AMD's AI GPU Business Surges, Rivaling Its CPU Operations in Less Than a Year

AMD's AI GPU business has barely been going for a year but it's already as big as its entire CPU operation

Server, cloud, client, laptop combined, AMD's CPU sales are now matched by its AI chips. AMD only announced its Instinct MI300, arguably the company's first truly competitive GPU for AI inferencing and training, back in November last year. It's really only been selling in big numbers for the last few months. But already its bringing in about as much money as AMD's entire CPU operations. That includes CPUs for servers, CPUs for cloud computing, for desktop PCs, for laptops, the works. So says AMD's CEO Lisa Su, speaking at the company's most recent earnings call for financial wonks, held yesterday (via Seeking Alpha). Su was responding to a high-ranking bean counter from Swiss bank UBS, who was trying to drill down into exactly how much money AMD has been raking in over recent months specifically from its nascent AI GPU operations. The analyst reckoned AMD probably notched up in the region of $1.5 billion in revenues for September alone and speculated that following months must be even more. To which Su had this to say: "We actually did better in the Data Center GPU business relative to our initial expectations. So, you would imagine that the business was actually greater than $1.5 billion [in September of this year]. I mean, we're actually seeing now our [AI] GPU business really approaching the scale of our CPU business." That's some achievement and no doubt helps explain why AMD's share price is about 2.5x what it was in mid 2022. Of course, everything is relative. While AMD's burgeoning AI GPU sales look stellar in isolation, they pale compared to Nvidia's. In the same investor call, another analyst pointed out that Nvidia would likely achieve an incredible $50 to $60 billion in AI GPU sales in 2025, while AMD might just hit $10 billion at best. Still, however you slice it anything specifically to do with gaming feels like awfully small beer for AMD these days. Revenues for its "Gaming" business, which includes not only gaming graphics cards for PCs but also the custom AMD-engineered APUs inside the Xbox and PlayStation consoles, amounted to just $462 million in the most recent quarter. That represents just 6.8% of AMD's overall revenues. And shrinking. Likewise, AMD's CPUs for desktop and laptop PCs pulled in nearly $2 billion over the same period. It does all make you wonder how motivated AMD can be when it comes to GPUs for PCs when there's so much more money to be made elsewhere. On the other hand, there's no guarantee that the AI GPU market will continue to boom. And graphics chips for PCs are a long-term core competence for AMD. So, here's hoping AMD is fully committed to improving its gaming graphics. We'll find out soon enough when it comes to AMD's next-gen cards, what with Lisa Su confirming that RDNA 4, likely in the form of the Radeon RX 8800 XT, will arrive "early" in 2025.

AMD sees AI revenue soar as Instinct MI300 GPU already rivals CPU business in sales

Serving tech enthusiasts for over 25 years. TechSpot means tech analysis and advice you can trust. What just happened? AMD's Instinct MI300 has quickly established itself as a major player in the AI accelerator market, driving significant revenue growth. While it can't hope to match Nvidia's dominant market position, AMD's progress indicates a promising future in the AI hardware sector. AMD's recently launched Instinct MI300 GPU has quickly become a massive revenue driver for the company, rivaling its entire CPU business in sales. It is a significant milestone for AMD in the competitive AI hardware market, where it has traditionally lagged behind industry leader Nvidia. During the company's latest earnings call, AMD CEO Lisa Su said that the data center GPU business, primarily driven by the Instinct MI300, has exceeded initial expectations. "We're actually seeing now our [AI] GPU business really approaching the scale of our CPU business," she said. This achievement is particularly noteworthy given that AMD's CPU business encompasses a wide range of products for servers, cloud computing, desktop PCs, and laptops. The Instinct MI300, introduced in November 2023, represents AMD's first truly competitive GPU for AI inferencing and training workloads. Despite its relatively recent launch, the MI300 has quickly gained traction in the market. Financial analysts estimate that AMD's AI GPU revenues for September alone were greater than $1.5 billion, with subsequent months likely showing even stronger performance. The MI300 specs are very competitive in the AI accelerator market, offering significant improvements in memory capacity and bandwidth. The MI300X GPU boasts 304 GPU compute units and 192 GB of HBM3 memory, delivering a peak theoretical memory bandwidth of 5.3 TB/s. It achieves peak FP64/FP32 Matrix performance of 163.4 TFLOPS and peak FP8 performance reaching 2,614.9 TFLOPS. The MI300A APU integrates 24 Zen 4 x86 CPU cores alongside 228 GPU compute units. It features 128 GB of Unified HBM3 Memory and matches the MI300X's peak theoretical memory bandwidth of 5.3 TB/s. The MI300A's peak FP64/FP32 Matrix performance stands at 122.6 TFLOPS. The success of the Instinct MI300 has also attracted major cloud providers, such as Microsoft. The Windows maker recently announced the general availability of its ND MI300X VM series, which features eight AMD MI300X Instinct accelerators. Earlier this year, Microsoft Cloud and AI Executive Vice President Scott Guthrie said that AMD's accelerators are currently the most cost-effective GPUs available based on their performance in Azure AI Service. While AMD's growth in the AI GPU market is impressive, the company still trails behind Nvidia in overall market share. Analysts project that Nvidia could achieve AI GPU sales of $50 to $60 billion in 2025, while AMD might reach $10 billion at best. However, AMD CFO Jean Hu noted that the company is working on over 100 customer engagements for the MI300 series, including major tech companies like Microsoft, Meta, and Oracle, as well as a broad set of enterprise customers. The rapid success of the Instinct MI300 in the AI market raises questions about AMD's future focus on consumer graphics cards, as the significantly higher revenues from AI accelerators may influence the company's resource allocation. But AMD has confirmed its commitment to the consumer GPU market, with Su announcing that the next-generation RDNA 4 architecture, likely to debut in the Radeon RX 8800 XT, will arrive early in 2025.

