Everything About AI
NextFin News - Apple is developing an enterprise artificial-intelligence server that could be configured with two or four of its planned M8 Ultra chips and may use Nvidia's NVLink Fusion networking technology to link them, according to a report published Wednesday — a move that would mark the iPhone maker's first return to dedicated server hardware since it discontinued its Xserve line in 2011.
The planned server, aimed at AI developers, businesses and governments, is not expected to reach the market before 2029 and could still be canceled or launched without any Nvidia technology, the report said, citing people familiar with the matter. Apple and Nvidia did not immediately respond to requests for comment.
The news lands in a server market that has grown into the most valuable real estate in technology while Apple sat on the sidelines. Worldwide server revenue reached $122.6 billion in the first quarter of 2026 alone, rising more than 30% year over year as GPU-rich AI systems and hyperscaler investment drove spending, according to International Data Corporation. Grand View Research estimates the global server market expanded from $205 billion in 2021 to $342.1 billion in 2025 and projects it to approach $1.03 trillion by 2033, a 14.8% compound annual growth rate. ABI Research puts the AI-server segment specifically at $245 billion in 2025, on track for $524 billion by 2030. Apple is not chasing a niche; it is eyeing a market that has roughly tripled since the company walked away from it.
What the Report Says: A Server Built on Apple Silicon, Possibly Wired by Nvidia
The core of the plan is an inference server powered by Apple's own M8 Ultra processors — the highest-performance variant of the chip family the company is developing for the AI era. Two configurations are under consideration, one with two M8 Ultra chips and another with four, the report said. To bind those chips together inside the rack, Apple has discussed using NVLink Fusion, Nvidia's high-speed chip-to-chip interconnect that lets processors behave as a single large accelerator rather than a cluster of discrete units.
The combination is striking because it pairs Apple's most closely guarded competency — in-house silicon design — with technology from a company it has regarded as an adversary for nearly two decades. Relations between the two soured after an alleged defect in Nvidia graphics chips that caused widespread MacBook failures, prompting Apple to sever the tie. That the two are now in talks at all signals how much the economics of AI infrastructure have rewritten old alliances.
Apple last sold dedicated server hardware under the Xserve brand, a rack-mounted machine introduced in 2002 and discontinued in 2011 after the company shifted its enterprise focus toward standard Mac hardware and consumer devices. When a dissatisfied customer emailed about the cancellation, Steve Jobs replied that "hardly anyone was buying them." The server market Apple walked away from has since become the most valuable real estate in technology: global cloud-infrastructure spending rose 29% year over year in the final quarter of 2025, according to analyst firm Omdia, as hyperscalers poured capital into AI compute capacity.
The timing is also revealing. Earlier this year, Apple tried running Google's Gemini models on its internal servers, but its Mac-based chips could not handle the large model, forcing the company to run parts of the revamped Siri assistant on Nvidia chips hosted in Google's cloud infrastructure. That episode laid bare the gap between what Apple's device-class silicon can do and what frontier AI workloads demand — a gap a purpose-built M8 server rack is designed to close.
"Some of those users will be enterprise employees," said Tom Mainelli, group vice president of devices and consumer research at IDC. "As developers build Apple Intelligence capabilities into their enterprise apps, we'll see its impact on the commercial side grow."
Why Apple Would Build a Server — and Why Nvidia Would Answer the Phone
The logic runs in two directions. For Apple, the server is a defensive play wrapped in an offensive one. On defense, it reduces dependence on third-party cloud providers for the AI features it bundles into iPhones, iPads and Macs. Every inference request Apple routes through someone else's data center is a margin leak and a privacy exposure waiting to be scrutinized. On offense, it gives Apple a foothold in the enterprise AI infrastructure market, where customers already running Apple hardware could see the rack as a natural extension of the devices on their employees' desks.
For companies with Apple hardware in their installed base, "that could be a significant cost savings across the company," Mainelli said. Apple's privacy and security focus, he added, could give enterprise IT decision-makers a "differentiating reason to consider Apple hardware for AI workloads."
For Nvidia, the calculus is about protecting the interconnect layer. If Apple builds AI racks with its own chips but wires them with NVLink Fusion, Nvidia still captures value even as its GPU share of Apple's stack potentially shrinks. That is a more resilient position than being disintermediated entirely — the fate Nvidia faces if Apple, like several hyperscalers, migrates wholesale to in-house silicon.
The arrangement also fits a broader pattern in the chip industry: hyperscalers and large device makers are increasingly buying from multiple silicon suppliers rather than relying on one. AWS continues to augment its own Trainium accelerators with Nvidia GPUs; the dual-sourcing strategy lets customers improve allocation and lower dependency risk on any single chipmaker. Apple's reported talks with Nvidia sit alongside its work with Broadcom on AI server chips. Broadcom disclosed last week that the two companies had expanded their long-standing technology collaboration through 2031, with Apple saying the agreement is expected to exceed $30 billion and lead to the production of more than 15 billion U.S.-made chips.
The Hard Part: 2029 Is a Long Way Off, and the Plan Can Still Die
The most important detail in the report may be the timeline. A server not expected before 2029 is not a product; it is a statement of direction. Three years in the AI infrastructure cycle is an eternity. Nvidia's current Blackwell generation will have been succeeded by Rubin-class systems; memory and networking standards will have shifted; and Apple's own M8 roadmap could slip, as its in-house Baltra server chip has already reportedly faced delays.
The report also notes the project could be canceled outright or launched without Nvidia technology. That caveat matters because it means the Nvidia angle is the least certain part of the story. Apple has form here: it explores architectures internally, keeps the promising ones, and quietly shelves the rest. The company has also been exploring acquisitions of server-chip companies to strengthen its AI server processors, suggesting the M8 server is one of several parallel paths rather than a single committed bet.
