According to a report by The Information, Apple could re-enter the server hardware market. The servers would be targeted at the AI development, with availability starting in 2029.
While Apple’s plans are tentative, the servers have reportedly been in the works for a year. The company has two types of servers centered around M8 Ultra chips, one with two M8 Ultra chips, while another has four M8 Ultras. The Information also reports that Apple could use Nvidia’s NVLInk Fusion technology, which is used for high-speed connections between chips. Nvidia in 2025 opened up NVLink to other companies for use in their servers.
Mac hardware has become very popular with AI developers. Online processing is a vital part of AI, so Apple’s desire to get into the AI server market is a good opportunity for the company. Apple used to sell Xserve, a line of servers, but it was discontinued 15 years ago. When the company discontinued Xserve, it recommended that customers use a Mac Pro or Mac mini, which is a viable solution for a single, stand-alone server or a very small cluster. With AI, however, large server clusters are needed, and companies want the blade form factor that is the standard in the industry.
The M8 Ultra is a few generations beyond the current fastest chip Apple offers, the M5 Ultra. The M5 Ultra is found only in the Mac Studio and requires a two-pound cooling system. Apple will have to create a cooling system that can fit in a 1U or 2U rack-mount design.
Bloomberg’s Mark Gurman, who has previously reported on Apple plans to release an AI server, responded to the report by stating that Apple could be using the M7 Ultra in its server for 2029, but it’s also possible Apple “is doing M8 Ultra simultaneously.”
But it is not just the rules concerning foldables the company has now defined; it has also set the standard competitors must follow in chip design with its A20 Pro, a super-powered processor that delivers noticeable performance improvements even compared to last year’s iPhone 17.
Memory bandwidth is 50% higher than last year’s chip.
GPU graphics performance increases up to 40% over the A19 Pro.
AI performance is expected to double.
These are all significant improvements that make for noticeable experience boosts for anyone using one of these phones. Prosaically, I anticipate first reviews will celebrate how fast and responsive these devices are, and this will be most evidenced in tasks such as video capture and photography. Pictures will capture more swiftly, open faster, and edit features will become (and already are) more advanced and sophisticated than before. Apple spent ample time celebrating the iPhone 18 Pro series for the new frontiers in digital photography they open up, and not just because of the new reflexive lens.
The point is that the devices have the horsepower to get through computationally intensive tasks, including AI, thanks to the Neural Engines and GPU.
Thermal management is the icing on the quake
What those tasks have in common is heat, which Apple’s product designers have addressed in two critical ways: liquid cooling and processor design.
The processor design is the important thing, because in this case Apple has moved away from the InFO POP packaging of the A19 Pro in which the DRAM is stacked above the chip. It’s now embraced a side-by-side design it described as “inspired by” M-series Apple Silicon.
That works because placing both components alongside each other, rather than in a perpetual embrace, reduces the overall temperature of both by putting a little space between them. As Apple explained it, this Wafer-level Multi-Chip Module (WMCM) approach means the memory gets out of the way of the processor, allowing the latter to make direct contact with the vapor chamber cooling system.
In layman’s terms, it means both memory and chip can perform more intensively while being more efficiently cooled, a move that helps Apple’s iPhone meet those astonishing performance achievements. The vapor chamber is also bigger, made of new material with three times the surface area of before, making it even better at cooling the device.
It also means Apple has achieved a lot of things competitors have not, most particularly in packaging, the use of 2nm chips in a mass market product, and overall design.
That’s where it gets interesting. Both Qualcomm and MediaTek now plan to follow in Apple’s footsteps with the adoption of 2nm chips and side-by-side packaging.
Digitimes suggests it is a move others will follow, as vendors work furiously to ensure their hardware can run AI effectively on device, which iPhones already do. Apple, of course, already knows its Macs, iPhones, and iPads have been designed from the ground-up to support on-device AI as it works to provide the world’s best mass market platforms for secure and private use of the tech.
Friends don’t let friends bet against Apple
All the same, even at the microprocessor level, it’s hard to ignore that where Apple goes, others follow — even while where others go, Apple frequently follows later, but seems to do it better. It has never been a company to bet against, after all. People really should stop doing that.
