AMD closed its fiscal second quarter of 2026 with the biggest revenue number in company history, and the engine behind it was not gaming or laptops. It was the data center, where AI demand has turned server CPUs and GPU accelerators into the fastest-growing line item on the balance sheet. Hours after AMD’s numbers landed, Arm moved to stake its own claim on that same market, unveiling the AGI CPU, its first internally designed data center chip built specifically for what the company calls agentic AI workloads. Together, the two announcements mark a shift in how the AI infrastructure race is being fought: not just GPU versus GPU, but CPU architecture versus CPU architecture, with x86 and Arm now competing directly for the same racks.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
AMD’s Record Quarter, By the Numbers
According to AMD’s official Q2 FY2026 earnings release, the company posted revenue of $11.54 billion, a company record and an increase of roughly 50% from the same period a year earlier. That alone would be a strong quarter for any chipmaker. What stands out is where the growth actually came from: AMD’s data center segment brought in $6.718 billion — rounded to $6.7 billion in AMD’s own release — up 107% year-over-year, and accounted for 58% of total company revenue, AMD’s largest-ever quarterly data center total. The acceleration looks even sharper set against where the segment stood just two quarters earlier: in its February 2026 report, AMD had put Q4 2025 data center revenue at $5.4 billion, up 39% year-over-year, meaning the segment’s growth rate nearly tripled in the span of two quarters.
AMD chief financial officer Jean Hu framed the results plainly in the earnings release, noting that revenue increased 50% year-over-year to a record $11.5 billion, driven by continued strength in the data center business, which represented 58% of company revenue in the quarter. The same release described data center segment revenue at $6.7 billion, up 107% year-over-year, driven by strong demand for AMD EPYC processors and AMD Instinct GPUs. That is the clearest official confirmation that AMD’s AI-driven data center business has become the primary growth story for the company, not a side bet. It is also a convincing beat of AMD’s own earlier guidance: back in February 2026, AMD had forecast total Q1 2026 revenue of $9.8 billion, plus or minus $300 million, a figure that included roughly $100 million in expected MI308 accelerator sales into China, and the company’s actual trajectory since then has run well ahead of that early-2026 projection.
Reporting on the results, CEO Lisa Su pointed to a fifth consecutive quarter of record server CPU revenue, with cloud and enterprise demand each growing more than 70% during the quarter. That streak matters because it shows the data center run is not a one-quarter spike tied to a single large customer order. It has now stretched across more than a year, dating back to a Q4 2025 data center print of $5.4 billion, up 39% year-over-year, that AMD disclosed in its February 2026 results, and spanning multiple product generations and multiple categories of buyer, from hyperscale cloud operators to traditional enterprise IT departments upgrading server fleets.
What’s Driving the Boom: EPYC Venice and Instinct MI450
Two product lines are doing most of the work inside AMD’s data center number. The first is the sixth-generation EPYC processor, code-named Venice, which is built on AMD’s Zen 6 architecture using a 2-nanometer manufacturing process. AMD has positioned Venice specifically around agentic AI workloads, the multistep, tool-using AI tasks that are increasingly running continuously inside cloud infrastructure rather than as one-off queries. AMD’s own performance claim, cited in the earnings commentary, puts Venice at up to 3.3 times the performance per watt of leading Arm-based CPUs, a specific jab at the architecture Arm is now trying to compete with directly using its own silicon.
The second product line is the Instinct MI450 series of AI accelerators, AMD’s answer to Nvidia’s dominant GPU lineup in AI training and inference. The MI450 series is shipping as part of AMD’s Helios rack-scale system, a fully integrated AI infrastructure platform that bundles CPUs, GPUs, networking, and memory into a single deployable unit rather than requiring customers to assemble components themselves. Microsoft is already deploying Helios racks on Azure, according to reporting on the earnings call, which gives AMD a marquee hyperscaler reference customer at a moment when cloud providers are racing to add AI capacity.
The wafer and memory constraints squeezing the rest of the PC industry, detailed in UBS’s warning that silicon wafer prices could jump 40% by 2027, make AMD’s data center execution look even more notable. The company is scaling a complex, multi-chip rack-scale product during the same window that memory and substrate costs are climbing across the industry, a dynamic also visible in the broader Chinese memory sector detailed in coverage of the CXMT and YMTC memory chip race.
