The global data center industry is entering one of the largest capital cycles in the history of modern infrastructure, and artificial intelligence is the reason. After years of steady cloud build out, spending is now tilting decisively toward hardware designed for accelerated compute, with trillions of dollars in play and a handful of key suppliers positioned squarely in the flow of that money.
AI is driving an unprecedented data center capital supercycle, reshaping infrastructure economics worldwide
From cloud boom to AI buildout
A decade ago, cloud providers were mostly scaling fleets of general purpose servers to host virtual machines, storage and web applications. The story was about elasticity, multitenancy and shifting workloads out of corporate data centers.
That playbook is being rewritten. Training frontier scale AI models and serving them to hundreds of millions of users demands a very different kind of infrastructure: tightly coupled clusters of GPUs and custom accelerators, ultra low latency networks and power hungry racks that push the limits of cooling. This shift in infrastructure demand is reflected in agreements like the one between Microsoft and Mistral, which aims to enhance AI capabilities in Europe.
Industry researchers now expect annual data center capital spending to reach on the order of 1.6 to 1.7 trillion dollars by 2030, up from only a few hundred billion just a few years ago. Several analyses estimate that cumulative spending on data centers through the end of the decade could approach 7 trillion dollars, with the majority required for AI capable sites.
In parallel, global data center capacity is expected to nearly double from around 100 gigawatts today to roughly 200 gigawatts by 2030. This is not a normal upgrade cycle. It is an AI infrastructure super cycle, and it is reshaping the economics of every company that builds, equips or finances large scale data centers.
The new math of data center capex
The spending mix is shifting fast. Analysts now see hyperscale capital expenditure crossing 600 billion dollars in 2026, with something like three quarters of that directed to AI oriented servers, accelerators and the specialized facilities required to host them, according to industry forecasts.
That aligns with broader projections that accelerated servers could account for roughly two thirds of total data center infrastructure expenditures by 2030.
Zooming out, multiple research groups are converging on the same conclusion, even if their precise numbers differ:
- One major network and data center research firm projects that global data center capex will grow at a mid teens rate and reach about 1.7 trillion dollars annually by 2030, driven primarily by AI.
- McKinsey estimates that total capital outlays on data centers could reach 6.7 trillion dollars by 2030, of which about 5.2 trillion would be required for AI specific infrastructure.
- JLL and others see an infrastructure investment super cycle of up to 3 trillion dollars over roughly the second half of this decade when combining real estate, power and tenant IT fit out.
- Omdia expects annual data center investment could reach about 1.6 trillion dollars by 2030, underscoring that even conservative scenarios imply enormous growth.
- Goldman Sachs and JPMorgan both estimate hyperscale operators will invest between about 5.3 and 5.5 trillion dollars on AI and data centers by 2030, much of it financed in private credit and bond markets.
The numbers differ because each firm draws the boundaries differently. Some include only facility and power infrastructure, others fold in IT hardware, and some attempt to account for sovereign AI buildouts and edge sites.
But the direction of travel is unmistakable. AI is now the primary driver of data center capex, and the industry is planning around sustained double digit growth rather than a short, speculative spike.
Within this macro picture, AI infrastructure spending itself is already scaling. Recent estimates put quarterly AI related outlays near 90 billion dollars in early 2026, with full year investment around 500 billion dollars and a path to more than 1 trillion dollars annually before the decade ends, based on the combined signals from these forecasts.
Why hardware integrators like SANM sit in the middle
In this environment, companies that design, assemble and integrate complex hardware systems have moved from the periphery to the center of the story. Sanmina, often referred to by its ticker SANM, is a good example. The company’s leadership has articulated an ambition to double revenue within roughly two years as AI and data-center programs scale, underscoring how central this buildout has become to its growth story.
Historically, firms like Sanmina were known as electronic manufacturing services providers. They built servers, storage systems and telecom gear to exact specifications for the big original equipment makers. Margins were slim, and they were mostly invisible to end users.