Please subscribe to my daily, human-curated Apple-related news headline feed at The Core, or follow me on BlueSky, LinkedIn, or Mastodon.
The memory crisis is getting worse for Apple, which is struggling to get enough memory chips together for its upcoming iPhone 18 Pro series smartphones. That’s according to tech journalist Tim Culpan.
As he details it, Apple and its assembly partners are still attempting to secure sufficient quantities of memory, and though they’re confident they can meet initial demand once the devices are introduced, they apparently remain concerned that wait times could rapidly extend as retail inventory evaporates.
Apple’s manufacturing process compounds the problem. The A20 processor is packaged with memory using a new TSMC process, meaning processors are reportedly piling up while manufacturers wait for memory chips. You should read Culpan’s report to get the full picture.
What’s the frequency?
You don’t need to read between the lines to see the challenge. Despite demand for TSMC’s new processors, Apple seems to have been able to secure the supply it needs. But when it comes to churning out the final packaged chips memory, supply constraints have created significant obstacles to producing in quantity.
This is bad for Apple, particularly as the challenge doesn’t appear to be confined to iPhones; customers are experiencing delays getting new Macs. “Many new orders are now not arriving until September,” Bloomberg’s Mark Gurman wrote earlier this week.
It isn’t just Apple that will be impacted by the AI-driven memory drought. The scale of Apple’s orders is among the greatest in the industry, and if its product plans are feeling the pain, every other manufacturer will be feeling it as well.
Conscious uncoupling
This is certainly in tune with expectations voiced at the beginning of the year when Ranjit Atwal, senior director analyst at Gartner, warned: “This is the steepest contraction in device shipments witnessed in over a decade. Higher prices will narrow the range of devices available, prompting buyers to hold on to devices for longer, fundamentally altering upgrade cycles.”
Gartner in February predicted a 10.4% decline in global PC shipments and an 8.4% drop in smartphone shipments as a result. The analyst also predicted a 130% surge in combined memory and SSD storage prices by the end of this year, with steep product price increases to follow. Recent data from IDC, Gartner, Counterpoint, and Omdia confirm PC market declines, but only at around 4% (the estimates vary).
Today’s report from Culpan suggests we’ve not yet experienced the full extent of this decline — hinting that while the initial fall reflected price, the next impact will be defined by lack of supply. While this hurts big brands like Apple, smaller entities could be left high and dry.
When the chips are down
It is interesting to reprise Atwal’s warning in February that, “the sub-$500 entry-level PC segment will disappear by 2028,” as this seems to be what’s happening. That’s something long-time Mac users like me find particularly ironic, given it was only this year Apple briefly offered up its superbly priced $499 MacBook Neo. The industry direction we’re seeing now suggests we’ll never see that again, though the success of that device gave Apple a phenomenal 28.7% increase in sales in its just-revealed June quarter.
Despite memory supply challenges, Apple seems to be faring fairly well, with market share increasing across its business. Counterpoint data reveals that Apple has achieved an astonishing 65% share of the premium smartphone market. In part, that’s because as an existing premium brand, Apple was able to better absorb rising memory costs through higher margins and reduced promotions. Realistically, this means we can expect an overall increase in iPhone prices when the new range is announced up to $300 more, Jeff Pu, of GF Securities, recently claimed.
Building the moat
Once again, what’s sustainable but difficult for larger brands such as Apple is existential disaster for smaller players — and it’s only now a matter of time before we see some real blood. That’s particularly true in smart home and device markets, where manufacturers lack the margins to sustain higher memory prices while delivering products customers can afford. A recent Global Electronics Association report tells us 62% of electronics manufacturers are already experiencing constrained availability or extended lead times. It also tells us 82% expect rising prices, including 33% who cite a “significant increase.”
This is already being felt by consumer and business users, as networking equipment is experiencing significant shipping delays. So, while we may find ourselves waiting a month or more for an iPhone, the wait for new routers, external storage devices, and home automation systems could be even longer once available inventories disappear.
This doesn’t appear to be a short-term challenge; a recent Digitimes report warns that vendors have already sold their entire allocation of memory capacity for 2027.
Don’t even get me started on the likely impact on the military and defense markets as high-performance memory, storage, and processor supplies become constrained. Just like declining river levels in Europe, lack of memory threatens severe disruption. With so much turbulence impacting the tech economy, all we need now is for one or more of theinvestor-supported AI companies that have helped create the memory shortages to default on loan payments.