AMD's Dr. Lisa Su predicts AI GPU market will grow to $500 billion by 2028 or 'roughly equivalent to annual sales for the entire semiconductor industry in 2023'

It's increasingly difficult to walk through our front doors, turn on our TVs, or even steal a quick glance at our phones without being bombarded by the promises of a brave new world of AI. That's not just the case for AI tools and end-user benefits, either, because there also seem to be endless proclamations of limitless AI market growth and money-making potential. AMD CEO Dr. Lisa Su has once again thrown her name into the hat of AI market hopefuls -- not without merit, either, I might add. In AMD's Q3 2024 earnings call (via Seeking Alpha), Su said, "In the Data Center alone, we expect the AI accelerator TAM will grow at more than 60% annually to $500 billion in 2028. To put that in context, this is roughly equivalent to annual sales for the entire semiconductor industry in 2023." It's no surprise that companies which stand to benefit from the AI industry doing well are often those who big it up, predicting the most wondrous successes. But, at the same time, it's difficult not to agree with them. The AI industry, after all, has grown massively over the last few years and had a market size of almost $200 billion last year. There's certainly debate over whether the AI boom might be a financial bubble that's destined to burst, but bubble or not, unless and until it does burst, the market's still there. Su's comments need to be understood with a little more context, however. For one, she's talking about the predicted "TAM" for AI GPUs in 2028, which means "total addressable market". This means the absolute maximum market value that the AI GPU market could achieve: an ideal number. The semiconductor market that Su compares this to did rake in about $500 billion in sales in 2023, but these were actual sales, not TAM. So, Su's not really saying that the AI GPU market will be as big in 2028 as the semiconductor market is today, but the numbers do help get across the sheer scale of the predicted AI GPU boom. It's also important to remember the obvious: GPUs are made from silicon, so if the AI GPU market increases, so does the semiconductor market. As for AMD's role in all this, the company's AI GPU business has barely been going for a year and is already as big as its CPU operation, and its share price has more than doubled since mid 2022. In the wider world, GPUs are now all but synonymous with AI, so it's not surprising that AMD's GPU business has exploded, even if Nvidia Blackwell ends up taking the lion's share with AMD's CDNA 4 picking up the proverbial scraps. It's also no surprise that this explosion and AMD's record revenues isn't too impacted by its gaming revenues being down year-on-year. Even with today's biggest consoles donning AMD graphics, gaming's peanuts compared to AI and all the rest. Still, these are our peanuts, as PC gamers, so fingers crossed the next generation of gaming GPUs from both AMD and Nvidia put some spark back into the gaming side of the business. I suppose if Su's AI market predictions are correct, we can at least be hopeful for continual investment into GPU architectures in the form of AI accelerators, and these can trickle down into our gaming GPUs. Hand-me-downs they might be, but powerful ones nonetheless. I'll take that.