The AI data center buildout changes that calculus for several reasons:
- The systems themselves are more complex. A modern GPU server or custom accelerator platform involves intricate power delivery, high speed signaling, advanced printed circuit boards and tight mechanical tolerances.
- The cost per rack has exploded. Specialized GPUs and accelerators have become the single largest line item in AI data center builds, so any design or integration error is extremely expensive.
- Time to deploy matters. Hyperscalers are racing to clear AI training backlogs, so partners who can take a reference architecture and deliver thousands of fully integrated, tested nodes quickly are strategically important.
As hyperscale and enterprise customers rework their data center architectures to prioritize accelerated compute, high bandwidth storage and dense interconnects, order patterns to integrators like Sanmina are naturally skewing toward GPU accelerated servers, custom accelerator boards and co designed systems optimized for training and inference. This is exactly where the highest value budgets sit, and that is why this category of supplier is drawing so much attention from investors.
Accelerators as the new center of gravity
The center of gravity in data center design has shifted from general purpose CPUs to accelerators built for linear algebra and parallel workloads. That shift shows up clearly in the numbers.
Industry estimates suggest that hyperscale operators deployed more than 3.5 million AI accelerators in 2025, roughly tripling the installed base in a single year according to some forecasts.
Separate projections see the global AI accelerator market approaching the high hundreds of billions of dollars by 2029, as training clusters scale and inference spreads across cloud, colocation and on premise environments.
These projections align with the broader view that by 2030, accelerated servers will represent around two thirds of total data center infrastructure capex. They also help explain why AI specific data center investments are expected to require around 5.2 trillion dollars through the decade.
If most of the value in a data center now sits in its compute fabric, the companies that design and supply that fabric attract a disproportionate share of spending and negotiating leverage.
For integrators and original design manufacturers, this has two implications. On the upside, it supports richer engineering engagements, closer collaboration with chip makers and hyperscalers, and a longer, higher margin revenue stream tied to each AI platform generation.
On the downside, it concentrates risk around the product cycles and supply constraints of a small number of leading accelerator vendors.
Capacity, power and the physical limits of growth
Capital alone does not guarantee capacity. Power, land and cooling are emerging as hard constraints.
Research indicates that global data center capacity is on track to nearly double to about 200 gigawatts by 2030, with AI workloads likely to account for around half of that capacity by that time.
Other analyses go further and suggest the overall need for data center capacity could roughly triple by 2030, with AI constituting roughly 70 percent of the incremental expansion.
At the same time, power demand is rising even faster. Goldman Sachs Research estimates that AI could drive a roughly 165 percent increase in data center power demand by 2030, requiring substantial upgrades to grids and generation capacity in key markets.
JLL and other observers argue that power availability, not demand, is becoming one of the main constraints on data center growth.
These constraints are already changing the specifications that companies like Sanmina receive. Customers are asking for higher power racks, more advanced thermal management and higher density interconnect solutions so that they can pack more accelerators into each physical footprint.
This pushes integrators into areas that once belonged primarily to facility designers and cooling specialists.
The result is a more tightly coupled ecosystem. Data center developers, utilities, chip makers and hardware integrators must coordinate earlier and more deeply than in previous cloud cycles, because thermal envelopes, grid connections and chip roadmaps now intersect directly.
Financing, risk and the possibility of an AI hangover
The scale of the AI data center buildout has already attracted warnings about over investment and the risk of a future correction.
On one side, the capex numbers are striking. Estimates of cumulative data center and AI infrastructure spending by 2030 range from about 5 trillion dollars for AI specific sites to nearly 7.6 trillion dollars when combining power, facilities and computing over just a handful of years.
On another, AI adoption in many enterprise workflows remains early, and some use cases are still searching for durable business models.
Financing is another source of both strength and vulnerability. Analysts expect that hyperscalers will lean heavily on debt markets, with several trillion dollars of AI and data center investment funded by corporate bonds, private credit and infrastructure vehicles.