Got to keep the customer satisfied
Eager to protect sales, Apple recently introduced the Apple Upgrade leasing service in the US. This should enable consumers to purchase new devices at prices more sustainable to them over time. Apple isn’t alone in taking such action, which will inevitably extend beyond America.
“OEMs are expanding financing, trade-in and buyback programs to improve affordability,” said Counterpoint’s Harshit Rastogi. “Samsung has also expanded its Galaxy Forever program to several markets to make flagship devices more accessible.”
Such schemes are all well and good, of course. Consumers will embrace them in hopes of a better tomorrow. But the tech industry is not immune to the wider constellation of existential challenges impacting economic environments.
In the end, if Apple, the industry’s biggest buyer of advanced components, is struggling to secure memory, the rest of the electronics sector is likely to face even greater challenges. For consumers, the initial impacts will be longer waits and higher prices. But the longer term consequences could be slower innovation and increased consolidation across the industry.
AMD executives took to the stage at its Advancing AI 2026 event in San Francisco today to detail the company’s next generation of AI infrastructure solutions, from Instinct MI455X AI accelerator GPUs and 6th Gen EPYC “Venice” CPUs, to Pensando networking, ROCm.AI software and its Helios rack-scale platform that ties it all together.
AMD has been working towards rack-scale AI system solutions for years. Its ZT Systems acquisition last year added valuable engineering talent and intellectual property that is now finally bearing the real fruits. Its Helios AI platform is a major platform evolution for AMD, with shipments scheduled to begin in the second half of this year (which is here and now).
The announcements at Advancing AI show how the company has engineered its AI platform solutions for large reasoning models, sustained inference and agentic workflows. These workloads pressure memory capacity, data movement, networking and CPU orchestration. AMD’s approach is to keep as much data close to the compute engines as possible and move it more efficiently throughout the system, but there’s deeper nuance here that’s obvious versus AMD’s chief rival, NVIDIA.
AMD’s MI455X targets the AI memory wall
The Instinct MI455X GPU is the compute engine that fuels the Helios rack, and the first GPU based on AMD’s new CDNA 5 architecture. Built with a modular mix of 2nm and 3nm chiplets, it carries 432GB of HBM4 and 23.3TB/s of peak memory bandwidth.
Compared to AMD’s current MI355X, the MI455X offers 1.5 times the memory capacity, up to 2.9 times the peak memory bandwidth and up to four times the peak matrix performance with MXFP4 and MXFP8 data types, which are lower-precision numerical formats designed to accelerate AI processing while reducing memory demands. With MXFP6 (6-bit floating point), performance is rated at up to twice that of MI355X.
AMD also shared some actual, measured internal results using production silicon. The company claims MI455X delivers 3.8 times higher FP8 decode performance, 3.5 times more measured FP4 compute performance and between 2.5 and 3.5 times more networking bandwidth than MI355X, depending on the transfer path tested. Those figures provide more context than just numerical specifications, though they remain AMD-provided comparisons that will need independent validation.
AMD
The architectural choices behind the numbers are important. Reasoning models and long context windows require sizeable KV caches for maintaining AI attention states, while mixture-of-experts models frequently move large amounts of data across accelerators. MI455X should let more model data, activation states and cache remain local. New dedicated IP in hardware can transfer data while the GPU continues processing, and expanded cache and multicast capabilities are designed to reduce redundant data movement to further improve efficiency.
The aforementioned lower-precision formats can also raise throughput and reduce memory use, but model developers still have to determine where they can be applied without unacceptable accuracy loss.
AMD’s Helios rack takes aim at Vera Rubin
Dave Altavilla
Helios is AMD’s primary rack-scale competitor to NVIDIA’s Vera Rubin platform. Each liquid-cooled rack combines 72 MI455X GPUs, 18 single-socket Venice host CPUs and Pensando networking technologies.
In its most complete, premium configuration, AMD rates Helios for 2.9 exaflops of low-precision AI compute, with 31TB of aggregate HBM4 capacity, 1.7PB/s of memory bandwidth, 260TB/s of bidirectional scale-up bandwidth and 43TB/s of scale-out bandwidth.