AMD is Finally Catching up with NVIDIA Blackwell

Su cautioned about an uncertain future as AMD prepares to ramp up GPU production and optimise for larger deployments -- quite the opposite of NVIDIA's rapid, consistent growth in AI segments. AMD CEO Lisa Su said on October 30 that the company's MI325 advanced GPU accelerator is a game-changing product that will set off against NVIDIA's H200, while its MI350 series will compete against Blackwell. With NVIDIA beating them in the GPU race, Su expects the company to close the performance gap faster than anticipated with their new hardware series. Previously, Andrew Dieckmann, CVP and GM of data centre GPU at AMD, also acknowledged the same. "We are trying to take very representative benchmarks that are realistic. I can tell you that in our customer engagements, especially regarding inference workloads, we have yet to find a single workload that we cannot outperform NVIDIA on," he said, adding that AMD doesn't always outperform NVIDIA. "However, if we optimise for a specific solution, we can beat them", added Dieckmann. Su also subtly pointed out AMD's aggressive stance in the AI GPU market, positioning itself as a challenger to NVIDIA, during the quarterly earnings call. Announcing the third quarter earnings, Su raised its AI guidance revenue to exceed $5 billion, up from the previous quarter's forecast of $4.5 billion in the wake of the AI boom. Betting big on AI developments, Su calls it the beginning of the "AI super cycle", referring to anticipated demand across all computing domains. That said, she acknowledged the challenges of sustaining growth in an intensely competitive market, with hyperscalers selectively deploying AMD products. "This is really the beginning of what I like to call an AI super cycle. We believe AI is going to be in every aspect of computing" said Su. In Q2 FY24, AMD's revenue stood at $6.8 billion, an 18% increase year-on-year, with a gross margin of 50%. The company's operating income remained at $724 million, while the net income was $771 million. Su said, "We delivered strong third quarter financial results with record revenue led by higher sales of EPYC and Instinct data centre products and robust demand for our Ryzen PC processors," AMD CFO Jean Hu also said that it is making strategic investments to position AMD as the end-to-end AI infrastructure leader and drive long-term profitable growth. During the quarter, AMD generated $3.5 billion in revenue from its data centres, up 122% year over year. This growth can be attributed to increased shipments of AMD's Instinct GPU and EPYC CPU. "Our gross margin improvement has been primarily driven by the mix, especially Data Center business continues to be the strong growth driver of our business, accounting for more than 50% of our revenue mix," revealed Hu, emphasizing AMD's gross margin improvement driven by the data centre segment. Lisa Su also emphasised AMD's software advancements with the ROCm stack, aiming to offer an "open alternative" to NVIDIA's closed software ecosystem, appealing to customers wanting flexibility and cost-effective AI compute options. "ROCm now provides AI developers with a truly open software alternative that has been deployed and validated at scale," said Su. Their newly launched Ryzen AI Pro 3000 series focuses on industry-leading performance in the personal computing sector. "Our third-generation AI-enabled processors deliver unprecedented AI processing capabilities with incredible battery life and seamless compatibility for the applications users depend on", said Jack Huynh, senior VP and GM of Computing and Graphics Group at AMD. Lisa Su also expects the data centre AI accelerator's TAM to reach $500 billion by 2028 and is projected to grow by more than 60% annually. She, however, cautioned about an uncertain future as AMD prepares to ramp up production and optimise for larger deployments -- quite the opposite of NVIDIA's rapid, consistent growth in AI segments. "We feel very good about the growth opportunities. I would say that it might be lumpy. In general, these are large customer acquisitions and it's not always predictable exactly which quarters you would expect the significant build-out" said Su, pointing towards AMD's realistic approach. Su hinted that the faster cadence of AMD's AI products (the likes of MI300, MI325, and MI350) was deliberate, helping them close the competitive gap more quickly. This accelerated pace could strain AMD's resources but underscores their commitment to becoming a serious contender. "We have successfully accelerated our product development pace to deliver an annual cadence of new Instinct products. Our next-gen MI350-series silicon is looking very good and is on track to launch in the second half of 2025 with the largest generational increase in AI performance we have ever delivered", said Su. However, NVIDIA is already a step ahead of Blackwell, with their Rubin GPUs set to release in 2026, and AMD is also planning to launch their MI400X series GPU at the same time. In the client segment, their revenue was up by 29% year-on-year at $1.9 billion. AMD's recent partnerships inside the ecosystem include industry leaders like Oracle, which is using its Instinct GPU accelerators to power its OCI computer supercluster to manage AI workloads. AMD extended its support to the development of Meta's Llama 3.2 open-source models and wants to assist developers in building agentic applications and other groundbreaking AI products using its hardware. They've also partnered with Microsoft to enable Copilot + on their AMD-powered AI PCs. Su acknowledged that capturing market share from entrenched competitors in the data centre GPU segment requires time and patience. Building trust with hyperscalers like Microsoft, Meta, and others has been a focus, suggesting that AMD still faces hurdles in gaining broader adoption but remains committed to long-term partnerships. "This is a multi-generational journey. We've always said that. We feel very good about the progress. I think next year is going to be about expanding both customers as well as workload," said Su. AMD closed the acquisition of Silo AI to accelerate the deployment of AI models on AMD's hardware. AMD also recently announced an agreement to acquire ZT Systems, an AI computing infrastructure provider. The acquisition is subject to regulatory clearances and is likely to be completed by the first half of 2025. Moreover, AMD and Intel also announced a collaboration to broaden the x86 ecosystem by developing an advisory group that will include Dell, HP, Lenovo, Meta, Microsoft, Oracle, and Red Hat. "Establishing the x86 Ecosystem Advisory Group will ensure that the x86 architecture continues evolving as the compute platform of choice for developers and customers," said Lisa Su. AMD's projected revenue in Q4 2024 is set to rise towards $7.5 billion, with an approximate gross margin of 54%. Jean Hu also said, "We are on track to deliver record annual revenue for 2024 based on significant growth in our Data Center and Client segments."