The Anthropic Mega-Deal Behind the Instinct Numbers
One customer commitment stands out in the reporting around AMD’s quarter: Anthropic has committed to as much as 2 gigawatts of AMD Instinct MI450-series GPUs. A 2-gigawatt AI compute commitment from a single AI lab is a substantial figure by any measure in this industry, and it signals that AMD’s accelerator business is no longer dependent solely on a handful of hyperscaler cloud deals. Frontier AI labs building and training their own large models are now treating AMD’s Instinct line as a credible, capacity-scale alternative to Nvidia’s GPUs, at least for a meaningful slice of their infrastructure needs.
The scale of that commitment also explains why AMD’s stock has been treated as an AI infrastructure bellwether alongside Nvidia over the past year, a dynamic covered in depth when AMD’s market capitalization first crossed the $1 trillion mark and in the follow-up analysis of Wall Street’s split reaction to that milestone. Anthropic’s GPU commitment, layered on top of Microsoft’s Helios deployment, gives AMD two of the most closely watched names in AI as public reference customers heading into fiscal 2027.
Arm’s Answer: Introducing the AGI CPU
Arm’s response arrived through its own newsroom, where the company described the AGI CPU as the first offering of Arm’s new data center silicon product line, adding that it is available to order now, according to Arm’s newsroom announcement. That single sentence marks a genuine break from how Arm has operated for decades. Historically, Arm licensed its chip designs and instruction set architecture to other companies, such as AMD’s own Arm-based competitors, and collected royalties on every unit shipped, rather than designing and selling a finished data center CPU under its own brand.
Arm’s own materials frame the AGI CPU around a specific workload profile. In a product introduction, Arm described agentic AI as workloads that run continuously, scale across thousands of parallel tasks, and require sustained performance at gigawatt scale, according to Arm’s product introduction page. That description is effectively identical to the workload category AMD is targeting with EPYC Venice, which confirms that both companies now see the same emerging category, continuously running, multistep AI agents operating at data-center scale, as the next major battleground for CPU architecture.
Arm has also made specific performance claims for the new chip. A company exhibit filed with securities regulators states that the AGI CPU delivers more than 2 times the performance per rack versus x86 CPUs, enabling up to $10 billion in capital expenditure savings per gigawatt of AI data center capacity, according to Arm’s SEC-filed exhibit. That is a direct challenge to the x86 architecture that AMD and Intel have built their data center businesses on for decades, and it puts a specific dollar figure behind Arm’s pitch to hyperscalers who are watching every watt and every dollar of capital spending as they scale AI capacity.
Why Arm Is Moving From Licensing to Silicon
The strategic logic behind Arm’s shift is straightforward: licensing revenue and per-unit royalties, while durable, cap how much value Arm can capture from the total AI infrastructure boom compared with companies that sell finished silicon directly. By designing and selling its own data center CPU, Arm positions itself to capture a share of the value chain closer to what AMD, Nvidia, and Intel already capture on every chip they ship, rather than a smaller royalty slice on designs licensed to those same companies.
It is a bet that carries real execution risk. Arm’s traditional customers, including companies building their own Arm-based server chips, may view Arm’s move into finished silicon as new competition rather than a neutral architecture partner, a tension that echoes how Qualcomm’s own $60 billion AI chip deal with AWS reshaped assumptions about which companies can credibly ship data-center-scale AI silicon. Arm now has to convince hyperscalers that a chip from the architecture licensor itself is worth choosing over both the x86 incumbents and Arm-based designs from its own existing licensees.
Rene Haas and the $2 Billion Demand Signal
Arm CEO Rene Haas has put a specific number behind early demand for the AGI CPU. According to reporting on Arm’s recent results, Haas said demand for the AGI CPU had exceeded $2 billion across fiscal 2027 and fiscal 2028, and that Arm was adding customers, including multiple customers in the United States and China. That figure covers committed or projected demand rather than shipped revenue, since the chip is only just becoming available to order, but it gives investors a concrete benchmark to check Arm’s execution against over the next two fiscal years.