These are formidable figures, but they are technical specifications rather than actual application benchmarks. The more consequential development is AMD’s move from collections of eight-GPU servers to a 72-GPU shared-memory domain. Models too large for one node can operate across the rack without treating every exchange as a scale-out networking transaction, which benefits large-model inference as well as training.
AMD uses UALink over Ethernet, or UALoE, for an open standard scale-up fabric. Each MI455X provides 3.6TB/s of bidirectional scale-up bandwidth, while the complete rack delivers all-to-all connectivity through a single switch layer. AMD also claims six times more scale-out bandwidth per GPU than MI355X when MI455X is configured with three Pensando Vulcano 800 AI NICs.
While open standards give cloud providers more control over suppliers and system design, AMD and its partners now have to prove those components can deliver the predictable performance, reliability and deployment experience customers expect from a tightly controlled, more vertically integrated platform.
Finally, AMD designed Helios with automatic rerouting around failed links, virtual rack partitions, tray-level serviceability and rack-wide power, cooling and health monitoring. Major hyperscalers and potentially large-scale enterprise customers will likely key in on these capabilities, which can affect the availability, total cost and consistency of the AI services they consume.
Kind of like cowbell, AMD Venice gives agentic AI more CPU
AMD
AMD’s agentic CPU messaging regarding its upcoming Venice-based EPYC processors is mostly marketing speak, but the underlying requirement is very real. An AI agent can invoke retrieval, databases, security checks, code execution and other tools before a GPU generates a response. Running many agents concurrently increases the amount of conventional compute requirements surrounding the accelerators.
Venice scales to 256 Zen 6 cores with support for 512 threads, 16 memory channels, up to 1GB of L3 cache per socket, along with PCIe 6.0 and CXL 3.1 connectivity. AMD is also offering several Venice configurations for other applications, including general-purpose servers, high-frequency workloads, GPU hosts and high-density CPU sandbox systems used to execute agent tools.
Treating the CPU solely as a GPU host understates its role. Gateways, tokenization, vector search, databases and short-lived code execution stress different mixes of per-core performance, thread count, memory bandwidth and I/O. Specifically, AMD’s internal testing shows Venice significantly outperforming its current EPYC 9965 Turin CPU across five parts of the agentic AI pipeline, including gateway processing, context assembly, vector search, enterprise applications and short-lived tool execution. Individual gains vary by workload, but AMD details the overall generational improvement at up to a 1.7 times lift. As with the MI455X figures though, these comparisons come from AMD and will require independent validation.
Pensando networking and ROCm software advance
Keeping GPUs fed with data and coordinating traffic across racks directly affects utilization and operating costs. In fact, GPU utilization is a pretty sad state of affairs currently for some of the major frontier model providers.
As such, Pensando networking has become central to AMD’s roadmap. Helios can connect each MI455X to as many as three 800Gbps Vulcano AI NICs, while Salina DPUs handle front-end networking and infrastructure services.
On the software side, which is an equally critical component, AMD also introduced ROCm.AI, an AI-assisted development layer due to arrive in August. It includes reusable skills for coding agents, simplified management and Hyperloom, which can profile workloads, tune serving configurations, modify kernels and validate results.
These tools address two persistent AMD challenges: developer efficiency and ease of use, and software tuning. Automated optimization still has to produce repeatable gains without creating hard-to-maintain code, however. And while ROCm has progressed significantly over the last few years, NVIDIA’s CUDA retains an advantage in maturity, tooling and developer familiarity.
AMD now has commitments that give its MI450 generation and Helios considerably more weight. Meta and OpenAI have announced multi-generation agreements composed of up to 6GW of AMD compute capacity, with initial 1GW deployments planned for the second half of 2026.
Oracle plans a 50,000-GPU public cloud cluster beginning in the third quarter, while Microsoft will deploy Helios for Azure AI inference. Finally, just before the AMD event, Anthropic announced a strategic partnership for up to 2 Gigawatts of AMD-fueled AI compute, with its first gigawatt expected online in the first half of 2027.
Commitments of this scale reflect confidence in more than just MI455X performance. These customers are evaluating the complete architecture, including Venice CPUs, Pensando networking, ROCm software, rack integration, serviceability and AMD’s ability to deliver and execute across multiple product generations.
