AMD, Cisco and the Saudi state-backed AI firm HUMAIN flipped the switch on a production AI computing cluster in Saudi Arabia on August 31, 2026, running AMD Instinct MI355X GPUs over a Cisco Silicon One network fabric. The announcement, distributed via GlobeNewswire and picked up by outlets including The National and Network World, marks the first time AMD’s newest Instinct accelerators have gone live at data-center scale in the Gulf, and it lands one week after AMD posted a data center revenue quarter unlike anything in the company’s history.
The timing is not an accident. AMD, Cisco and HUMAIN are trying to prove that Nvidia is not the only supplier capable of building sovereign AI infrastructure at gigawatt scale, and Saudi Arabia is trying to prove it can become a genuine third pole in global AI compute, alongside the United States and China. What went live this week is a small first step toward a stated goal of 1 gigawatt of AI infrastructure in the kingdom by 2030.
AMD, Cisco and HUMAIN switch on Saudi Arabia’s first MI355X AI cluster
According to the joint press release issued by the three companies, the newly live infrastructure is already serving HUMAIN customers “in the Kingdom and beyond.” HUMAIN, described in its own materials as a company of Saudi Arabia’s Public Investment Fund (PIF) delivering full-stack AI solutions, is operating the system as a commercial GPU-as-a-service platform for AI training and inference workloads.
AMD chair and chief executive Dr. Lisa Su called the milestone significant in a statement carried in the release, saying, “Bringing AMD Instinct systems online in Saudi Arabia is an important milestone in our work with HUMAIN.” HUMAIN chief executive Tareq Amin framed the launch around a broader shift in enterprise AI demand, stating, “Demand for sovereign, high-performance AI infrastructure continues to grow.” Cisco chair and chief executive Chuck Robbins tied the deployment to Cisco’s networking ambitions, saying, “The true power of AI lies in its potential to solve some of the world’s most complex challenges.”
Every major hardware vendor now wants a piece of the Gulf AI build-out, and this deployment is AMD and Cisco’s clearest joint answer yet to Nvidia’s dominant position in the region.
Inside the deployment: MI355X GPUs, EPYC CPUs and Cisco Silicon One networking
The production system pairs AMD Instinct MI355X GPU accelerators with AMD EPYC server CPUs, connected through Cisco’s Silicon One-based AI networking fabric. Per the companies’ announcement, the networking layer runs on Cisco’s N9000 Series switching platform using Silicon One silicon and 800G optical connectivity, the kind of high-bandwidth, low-latency interconnect large GPU training and inference clusters require to keep hundreds or thousands of accelerators working in sync.
AMD’s own product materials list the Instinct MI355X’s launch date as June 12, 2025, and cite roughly 5 petaflops of AI compute performance per unit, though AMD has not published a detailed public breakdown of FP8 versus FP4 throughput or exact HBM3E memory capacity for the chip in the way Nvidia does for its Blackwell Ultra line. That gap in public disclosure is itself notable: AMD is leaning on deployment speed and openness rather than a spec sheet arms race to make its case in the Gulf.
HUMAIN’s service model treats the cluster as shared infrastructure rather than a dedicated supercomputer for one customer. Enterprises and government entities inside Saudi Arabia, and reportedly customers in neighboring markets, can rent MI355X compute for model training or inference the same way they would rent GPU instances from a hyperscale cloud provider, except the data stays inside Saudi borders. That data-sovereignty pitch is central to why HUMAIN exists at all.
AMD’s data center business just posted its biggest quarter ever
The Saudi deployment lands roughly four weeks after AMD reported its second-quarter 2026 financial results. Per AMD’s official earnings release, dated August 4, 2026, total company revenue reached $11.5 billion, up 50% year-over-year. The Data Center segment, which includes both EPYC CPUs and Instinct GPUs, generated $6.7 billion of that total, up 107% year-over-year and accounting for 58% of AMD’s overall quarterly revenue. AMD attributed the jump specifically to demand for its Instinct MI350-series GPUs, the family that includes the MI355X now running in Saudi Arabia, alongside continued EPYC server CPU adoption.
That growth rate outpaces AMD’s overall company growth by a wide margin, and it shows the data center segment has become the primary engine of AMD’s business rather than a side project next to its PC and gaming GPU lines. A live, revenue-generating cluster in Saudi Arabia gives AMD a concrete reference deployment to point to when it tells investors and prospective customers that MI350-series chips are winning business outside AMD’s traditional US hyperscaler base.