As long as rates remain manageable and revenue growth from AI services continues, this is sustainable. If either assumption cracks, financing conditions for new projects could tighten quickly.
There are also technological uncertainties. Rapid advances in model efficiency, new chip architectures and more effective inference techniques could reduce the amount of hardware required per unit of AI capability.
That would be good news for energy and climate, but it could compress the revenue opportunity for some parts of the value chain.
None of this invalidates the core thesis that AI will require a much larger and more capable compute infrastructure. It does suggest that investors and operators should treat the more aggressive forecasts as scenarios rather than inevitabilities, and pay close attention to utilization, pricing and regulatory developments over the next few years.
What this means for enterprises and the wider ecosystem
For enterprises that are not themselves building hyperscale data centers, this transformation still matters.
First, it shapes cloud pricing and availability. As hyperscalers spend hundreds of billions of dollars a year on AI infrastructure, they will seek to recoup that capital through usage based pricing, reserved capacity deals and new platform services.
That will influence the cost and design of AI projects in almost every sector.
Second, it restructures supply chains. Hardware integrators, contract manufacturers and component suppliers become strategic partners rather than interchangeable vendors.
Enterprises that depend on specific accelerator platforms may find that their own deployment timelines are indirectly constrained by the lead times and integration capacity of firms like Sanmina.
Third, it raises questions about resilience and sovereignty. Governments are increasingly framing AI infrastructure as a strategic asset and exploring sovereign data center projects, local content requirements and incentives to ensure domestic capacity.
That will create both opportunities and compliance obligations for international hardware and cloud providers.
Finally, it sharpens the sustainability debate. More AI compute means higher energy consumption and greater pressure on grids, even as many operators commit to renewable power and aggressive efficiency targets.
How quickly AI hardware improves on a performance per watt basis, and how effectively data centers integrate new cooling and power technologies, will determine whether this buildout can be reconciled with climate goals.
Key takeaways and what to watch next
Several conclusions emerge from the current research and from the evolution of past cloud cycles.
- AI has become the dominant force in data center design and investment, shifting the focus from general purpose compute to accelerator centric architectures.
- The industry is planning for a long running capex cycle measured in trillions of dollars, not a short speculative spike, although the exact scale remains uncertain.
- Hardware integrators and system design partners are moving into more strategic positions as the complexity and cost of AI systems rise, but they also face greater dependence on a small set of chip makers.
- Physical constraints around power, land and cooling are likely to be as important as capital in determining where and how fast AI data centers can grow.
- Financing, regulation and real world AI adoption will decide whether the most optimistic spending scenarios are realized or moderated.
For companies like SANM, the message is clear. As long as AI workloads keep growing and hyperscalers remain in an arms race for capability, demand for GPU accelerated servers, custom accelerator boards and tightly integrated systems should remain robust.
The challenge will be to navigate component cycles, customer concentration and policy shifts while continuing to invest in the engineering depth that this new era of infrastructure requires.
The AI compute boom has only just begun, and the decisions made over the next few years about where to build, what to deploy and how to power these facilities will shape the trajectory of the digital economy for the next decade and beyond.
Conclusion
Sanmina’s latest data center results underline how quickly artificial intelligence is reshaping the hardware that sits underneath modern computing. Stronger demand for AI servers storage and networking equipment is not just a short term bump for SANM. It is a signal that the company has successfully repositioned itself into the core of a long running build out of high density compute infrastructure that is likely to define the next decade of cloud and enterprise technology.
Why Sanmina’s AI pivot matters right now
Over the past two years hyperscale cloud providers and large internet platforms have moved from cautious AI experiments to full scale build programs. Capital spending related to AI data centers is now measured in the hundreds of billions of dollars annually as firms such as Amazon Microsoft Meta and Alphabet race to add capacity for model training and inference. Industry research suggests that in 2026 alone AI related investments by major hyperscalers fall in a broad range between roughly 630 billion and 680 billion dollars which is comparable to the economic output of mid sized nations.