There is some financial alignment behind the agreements as well. AMD issued OpenAI performance-based warrants and committed to investing up to $5 billion in Anthropic. That context matters when evaluating these deals as market validation, but these planned deployments are substantial nonetheless and put Helios on a much stronger foundation as it begins shipping.
AMD expands its robotics and embedded foundation
AMD also expanded its physical AI portfolio, building on credible traction from its Xilinx-derived Kria adaptive system-on-modules and embedded technologies that are already powering robotics, machine vision and industrial automation applications.
The new Ryzen AI Embedded X100 combines up to 16 Zen 5 CPU cores, integrated Radeon graphics, a second-generation NPU and as much as 128GB of unified LPDDR5X memory shared across its compute engines. To me this looks a lot like a repackaging and optimization of the company’s Strix Halo platform, but with specific optimizations for the embedded space. Regardless, AMD is pairing X100 with the Kria AI Robotics Developer Platform, which includes a System Module or SOM, and a new Robotics Partner Network spanning hardware, software and platform providers.
Samples began shipping in June, with full production expected in the fourth quarter. This broader objective is to give developers a path across AMD x86 CPUs, GPUs, NPUs and FPGAs for real-time autonomous systems, rather than requiring them to assemble those hardware engines and software components independently.
Execution for AMD is now the test
AMD has assembled a credible platform for the burgeoning agentic AI market that’s blowing up currently with no signs of stopping. MI455X addresses memory and data movement, Venice handles dense agentic CPU workloads, Pensando networking connects global system resources, and ROCm.AI addresses software complexity. Finally, Helios assembles these components into a true competitive threat for NVIDIA’s latest Vera Rubin platform.
AMD’s open architecture may appeal to customers seeking supplier choice, but openness must also translate into reliable deployments, competitive total cost and software that does not require a significant rip-up. NVIDIA enters this cycle with a stronger ecosystem and far more rack-scale deployment experience. The true test will be how easily and reliably customers can integrate, operate and maintain these AMD solutions at scale.
As it stands, AMD now has major customers and a clearly defined architecture with systems engineering expertise behind it. Delivering Helios on schedule and showing that its performance claims translate into a real production workload throughput advantage and total cost of ownership gains will determine how much the competitive gap narrows. And of course, this is in a market that is clamoring for ever-more compute resources with a seemingly insatiable demand for AI services and capacity. That’s an environment for big iron success. Now AMD just has to deliver optimized, turnkey AI platforms. This is far easier said than done, but time will soon tell as deployments take shape this year.
This article is published as part of the Foundry Expert Contributor Network. Want to join?
Nvidia and Microsoft this month touted the reinvention of computers with a new class of “agentic AI PCs” that will “reinvent the way PCs work.”
That’s how Nvidia CEO Jensen Huang described the computers at the recent Computex trade show. At the event, Nvidia introduced its first AI-focused PC chip called N1X, which has an integrated CPU and GPU and will be used in agentic AI PCs.
Nvidia’s new RTX Spark PCs are the first in a major “PC reinvention for 40 years,” Huang said, likening them to AI phones. “You could talk to it, it could look at you. You could ask it to read files… [or] go help you do research.”
Not so fast, say analysts, who argue the computers are mostly repackaged AI PCs that shouldn’t necessarily drive enterprise upgrades.
“Agentic AI PCs is a strange term that should probably be deemphasized,” said Leonard Lee, principal analyst at neXt Curve. “Depending on use case, PCs of the last two generations are ‘agentic AI’-capable.”
Skeptical of the hype, analysts said many current PCs are already capable of running agents, such as those spun up on device by OpenClaw. Those agents don’t need to communicate with AI models in the cloud.
In fact, Lee said, Apple has already shown that Windows isn’t needed to run agentic workloads, with the Mac Mini among the most talked-about device for hosting personal AI agents and OpenClaw.
Analysts said the definition of agentic AI PCs really comes down to the hardware inside, which largely defines their capabilities. Most are really just an evolution of earlier AI PCs, which arrived a few years ago amid similar hype.