AMD Instinct MI355X vs Nvidia GB300: how the specs compare
Nvidia remains the company AMD is chasing in data center AI compute, and its Blackwell Ultra GB300 platform is the chip HUMAIN and other Gulf AI buyers are weighing MI355X against. According to Nvidia’s own GB300 NVL72 product materials, the rack-scale system delivers up to 20 terabytes of GPU memory and up to 576 terabytes per second of aggregate memory bandwidth across the full 72-GPU rack, with fifth-generation NVLink providing 130 terabytes per second of total interconnect bandwidth inside the rack. Third-party spec sheets referencing Nvidia’s disclosures put individual GB300 GPUs at roughly 252 to 288 gigabytes of HBM3E memory and 7 to 8 terabytes per second of per-GPU memory bandwidth.
AMD has publicly disclosed far less granular detail about the MI355X than Nvidia has for GB300, which makes a precise flops-for-flops comparison difficult using only public sources. What’s clear from the deployment itself is that HUMAIN considers MI355X clusters, paired with Cisco networking, capable of running production AI training and inference workloads today, while its separately announced GB300 purchases from Nvidia, covered below, represent a parallel and larger track of the same build-out. HUMAIN is not choosing one vendor. It is running both.
| Platform | Vendor | Status in Saudi deployment | Key disclosed spec | Networking partner |
|---|---|---|---|---|
| Instinct MI355X | AMD | Live in production, Aug 31 2026 | ~5 petaflops AI compute (AMD) | Cisco Silicon One, N9000, 800G optics |
| Blackwell Ultra GB300 NVL72 | Nvidia | 18,000 GB300 chips ordered by HUMAIN (Network World) | 20TB GPU memory, 576TB/s bandwidth per rack (Nvidia) | Nvidia NVLink / Spectrum-X (per Nvidia materials) |
| Instinct MI400 series | AMD | Planned for 2027 expansion phase | Not yet publicly detailed by AMD | Cisco Silicon One (planned) |
| AWS Trainium | Amazon | Part of AWS-HUMAIN $5B AI Zone (per reports) | Not detailed in this announcement | AWS networking stack |
Who is HUMAIN? Saudi Arabia’s PIF-backed AI champion
HUMAIN is the artificial intelligence company formed by Saudi Arabia’s Public Investment Fund, the sovereign wealth vehicle that also controls stakes across gaming, sports and technology, including the just-closed $55 billion take-private deal for Electronic Arts. Where PIF’s other bets have focused on acquiring stakes in existing companies, HUMAIN is built to be Saudi Arabia’s own full-stack AI operator, running everything from data centers and networking to foundation models and enterprise AI services under one roof.
The company has moved unusually fast by industry standards. In roughly a year, HUMAIN has signed infrastructure and technology agreements with chipmakers and cloud providers including Nvidia, AMD, Amazon Web Services, Qualcomm and Cisco, according to reporting from Network World. HUMAIN’s stated ambition is not simply to buy compute, but to become a regional AI utility that Gulf governments, banks and enterprises can build on without sending sensitive data outside Saudi borders.
The roadmap: 250 megawatts by 2027, 1 gigawatt by 2030
AMD, Cisco and HUMAIN have laid out a specific expansion timeline alongside this week’s launch. The next phase calls for up to 250 megawatts of additional AI compute capacity, beginning construction in 2027 and expected online in the second half of that year. That expansion phase is planned around AMD’s next-generation Instinct MI400-series GPUs and EPYC CPUs, running on AMD’s open ROCm software stack, rather than another round of MI355X hardware.
Beyond that, the three companies have set a long-term target of up to 1 gigawatt of AI infrastructure capacity in Saudi Arabia by 2030. For context, 1 gigawatt is roughly the output of a large nuclear reactor unit, and it puts the AMD-Cisco-HUMAIN venture in the same capacity conversation as the largest AI data center campuses being built in the United States by hyperscalers such as Microsoft, Meta and Amazon. Reaching that figure would require years of sustained capital spending well beyond what has been publicly committed so far.
Why Saudi Arabia is racing to build sovereign AI infrastructure
Saudi Arabia’s push into AI infrastructure sits inside its broader Vision 2030 economic diversification program, which aims to reduce the kingdom’s dependence on oil revenue by building new technology, entertainment and industrial sectors. AI compute has become one of the most capital-intensive fronts of that strategy, because owning the physical infrastructure, not just licensing software from US AI labs, is what lets a government keep sensitive data and model training inside its own borders.