That surge in spending is colliding with real world constraints. Analysts and operators describe an environment where the limiting factor is no longer just silicon supply but power cooling and site readiness. Demand for AI computing is outstripping available capacity across most large cloud providers with power and thermal budgets becoming central design challenges. In this context companies that can reliably design manufacture and integrate dense racks of accelerators with advanced power and cooling are moving from behind the scenes suppliers to strategic partners. Sanmina is one of those companies.
How Sanmina arrived at this moment
Sanmina historically operated as an electronics manufacturing services provider focused on high reliability systems for communications industrial and automotive customers. That profile began to change sharply in late 2025 when the company acquired the data center infrastructure manufacturing business of ZT Systems from AMD. The deal is described in industry research as transformative because it effectively doubled Sanmina’s scale and shifted it from a component oriented role into a rack scale integrator for AI and cloud infrastructure.
Following the acquisition AMD selected Sanmina as a key manufacturing and integration partner for its Helios rack level AI architectures. This pairing combines AMD’s processor and accelerator design strengths with Sanmina’s ability to engineer assemble and validate complete racks that integrate GPUs custom accelerators high bandwidth memory storage and high speed networking. The company now focuses heavily on AI servers storage systems and networking gear for hyperscalers and large enterprises rather than just traditional communications equipment.
Financial results from the second quarter of fiscal 2026 illustrate how quickly this pivot has changed the business. Sanmina reported that its revenue effectively doubled year over year driven by surging demand for AI and cloud infrastructure and by stronger than expected contributions from ZT Systems. ZT Systems alone delivered approximately 1.88 billion dollars in revenue in the quarter significantly above internal plans according to management. Communications networks together with cloud and AI infrastructure represented about 69 percent of total revenue or 2.77 billion dollars compared with 733 million dollars in the same segment a year earlier. These are not marginal gains. They reflect a structural shift toward AI data center work as the new core of the company.
What the demand signals say about AI infrastructure
The uptick in Sanmina’s data center backlog and shipments is part of a broader pattern. Research firms tracking global data center capacity find that AI workloads are now the primary engine of growth across regions rather than a niche on top of legacy computing. Forecasts from ABI Research and others indicate that active data center capacity dedicated to AI workloads could grow from around 11.5 gigawatts in 2026 to roughly 43.6 gigawatts by 2031 with AI expected to account for more than half of total capacity early in the next decade.
Hyperscaler oriented studies show a similar trajectory. One recent outlook projects that hyperscaler data center capacity could increase more than sixfold by 2035 with power cooling and workload density rising faster than the number of physical sites. This reflects simultaneous build outs for AI training clusters mainstream cloud services storage and enterprise migration workloads rather than AI alone. Analysts at firms such as Dell Oro have documented sharp increases in spending on accelerated servers which pull through demand for GPUs custom accelerators high bandwidth memory solid state drives and high speed network interfaces that tie large clusters together.
Within that environment Sanmina’s role is to convert component level demand into systems that are ready for deployment. Industry research notes that much of the company’s current growth comes from high level assembly of AI servers and complete racks where it integrates large numbers of accelerators specialized cooling and fast interconnects. The company’s backlog in AI infrastructure reflects not only near term orders but also multiyear programs as hyperscalers schedule phased deployments of training and inference capacity. Stronger data center demand therefore confirms that Sanmina’s pivot is aligned with a secular expansion in AI oriented infrastructure rather than a single product cycle.
Strategic implications for Sanmina and its customers
From a strategic standpoint Sanmina is moving from a traditional manufacturing story toward an infrastructure growth narrative anchored in rack level integration and deeper capture of the hardware value chain. As the company builds complete racks instead of just boards or subassemblies it can participate in a larger share of the bill of materials including power distribution cooling subsystems cabling and mechanical structures. That naturally raises its revenue opportunity per deployment and can support higher margins when execution is strong.
