high-bandwidth memory for the first time. This will coincide with several other new chips, including a follow-up to Nvidia’s Arm-based Grace CPU called Vera.
At its GTC 2025 event, Nvidia CEO Jensen Huang provided more details about Rubin, which he said would be part of the liquid-cooled Vera Rubin NVL144 platform that will debut in the second half of 2026 and connect 144 Rubin GPUs with Vera CPUs, which will sport 88 custom Arm cores, using its new, sixth-generation NVLink chip-to-chip interconnect.
The Vera Rubin NVL144 platform will have the ability to hit 3.6 exaflops of 4-bit floating-point (FP4) inference performance and 1.2 exaflops of 8-bit floating-point (FP8) training performance, which Nvidia said will make it 3.3 times faster than the new GB300 NVL72.
The platform will feature 13 TBps of HBM4 memory bandwidth and 75 TB of fast memory, a 60 percent increase from the GB300 NVL72. The NVLink 6 bandwidth will hit 260 TBps, double that of the GB300 NVL72. The ConnectX-9 SmartNIC will hit 28.8 TBps, also double.
Nvidia plans to follow up the Vera Rubin NVL144 with the liquid-cooled Rubin Ultra NVL576 in the second half of 2027. While it will keep the Vera CPU, the platform will come with a new GPU package called Rubin Ultra that will expand in size to four reticle-sized GPUs, featuring 1 GB of HBM4e memory and 100 petaflops of FP4 performance.
As the name implies, the Rubin Ultra NVL576 will connect 576 Rubin Ultra GPUs with Vera CPUs using a seventh generation of Nvidia’s NVLink. It will be capable of 15 exaflops of FP4 inference performance and 5 exaflops of FP8 training performance, which Nvidia said will make it14 times faster than the GB300 NVL72 platform.
The platform will feature 4.6 PBps of HBM4e memory bandwidth and 375 TB of fast memory, eight times faster than the GB300 NVL72. The NVLink 7 bandwidth will run 12 times faster at 1.5 PBps while the ConnectX-9 SmartNIC will hit 115.2 TBps, eight times greater than the GB300 NVL72.
On top of providing those details, Huang disclosed that Nvidia plans to deliver a next-generation platform with a new GPU called Feynman, which will feature a new high-bandwidth memory format, in 2028. This platform will also feature Vera CPUs, a next-generation NVLink interconnect, an eighth-generation NVSwitch and ConnectX-10 SmartNICs while using 204 TBps Spectrum 7 Ethernet switches.
3. Nvidia Announces Investment In US Manufacturing
Nvidia said in April that it will build entire AI supercomputers in the United States for the first time thanks to investments it’s making with Taiwanese manufacturing partners TSMC, Foxconn and Wistron.
In its announcement, the AI infrastructure giant said that it has “commissioned more than a million square feet of manufacturing space to build and test Nvidia Blackwell chips in Arizona and AI supercomputers in Texas.”
Nvidia announced the move as President Trump continued to adjust the wide-ranging tariffs he unleashed on nearly 60 countries and regions in early April with the goal of fixing perceived trade imbalances with other countries and grow domestic manufacturing.
This new production capability will allow the company to “produce up to a half trillion dollars of AI infrastructure” in the U.S. within the next four years, according to Nvidia.
The company said it is also working with Amkor and SPIL to handle the chip packaging and testing needs of its AI supercomputer products.
The investments Nvidia is making in U.S. manufacturing for its chips and AI supercomputers is “expected to create hundreds of thousands of jobs and drive trillions of dollars in the economic security over the coming decades,” according to the company.
The company said it will use its “advanced AI, robotics and digital twin technologies to design and operate the U.S. facilities run by parts partners.”
2. Huang Counters DeepSeek Concerns With New Growth Narrative
Nvidia CEO Jensen Huang (pictured) used his company’s GTC 2025 event in March to counter the narrative that the rise of efficient reasoning models like DeepSeek-R1 will undercut demand for its GPUs and associated componentry.
“The amount of computation we need at this point as a result of agentic AI, as a result of reasoning, is easily 100 times more than we thought we needed this time last year,” Huang said during his keynote at GTC 2025.
Huang made the assertion after DeepSeek, the Chinese company behind its eponymous R1 model, claimed in late January that it spent significantly less money than Western competitors such as OpenAI and Anthropic to develop its model. This fueled concerns that AI model developers will require fewer GPUs to train and run models.
By the contrary, Huang said, reasoning models like DeepSeek-R1 and the agentic AI workloads they power will create a need for more powerful GPUs in greater quantities. That’s because of how reasoning models have significantly increased the number of tokens used for queries as well as answers when compared to traditional large language models.
To that end, Huang pointed to Nvidia’s upcoming GB300 NVL72 rack-scale platform powered by its new Blackwell Ultra GPU as well as more powerful computing platforms coming out over the next two years as necessary for keeping up with the computational demands of reasoning models.
This point was outlined by one of Huang’s top lieutenants, Ian Buck, in a briefing with journalists the day before his keynote.
“While DeepSeek can be served with upwards of 1 million tokens per dollar, typically they’ll generate up to 10,000 or more tokens to come up with that answer. This new world of reasoning requires new software, new hardware to help accelerate and advance AI,” said Buck, whose title is vice president of hyperscale and high-performance computing.
With data centers running DeepSeek and other kinds of AI models representing what Buck called a $1 trillion opportunity, Nvidia is focusing on how its GPUs, systems and software can help AI application providers make more money, with Buck saying that Blackwell Ultra alone can enable a 50-fold increase in “data center revenue opportunity.”
The 50-fold increase is based on the performance improvement Buck said Nvidia can provide for the 671-billion-parameter DeepSeek-R1 reasoning model with the new GB300 NVL72 rack-scale platform over an HGX H100-based data center at the same power level.
“The combination of total token volume [and] dollar per token expands from Hopper to Blackwell by 50X by providing a higher-value service, which offers a premium experience and a different price point in the market,” he said.
“As we reduce the cost of serving these models, they can serve more with the same infrastructure and increase total volume at the same time,” Buck added.
More than two months later, Nvidia defied concerns about the impact of DeepSeek-R1 by showing that demand for its GPUs remained strong. Its first-quarter revenue grew nearly 70 percent year over year to $44.1 billion despite a multibillion-dollar write-off that was caused by new U.S. export controls on the company’s H20 GPUs being shipped into China.
Moreover, Nvidia said it expected second-quarter revenue to be roughly $45 billion, which would amount to an approximately 2 percent increase sequentially and a 50 percent year-over-year increase. This also took into account the impact of the new U.S. export controls.
1. Nvidia Becomes First Company To Hit $4T Market Cap
Nvidia Wednesday became the first company to hit a $4 trillion market capitalization.
The AI infrastructure giant achieved the milestone after its share price grew 2.4 percent to $164, according to Reuters, making it the world’s most valuable company once again. Its stock price has risen 18 percent since the beginning of the year.
At the time of Nvidia reaching the sky-high market cap Wednesday morning, Microsoft had the second largest at roughly $3.8 trillion and Apple had the third largest at $3.1 trillion.
The Santa Clara, Calif.-based company exceeded $3 trillion in market cap for the first time last June and weeks later surpassed Microsoft to become the world’s most valuable company for the first time. It hit the No. 1 spot at least two more times last year.