Arm’s overall business is still much smaller than AMD’s. The company reported Q1 FY2027 revenue of $1.29 billion, up 22.4% year-over-year, a fraction of AMD’s $11.54 billion quarter. That gap illustrates the scale of the bet Arm is making: the AGI CPU is not yet a meaningful percentage of Arm’s revenue, but the $2 billion demand pipeline Haas cited would represent a substantial new revenue stream if it converts into actual shipments on the timeline Arm is targeting.
AMD’s 2027 Outlook: Why Data Center Revenue Could Double Again
AMD is not treating its record data center quarter as a ceiling. In commentary reported October 1, 2026 by 247WallSt, AMD management said Data Center segment revenue is expected to “more than double year-over-year in 2027,” a forward target that builds directly on the 107% growth and $6.718 billion the segment already posted in Q2 FY2026. Clearing that bar would mean the segment roughly doubles again from a base that has already more than doubled, a second straight year of triple-digit-equivalent growth.
The case for taking that guidance seriously rests on commitments that are already locked in rather than purely aspirational. Anthropic’s up-to-2-gigawatt commitment to AMD’s Instinct MI450-series GPUs and Microsoft’s deployment of Helios racks on Azure, both detailed above, represent multi-year capacity build-outs that will not fully show up in a single quarter’s revenue. If those deployments ramp through fiscal 2027 the way both companies have indicated, they would supply a meaningful share of the volume AMD needs to approach a doubling target without signing a single additional customer. Add in five consecutive quarters of record server CPU revenue spanning both hyperscale and enterprise buyers, and AMD’s 2027 guidance reads less like a stretch goal and more like an extrapolation of demand the company says it can already see on its books.
The case for skepticism is just as concrete. Hitting “more than double” twice in a row requires AMD to keep securing wafer capacity and high-bandwidth memory at a moment when, as covered above, UBS has warned silicon wafer prices could climb 40% by 2027. Every additional Instinct accelerator and EPYC Venice chip AMD ships in 2027 competes for the same constrained manufacturing capacity that is also getting more expensive industry-wide. Arm’s AGI CPU adds a second variable: if its performance-per-rack and capex-savings pitch persuades even a handful of hyperscalers to shift new server orders toward Arm-based silicon instead of EPYC, that would chip away at the server CPU side of AMD’s data center number at precisely the moment AMD needs it to keep accelerating rather than plateau.
The guidance also sharpens the contrast with Arm’s own numbers. Rene Haas’s demand pipeline of more than $2 billion cumulative across fiscal 2027 and fiscal 2028, discussed above, is for a brand-new product line starting from a standing start. AMD, by comparison, is guiding for a Data Center segment that already generates more than $6.7 billion in a single quarter to more than double again, year-over-year, in 2027. The two figures are not directly comparable, since one is a new-product demand pipeline and the other is guidance layered on an established multibillion-dollar run rate, but the gap underscores how much larger AMD’s existing data center business already is relative to the challenge Arm is mounting against it.
For now, the 2027 target is a management projection rather than a locked-in number, and AMD has not attached a specific dollar figure to it beyond the “more than double” framing. The clearest way to track it will be the same way AMD’s current streak became visible in the first place: one quarterly data center print at a time, measured against the 107% year-over-year bar the company has already cleared once, with October 2026 marking the first time that 2027 target entered the public record.
AMD vs Arm: Comparing the Data Center AI Chip Bets
Laid side by side, AMD and Arm are pursuing the same customer base with fundamentally different starting points. AMD is defending and extending a data center business that already generates billions in quarterly revenue and has a five-quarter record streak behind it. Arm is trying to build a brand-new hardware revenue line from a standing start, leaning on its decades of architecture credibility and a specific efficiency pitch around performance-per-rack and capital expenditure savings.