That sovereignty argument is why HUMAIN is buying chips from multiple competing vendors instead of picking one. A Nvidia-only or AMD-only strategy would leave Saudi Arabia exposed to a single supplier’s pricing, export controls or delivery delays. Running MI355X clusters alongside separately reported GB300 purchases and AWS Trainium capacity gives HUMAIN leverage with each vendor and reduces the risk that any single geopolitical or supply chain shock stalls the entire build-out.
| Partner | Reported deal scope | Source |
|---|---|---|
| AMD | $10 billion joint investment in AI data centers over five years | Network World |
| AWS (Amazon) | $5 billion “AI Zone” featuring AWS infrastructure, Nvidia GB300 and AWS Trainium chips | Network World |
| Nvidia | 18,000 GB300 chips ordered, with additional volume planned | Network World |
| Cisco | Silicon One networking fabric across MI355X and future MI400 deployments | GlobeNewswire joint release |
| Qualcomm | Named partner in HUMAIN’s broader technology ecosystem | Network World |
What this means for AMD’s stock and Wall Street’s AI trade
AMD shares have spent 2026 tracking the broader AI infrastructure spending story, with the company’s data center revenue growth serving as the main evidence that AMD can win meaningful share away from Nvidia in AI accelerators, not just in traditional server CPUs. The Saudi Arabia deployment is a proof point for that thesis: a live, paying commercial customer running AMD’s newest Instinct chips at data-center scale, outside the small circle of US hyperscalers AMD has traditionally depended on for Instinct volume.
No named analyst commentary or specific share-price move tied directly to this announcement had been published as of the article’s writing. What matters more for investors watching AMD’s stock is the pattern: a 107% year-over-year jump in data center revenue last quarter, followed within weeks by a new international reference deployment. That combination is what analysts covering AMD tend to point to when arguing the company’s AI accelerator business has crossed from promising to material.
AMD vs Nvidia: the data center GPU rivalry moves to the Gulf
For years, the AMD-versus-Nvidia data center story played out mostly in US hyperscaler earnings calls and GPU allocation reports. The Saudi Arabia deployment shows that rivalry has fully expanded into sovereign AI infrastructure deals, where governments and state-backed companies, not just Microsoft, Google and Meta, are now major GPU buyers. HUMAIN’s decision to run both AMD Instinct and Nvidia GB300 hardware inside the same national AI strategy suggests Gulf buyers see real value in a multi-vendor approach, rather than betting the kingdom’s AI future on a single chipmaker.
That dynamic is good news for AMD specifically, because it means AMD does not need to displace Nvidia outright to win meaningful business in the region. It only needs to be a credible second source that HUMAIN and similar buyers trust enough to run production workloads on, which is exactly what going live in Saudi Arabia this week was designed to demonstrate.
From MI300X to MI355X: how fast AMD’s Instinct line has moved
AMD’s Instinct accelerator line has moved through several generations in a compressed timeframe, evolving from the MI300 series into the MI325X and now the MI350 series, which includes the MI355X now running in Saudi Arabia. That cadence reflects the same pressure Nvidia faces: AI customers expect a new flagship accelerator roughly every year, not every two to three years as was typical in prior GPU generations. AMD’s already-announced move to the MI400 series for the 2027 Saudi expansion phase confirms that pace is not slowing down.
For HUMAIN and other buyers, that rapid generational turnover cuts both ways. It means today’s MI355X cluster will be technically outdated within roughly two years of its own launch, but it also means HUMAIN can add MI400-based capacity in 2027 without waiting for a multi-year hardware refresh cycle, keeping its infrastructure closer to the frontier than a slower procurement cycle would allow.
Competing in the Gulf: who else is building AI data centers in the Middle East
Saudi Arabia is not alone in the region. The United Arab Emirates has pursued a parallel strategy through its own state-linked AI company, and both countries have separately courted the same pool of US chipmakers and cloud providers, including Nvidia, AMD, AWS and Qualcomm, for infrastructure partnerships. That regional competition is part of what is pushing HUMAIN to move quickly and announce concrete production milestones, like this week’s MI355X launch, rather than only unveiling future-dated pledges.