There is also the question of whether Apple wants to be a server vendor at all. Selling rack hardware to enterprises requires a sales, support and service machinery Apple has not maintained since 2011 — a very different business from shipping consumer devices through retail and carrier channels. The more plausible end state may be a server Apple deploys for its own AI services and offers to a narrow set of enterprise and government customers, rather than a broad commercial push into data-center hardware.
And the competitive bar is high. IDC's first-quarter data placed Dell first among named server original-equipment manufacturers with a 16.5% worldwide revenue share, while Hewlett-Packard Enterprise ranked among the five largest vendors at 3%. Apple would be entering a market where incumbents have spent a decade refining rack-level thermal design, power delivery, and enterprise support contracts — capabilities that matter more to buyers than the silicon brand on the spec sheet.
Second-Order Read: The Real Story Is Not the Rack, It Is the Interconnect
The first-order reading of this news is straightforward: Apple wants back into servers. The second-order implication is subtler and more important. What Apple is really testing is whether it can own the compute layer of AI while renting the networking layer — and whether Nvidia will let it.
Nvidia's durable advantage has never been just the GPU die; it has been the system-level moat built from CUDA, NVLink and full-rack engineering. By discussing NVLink Fusion for an Apple-built server, Nvidia is effectively saying it would rather license the moat than lose the customer. That is a meaningful adaptation for a company whose worst-case scenario is a world where every large buyer designs its own accelerator and leaves Nvidia with commodity exposure.
For the broader market, the signal is that the AI infrastructure stack is fragmenting along fault lines that were invisible two years ago. Chip designers, cloud providers, device makers and networking specialists are re-sorting themselves into overlapping coalitions rather than clean vendor-customer hierarchies. Apple-with-Nvidia-networking is not an anomaly; it is an early example of the new shape.
The Counter-Thesis: This Is a Distraction, Not a Strategy
The strongest case against reading too much into this report is simple: Apple has been here before, and it left. The Xserve was discontinued because the market was small and Apple's margins were better elsewhere. A 2029 server target is so far out that it may say more about internal exploration than about a committed product line. If the M8 server never ships, or ships only for Apple's internal use, then the Nvidia talks amount to due diligence rather than a strategic pivot.
There is also a capability argument. Building a competitive AI server is not just about the silicon; it is about thermal design, rack-level power delivery, networking software, and the enterprise support stack that keeps those racks running for customers who cannot tolerate downtime. Apple has not maintained that muscle in more than a decade, and rebuilding it would be a capital-intensive detour from the consumer businesses that generate the company's cash. Rack density is rising faster than Apple's server experience: GPU-dense racks now exceed 80 kilowatts, a thermal load that air cooling cannot manage economically and that liquid-cooling-ready designs are only beginning to address.
The falsifying signal is concrete: if Apple does not file the relevant server-hardware certifications or announce a server product by late 2027 — roughly two years before the reported 2029 window — the project should be treated as shelved rather than merely delayed. A second, sharper signal would be an Apple announcement that it has licensed or acquired an alternative interconnect technology, which would remove Nvidia from the story entirely.
Market Reaction: Modest Moves in a Story That Is Mostly Long-Duration
Apple shares, which traded lower earlier in the session, reversed course and rose modestly in premarket trading after the report surfaced. The muted reaction is appropriate: the news carries little near-term revenue content. A product that may not ship before 2029 does not move this year's earnings, and Apple's stock has already been rewarded for its AI spending plans — the shares are up 21% year to date and 57% over the past 12 months, outpacing the S&P 500's 21% twelve-month gain and trailing the Nasdaq-tracking QQQ's 28%.
The bigger market question is what the story implies for the contest between Apple and Nvidia for the title of the world's most valuable company. The two have traded the crown back and forth in 2026; in July, Apple briefly reclaimed the top spot with a market value near $4.88 trillion while Nvidia hovered around $4.84 trillion. If Apple's in-house silicon program succeeds in reducing its dependence on Nvidia's GPUs while still licensing Nvidia's interconnect, the long-term distribution of value between the two shifts in Apple's favor — but that is a multi-year thesis, not a trading signal.
What to Watch: Three Horizons
Short term (sentiment and positioning): Expect the stock reaction to remain contained. The news has no earnings impact before 2029, and Apple's next catalysts are the iPhone cycle and services growth, not server hardware.
Medium term (fundamentals, 2027–2028): Watch for the Baltra server chip's progress, any server-chip acquisitions Apple makes, and whether Broadcom's expanded collaboration produces AI server components ahead of the M8 rack. These are the leading indicators that the server program is real rather than exploratory. Dell and HPE's quarterly server-revenue trends will also show whether enterprise AI spending is broadening beyond the hyperscalers — the customer segment Apple would need to win.
Long term (structural, 2029 and beyond): If the M8 server ships with NVLink Fusion, Apple will have carved out a hybrid model — own silicon, Nvidia plumbing — that could become a template for other large device makers. If it ships without Nvidia, Apple will have replicated the hyperscaler playbook of full in-house integration. Either outcome narrows Nvidia's addressable market at the margin while leaving its interconnect franchise intact.
Base case: the server arrives late or in a limited form, aimed at Apple's own AI services and select enterprise customers. Upside case: Apple announces the rack ahead of schedule and pairs it with an enterprise software story that reframes the company as an AI infrastructure vendor. Downside case: the project is canceled, the M8 roadmap slips, and the Nvidia talks are revealed to have been one of several discarded architectures.
The takeaway: Apple's reported server plan is less a bet on hardware than a bet on leverage — the kind of move that matters most if it never ships, because the threat of it changes how Nvidia prices its moat.
Data as of September 16, 2026, reflecting premarket trading and market research available at the time of publication.
Explore more exclusive insights at nextfin.ai.