| Product / Line | Architecture & Process | Headline Claim | Business Model | Status (per company statements) |
|---|---|---|---|---|
| AMD EPYC Venice (6th-gen) | Zen 6, 2nm process | Up to 3.3x performance-per-watt vs. leading Arm-based CPUs | Direct CPU sales to OEMs and hyperscalers | Targeting agentic AI workloads, per AMD earnings commentary |
| AMD Instinct MI450 series | AI accelerator (GPU) | Anchor of Anthropic’s up-to-2GW compute commitment | Direct accelerator sales | Shipping as part of Helios racks |
| AMD Helios rack-scale system | Integrated CPU + GPU + networking rack | Deployed on Microsoft Azure | Systems-level sales | Shipping / ramping |
| Arm AGI CPU | Arm’s own data center CPU design | 2x+ performance-per-rack vs. x86; up to $10B capex savings per GW | Direct silicon sales (new for Arm) | “Available to order now,” per Arm newsroom |
| Arm overall business | IP licensing + new silicon | $1.29B Q1 FY2027 revenue, +22.4% YoY | Licensing royalties plus new hardware revenue | AGI CPU demand cited at over $2B across FY2027–FY2028 |
The comparison also highlights a subtlety in AMD’s own marketing: AMD is benchmarking EPYC Venice against Arm-based CPUs generally, not against Arm’s brand-new AGI CPU specifically, since the AGI CPU had not shipped in volume at the time AMD made that claim. Expect that specific matchup, AMD Zen 6 versus Arm’s own silicon, to become one of the most closely tracked benchmarks in the data center chip world once independent testing becomes available.
Where Nvidia and Intel Fit Into the Picture
Nvidia remains the dominant supplier of AI training and inference GPUs, and neither AMD’s earnings nor Arm’s AGI CPU announcement changes that baseline reality on their own. What both moves do is add pressure at the margins: AMD’s Instinct line and Helios racks give hyperscalers and AI labs a second sourcing option for accelerator capacity, which matters when GPU allocation itself has become a constrained resource, as detailed in coverage of how Nvidia B200 resale premiums have fueled a GPU loan boom. Intel, meanwhile, faces pressure on two fronts simultaneously: AMD’s server CPU share gains, which AMD has highlighted for several consecutive quarters, and now Arm’s direct entry into the same x86-dominated server CPU category Intel has anchored for decades.
The pressure is not limited to established Western chipmakers either. China’s push for AI compute independence, visible in reporting on Huawei’s Ascend 960DT and DeepSeek’s reported order for 160,000 chips, shows that the AI chip race Arm and AMD are fighting in the West has a parallel track running in China, where domestic silicon is being scaled specifically to reduce reliance on both x86 and Arm-based imports. AMD itself has tried to keep a toehold in that market despite export controls: its February 2026 guidance for the first quarter of 2026 called for total revenue of $9.8 billion, plus or minus $300 million, a figure that explicitly included about $100 million in expected MI308 accelerator sales into China, a modest sum next to AMD’s overall data center business but a telling sign of how closely even partial access to the Chinese market is being tracked.
Historical Context: How We Got Here
For most of the past two decades, the data center CPU market was effectively a two-horse race between Intel’s x86 Xeon lineup and AMD’s Opteron and, later, EPYC lines, with Arm-based chips serving mostly mobile devices and a small number of specialized server projects. That began changing as cloud providers like AWS built their own Arm-based server chips to cut costs and power consumption, proving that Arm architecture could handle serious server workloads at scale. AMD, for its part, spent years rebuilding credibility in the data center after a rough stretch in the early 2010s, a turnaround that accelerated sharply once generative AI created enormous new demand for both CPUs to run supporting infrastructure and GPUs to run the models themselves.
Arm licensing its architecture to countless chipmakers, rather than selling its own finished silicon, was the company’s business model for essentially its entire history. The AGI CPU announcement represents one of the more significant strategic pivots in Arm’s recent history, comparable in scale of ambition to Qualcomm and AWS deepening their own custom silicon ambitions. It is worth noting that Arm’s pivot comes at a moment when the broader AI infrastructure buildout is running into its own physical constraints, from the memory pricing pressure described in UBS’s wafer price warning to the desktop chip delays covered in reporting on AMD’s own Zen 6 desktop CPU timeline slipping to 2027, even as the data-center-focused Zen 6 Venice chips move ahead on schedule.