For hyperscalers and large enterprises this evolution offers a practical benefit. The modern AI data center is no longer a simple collection of general purpose servers. It is a tightly engineered environment where accelerators interconnects power delivery cooling and management software all need to be co designed. Industry forums describe a shift from traditional multi tenant data halls to power dense campuses optimized for model training and large scale inference with liquid cooling loops fast fabrics and floor plans tuned for accelerator efficiency. A manufacturing partner that can deliver validated racks that meet these constraints reduces integration risk and shortens time to deployment for customers that are already capacity constrained.
Sanmina is also positioned to serve edge AI scenarios where integrated compute and storage systems are deployed closer to users for low latency applications such as industrial automation and real time analytics. While the financial scale of edge deployments is smaller than hyperscale data centers they diversify the company’s portfolio and offer exposure to new classes of AI driven workloads in manufacturing logistics and network operations.
Risks constraints and what to watch
Despite the strong demand backdrop there are meaningful risks and constraints that deserve attention. The first is that AI infrastructure spending is currently concentrated among a small group of very large buyers. Public reports emphasize that a handful of hyperscalers account for the majority of the hundreds of billions in AI related capital expenditures now underway. Dependence on such customers can magnify both upside and downside. A change in the pace of spending or in architectural preferences at a few companies could ripple quickly through Sanmina’s order book.
The second constraint involves power and cooling. Multiple analyses of hyperscaler earnings and data center planning note that the bottleneck has shifted toward infrastructure rather than chips. Available power grid capacity cooling technologies and permitting timelines are emerging as the principal gates on how fast AI clusters can be added. If utilities regulators or local communities slow approvals for new power dense facilities the impact will be felt across the supply chain including rack integrators such as Sanmina. The company’s long term competitiveness will depend in part on how effectively it can design systems that fit within tightening power and thermal envelopes while still delivering the performance customers expect.
There is also the question of technological churn. Accelerated compute remains a rapidly evolving category with frequent changes in GPU architectures interconnect technologies and memory configurations. Industry commentary points out that accelerator cards and systems often have shorter lifecycles than conventional server gear due to the pace of innovation. For a manufacturer this creates both opportunity and operational complexity. Sanmina must maintain flexible engineering and supply chain processes so it can refresh designs quickly while managing component availability and inventory risk.
Finally AI infrastructure carries wider societal and regulatory implications. As more of the world’s power budget is allocated to AI data centers questions arise over environmental impact regional energy security and the balance of capacity between consumer services and foundational research. Forecasts that AI workloads may consume the majority of data center capacity by the early 2030s suggest that this debate will intensify. Companies at the heart of the build out including Sanmina will need to demonstrate not only technical proficiency but also transparency and responsibility in how and where infrastructure is deployed.
Key takeaways and what comes next
Taken together stronger data center demand confirms that Sanmina’s strategic pivot toward AI servers storage and networking hardware is well aligned with a long running shift in global computing infrastructure. The company has moved quickly from a traditional manufacturing profile to a central role in the AI ecosystem by acquiring ZT Systems data center business partnering closely with AMD and scaling its capabilities in rack level integration for hyperscale customers.
The broader backdrop remains supportive. AI workloads are driving most new data center capacity growth and are expected to account for an increasing share of total power and infrastructure investment over the coming decade. At the same time hyperscalers are pushing into new levels of power density and system complexity which creates ongoing demand for specialized integrators that can deliver complete solutions rather than just parts.
For technology leaders and investors the near term signal is clear. As long as AI demand continues to outrun infrastructure supply companies that sit in the critical path of turning chips into deployable racks should see sustained opportunity. For Sanmina the challenge now is less about proving that demand exists and more about executing consistently solving power and cooling constraints and managing reliance on a small set of large customers. How well the company navigates these issues will determine whether today’s surge in AI infrastructure revenue becomes a durable foundation for the next phase of its growth. reddit