US chipmakers, for their part, appear willing to sell into both Gulf AI buildouts simultaneously, treating the region as a genuinely new demand pool rather than picking a single national champion to back exclusively. That is consistent with AMD, Nvidia, Cisco and AWS all appearing across HUMAIN’s own partner list rather than any one vendor securing an exclusive regional arrangement.
Five predictions for the next phase of the Saudi AI compute race
Based on the scale and pace of commitments disclosed so far, several outcomes look likely over the next 12 to 24 months. First, AMD will use the live Saudi MI355X deployment as a reference case in sales conversations with other sovereign wealth funds and national AI initiatives outside the Gulf, given how few non-hyperscaler customers have Instinct GPUs running in commercial production today. Second, expect HUMAIN to announce additional vendor partnerships or expanded chip orders before the 2027 MI400-based expansion phase begins, since its current strategy depends on keeping multiple suppliers actively competing for its business.
Third, Cisco’s Silicon One networking business is likely to gain visibility as a distinct competitive front in the AI infrastructure market, separate from the GPU battle between AMD and Nvidia, since large GPU clusters are increasingly bottlenecked by interconnect bandwidth rather than raw compute. Fourth, AMD’s data center segment will remain the company’s primary growth story through at least its next several earnings reports, given the 107% year-over-year growth already logged in the most recent quarter and the pipeline of announced deployments still ramping up. Fifth, the 1-gigawatt target for 2030 will likely see its timeline debated publicly as construction phases begin, since infrastructure projects of this scale routinely face permitting, power supply and chip delivery delays that push announced dates later than originally stated.
The bigger picture: sovereign AI compute is becoming its own market
What happened in Saudi Arabia this week is bigger than one GPU deployment. It is early evidence that sovereign AI infrastructure, compute built and operated inside a country’s own borders under its own regulatory control, is becoming a distinct category of demand that chipmakers, networking vendors and cloud providers are now designing go-to-market strategies around. HUMAIN is the most visible example of that category today, but the same pressures, data residency rules, geopolitical hedging against single-vendor dependence, and national AI ambitions tied to economic diversification, are shaping similar projects elsewhere in the Gulf and beyond.
For AMD, Cisco and Nvidia alike, that means the next phase of AI infrastructure competition will not be decided only in US hyperscaler boardrooms. It will also be decided in national AI strategy documents in Riyadh, Abu Dhabi and other capitals now treating GPU procurement as a matter of both economic policy and geopolitical positioning.
Frequently asked questions
What went live in Saudi Arabia on August 31, 2026?
AMD, Cisco and HUMAIN announced that a production AI computing cluster built on AMD Instinct MI355X GPUs, AMD EPYC CPUs and Cisco Silicon One networking is now live and serving customers, according to their joint press release.
What is HUMAIN?
HUMAIN is an AI company owned by Saudi Arabia’s Public Investment Fund that operates full-stack AI infrastructure and services, from data centers and networking to enterprise AI platforms.
How does AMD’s MI355X compare to Nvidia’s GB300?
Nvidia has published detailed GB300 NVL72 specs, including up to 20 terabytes of GPU memory and 576 terabytes per second of rack-level bandwidth. AMD has disclosed less granular detail for MI355X publicly, citing roughly 5 petaflops of AI compute performance without a full precision-by-precision breakdown.
Is Nvidia also involved in HUMAIN’s infrastructure?
Yes. According to Network World’s reporting, HUMAIN has separately ordered 18,000 Nvidia GB300 chips and is building a $5 billion AI Zone with AWS that includes both Nvidia GB300 and AWS Trainium hardware, alongside its AMD and Cisco deployment.
How big is AMD’s data center business right now?
AMD’s Data Center segment generated $6.7 billion in revenue in the second quarter of 2026, up 107% year-over-year, and made up 58% of AMD’s total $11.5 billion in quarterly revenue, per AMD’s official earnings release.
What is the long-term capacity target for this deployment?
AMD, Cisco and HUMAIN have set a goal of up to 1 gigawatt of AI infrastructure capacity in Saudi Arabia by 2030, with an intermediate expansion of up to 250 megawatts targeted for 2027 using AMD’s next-generation Instinct MI400-series GPUs.
Why does Saudi Arabia want its own AI infrastructure instead of using US cloud providers?
Sovereign AI infrastructure lets Saudi Arabia keep sensitive government, financial and enterprise data inside its own borders under its own regulatory control, while also reducing dependence on any single foreign chip or cloud vendor as part of its Vision 2030 economic diversification strategy.