Market Reaction: Stock Moves and Investor Sentiment
Market data snapshots from the days around the announcements show both stocks pulling back slightly rather than rallying, a reminder that even record results can be met with muted reactions when expectations are already running high. Financial data platform Pluang showed Arm trading at $328.34, down 1.46% in a snapshot dated September 23, 2026, while market data provider TradingView showed AMD at $614.61, down 1.47% over the prior 24 hours in a snapshot from the same date. Neither figure represents a confirmed single-day closing move tied directly to the earnings or product news, but they do show that both stocks were trading with some volatility in the days immediately surrounding the reports.
| Metric | Figure | Period / Context | Source |
|---|---|---|---|
| AMD total revenue | $11.54 billion | Q2 FY2026, company record | AMD Q2 FY2026 earnings release |
| AMD revenue growth | +50% YoY | Q2 FY2026 | AMD CFO Jean Hu, earnings release |
| AMD data center revenue | $6.718 billion | Q2 FY2026, +107% YoY | AMD Q2 FY2026 earnings release |
| Data center share of total revenue | 58% | Q2 FY2026 | AMD CFO Jean Hu, earnings release |
| AMD 2027 Data Center guidance | More than double YoY (company target) | FY2027 outlook, cited Oct. 1, 2026 | AMD management, via 247WallSt |
| Arm quarterly revenue | $1.29 billion | Q1 FY2027, +22.4% YoY | Arm earnings reporting |
| Arm AGI CPU demand pipeline | Over $2 billion | Cumulative, FY2027–FY2028 | Arm CEO Rene Haas |
| AMD stock snapshot | $614.61, -1.47% | 24-hour move, Sept. 23, 2026 | TradingView |
| Arm stock snapshot | $328.34, -1.46% | Daily move, Sept. 23, 2026 | Pluang |
What This Means for Cloud Providers and Enterprises
For hyperscale cloud providers, the practical upshot is more negotiating leverage. When AMD, Arm, Nvidia, and, to a lesser extent, Intel are all pitching data-center-scale AI silicon at the same time, buyers like Microsoft, Amazon, and Google gain room to diversify sourcing and pressure pricing on every renewal cycle. Microsoft’s decision to deploy AMD’s Helios racks on Azure alongside its existing Nvidia and custom-silicon commitments is itself an example of that diversification strategy already playing out in production infrastructure.
For enterprises running their own AI infrastructure rather than renting it from a hyperscaler, Arm’s capital expenditure pitch, up to $10 billion in capex savings per gigawatt of capacity, is designed to appeal directly to finance teams evaluating multi-year AI infrastructure build-outs. Whether that claim holds up once the AGI CPU is deployed at scale, outside of Arm’s own marketing materials, is the open question that will determine whether Arm’s pivot into silicon becomes a durable new revenue line or a costly experiment.
Expert and Industry Perspective
AMD’s own numbers speak for themselves, but the company’s leadership has been explicit about what is driving them. As AMD CFO Jean Hu put it in the company’s Q2 FY2026 earnings release, “Revenue increased 50% year-over-year to a record $11.5 billion, driven by continued strength in our Data Center business, which represented 58% of company revenue in the quarter,” a statement that leaves little ambiguity about where AMD’s growth is concentrated (AMD investor relations).
Arm has been equally direct about the ambition behind the AGI CPU. The company’s newsroom stated that “the Arm AGI CPU is the first offering of Arm’s new data center silicon product line and is available to order now” (Arm newsroom), while Arm’s product materials describe the target workload directly: “Agentic AI runs continuously, scales across thousands of parallel tasks, and requires sustained performance at gigawatt scale” (Arm product introduction). In its SEC-filed exhibit accompanying the announcement, Arm also stated that “the Arm AGI CPU delivers more than 2x performance per rack versus x86 CPUs, enabling up to $10B in CAPEX savings per GW of AI data center capacity” (Arm SEC exhibit). Taken together, these statements show both companies talking directly to the same buyer: hyperscalers and AI labs trying to squeeze more useful AI compute out of every watt and every dollar of capital spending.
Five Predictions for AI Data Center Chips Through 2028
- AMD’s data center streak keeps extending toward its own 2027 target. With Anthropic’s up-to-2GW commitment and Microsoft’s Helios deployment both still ramping, and AMD management now guiding for Data Center revenue to more than double again in 2027, the five-quarter record run in server CPU revenue is likely to stretch further before it faces a genuine growth slowdown.