AI PCs, specifically CoPilot+ PCs, were originally designed with neural processing units (NPUs) to support the requirements of Windows Recall. “Some argue agentic AI PCs need more GPU compute, but…Qualcomm and Microsoft would counter that agentic AI PCs have been around for a while,” Lee said.
To highlight the capabilities of the RTX Spark PC, a demonstration at Computex showed how an architectural design workflow could be split between the PC and the cloud. (An MCP server managed data exchange to facilitate the cloud-PC workflow.)
Adobe has reengineered Photoshop and Premiere for RTX Spark PCs, allowing the software to communicate with AI agents on PCs and tools to run twice as fast, Huang said.
And at the company’s recent Build conference, Microsoft CEO Satya Nadella demonstrated the new Surface Laptop Ultra AI PC based on the RTX Spark design. He positioned agentic AI PCs as running a new execution layer that can act across files and devices, including generating and executing code.
While AI PCs with agents will be capable, they still lack applications, said Jim McGregor, principal analyst at Tirias Research. “However, PCs — especially the small form factor PCs and workstations — are likely to be the home AI appliance,” he said.
For enterprises, upgrading to RTX Spark PCs at this point could pose problems. That’s because Nvidia’s N1X CPU is based on the Arm processor architecture and could run into compatibility issues with x86 applications designed for Intel and AMD chips.
“Despite Arm processors being available for a while, enterprise penetration stays small, since compatibility with all apps, drivers, and corporate systems is critical and requires extensive testing…, even though Microsoft has done a reasonably good job with Windows 11 on Arm,” said Jack Gold, principal analyst at J. Gold Associates.
Because Microsoft is moving to natively incorporate AI into Windows, those PCs will land in enterprises ultimately, Gold said.
Regardless of hype, it might be a good time for enterprises chasing an agentic AI agenda to upgrade laggard Windows 10 devices, Lee said. “Considering the considerable hype about agentic AI, early adopters will need to upgrade their fleets if they pursue a broad agentic AI agenda for their organizations,” he said.
Intel is invading the physical AI space with a reentry into the robotics market it quit many years ago amid financial struggles.
The robotics strategy is part of the company’s larger plan to establish AI on the “edge,” in which devices have the computing capability to run AI locally. Many devices lack AI capabilities and have to offload processing to the cloud.
The chipmaker said its Intel Series 3 processors are now in 130 edge AI and robotics designs. It also had a design win with SensoryAI, which provides technology for robots that include Ella, a robotic barista made by Crown Digital.
The company’s Core Ultra Series 3 processors are derivatives of chip designs intended for laptops. But Intel has achieved a level of power efficiency for long battery life that allows those chips to be adapted for handheld devices and laptops.
Intel also said it can build advanced robotics chips thanks to its latest manufacturing technologies.
For example, many robotic functions, such as computer vision and real-time controls, can be integrated into a single chip. Previously, functions like graphics and movement and control were distributed among different cores in a chip.
SensoryAI, for example, has a chip architecture that provides the robotic barista — which is more like a robotic arm — with AI capabilities, Intel said.
The main “Avatar” agent handles customers as the main “Ella” agent reasons and executes the task. If Ella encounters errors, it passes on the issue to a Guardian agent, which helps with the recovery. Some issues could include making sense of an order, or cups that might be stuck.
The three agents are embedded in a single piece of Core Ultra Series 3 silicon.
Intel is displaying some of those robots at the Computex trade show in Taiwan. The company shared a video of a humanoid-style robot from the floor in a X.com post
This is not Intel’s first attempt at the robotics market. Intel sold robotics chips and kits when it was a dominant chip player in the field, but curtailed efforts in 2021 after Pat Gelsinger took over as CEO and restructured the company to focus on manufacturing.
Robotics is now back on the menu under new Intel CEO Lip-Bu Tan, who replaced Gelsinger last year. He has restructured to company to focus on high-growth areas that can generate high returns.
A Morgan Stanley study last year indicated the robotics market could be worth $5 trillion by 2050 — and more than 1 billion humanoid robots could be in operation.
However, challenges remain. There isn’t yet enough real-world data to train robots to do targeted work. And the AI models — generally called world models — they will need are still under development.
Training robots to do a specific job requires a sequence of events to happen in succession without any errors. Companies are still training robots to spot and understand errors, analyze possible resolutions, and take the right corrective action.
Login