- Independent AGI CPU benchmarks become the story of 2027. Arm’s performance-per-rack and capex-savings claims are currently company-sourced. Expect independent lab and hyperscaler benchmarking of the AGI CPU against both x86 chips and Arm’s own existing licensee designs to dominate coverage once shipments scale.
- More Arm licensees push back or diversify. Companies that license Arm’s architecture to build their own server chips may respond to Arm’s move into finished silicon by accelerating their own custom designs, viewing Arm’s AGI CPU as a competitor rather than a purely neutral IP supplier.
- Capital expenditure efficiency becomes the primary sales pitch. With memory and wafer costs climbing industry-wide, expect every major data center chip vendor, not just Arm, to lead with capex-per-gigawatt and performance-per-watt framing rather than raw clock speed or core count.
- Arm’s $2 billion demand pipeline becomes the key scorecard. Investors and analysts will use Rene Haas’s over-$2-billion FY2027–FY2028 demand figure as the benchmark to judge whether Arm’s silicon pivot is converting into real shipped revenue, checking it against actual reported results each quarter.
Related
- AMD Joins $1T Club, Stock Jumps 10% Past Intel [2026]
- AMD Hits $1T: Wall Street Splits $76 on Next Move [2026]
- Nvidia B200’s 58% Premium Fuels GPU Loan Boom [2026]
- UBS Warns Wafer Prices Could Jump 40% by 2027 [2026]
- AMD Zen 6 Desktop CPUs Delayed to 2027: 12-Core Chiplet Leak [2026]
Frequently Asked Questions
How much revenue did AMD’s data center segment generate in Q2 FY2026?
AMD’s data center segment generated $6.718 billion (rounded to $6.7 billion in AMD’s own release), up 107% year-over-year, according to AMD’s official Q2 FY2026 earnings release.
What is AMD’s outlook for data center revenue in 2027?
AMD management said Data Center segment revenue is expected to more than double year-over-year in 2027, according to commentary reported October 1, 2026, building on the 107% growth to $6.718 billion the segment posted in Q2 FY2026.
What is the Arm AGI CPU?
The Arm AGI CPU is Arm’s first internally designed data center chip, described by Arm’s newsroom as the first offering of the company’s new data center silicon product line, aimed at agentic AI workloads.
Is the Arm AGI CPU available to buy now?
Yes. Arm’s newsroom announcement states the AGI CPU is available to order now, though the company has not disclosed a broad general shipping date for volume deployment.
What is AMD’s EPYC Venice processor?
EPYC Venice is AMD’s sixth-generation EPYC server CPU, built on the Zen 6 architecture using a 2-nanometer manufacturing process, positioned for agentic AI workloads with a claimed performance-per-watt advantage over leading Arm-based CPUs.
How large is Anthropic’s AMD GPU commitment?
Reporting on AMD’s results describes Anthropic committing to as much as 2 gigawatts of AMD Instinct MI450-series GPUs, one of the largest publicly disclosed AI compute commitments tied to AMD’s accelerator business.
How much demand has Arm reported for the AGI CPU?
Arm CEO Rene Haas said demand for the AGI CPU had exceeded $2 billion across fiscal 2027 and fiscal 2028, with customers reported in both the United States and China.
Does Arm’s move into silicon threaten its existing licensing customers?
It could create tension. Companies that license Arm’s architecture to build their own server chips may now view Arm’s finished AGI CPU as direct competition rather than a neutral architecture license, though Arm has not publicly characterized it that way.
How did AMD and Arm stock react to the news?
Market data snapshots from around the announcements showed both stocks trading slightly lower: TradingView showed AMD near $614.61, down about 1.47% over 24 hours, and Pluang showed Arm near $328.34, down about 1.46%, in snapshots dated September 23, 2026.
Related Coverage
- AMD Joins $1T Club, Stock Jumps 10% Past Intel [2026]
- Arm's AGI CPU Takes Aim at AMD's $6.7B Surge [2026]
- RTX 60 Delay to 2028: AMD Radeon Caught in Same Freeze [2026]
- AMD Hits $1T: Wall Street Splits $76 on Next Move [2026]
- Aeon Pro vs Xbox Elite 2 vs DualSense Edge: $60 Gap [2026]
