Intel has officially unveiled its "Wildcat Lake" Core 300 series of processors for the entry-level PC segment. These CPUs are designed for value-oriented buyers looking for good CPU performance and basic GPU output in small form-factor commercial and edge AI PCs. Officially rated at 40 TOPS, these processors are capable of Copilot+ AI PC certification, meaning Intel has managed to provide basic local AI processing functionality to entry-level buyers. As part of the Core 300 series non-Ultra family, Intel's main selling point is all-day battery life in laptops with local AI processing. There are three segments, each featuring a hybrid core configuration, pairing two "Cougar Cove" P-cores with four LPE "Darkmont" cores.

The actual SoC package integrates two dies. The first and most important one, built on the 18A internal Intel node, features a 6-core CPU configuration, NPU 5 with 40 TOPS of INT8 data, Xe display and media engine, and a GPU that includes up to two Xe3 cores. This die also contains the memory controller and cache pool, supporting LPDDR5X memory running at 7,467 MT/s or up to 6,400 MT/s for DDR5. The memory-side cache includes 4 MB to help tasks "buffer" into this cache before and after accessing memory. Intel dedicates the second die to I/O handling, as the platform supports 6 PCIe Gen 4 lanes, two Thunderbolt 4 connections, two USB 3.2 connectors, and up to eight USB 2.0 connectors. For Wi-Fi, there is the Wi-Fi 7 standard, and Intel also includes Bluetooth 6.0.

Intel's upcoming "Crescent Island" series of GPUs, based on the newly developed Xe3P graphics IP, will be hitting the market this year. However, it seems this IP won't be featured in the desktop Arc gaming series of discrete GPUs. According to a reliable leaker, Jaykihn, Intel will not release this GPU generation as dedicated Arc gaming graphics cards. Instead, Xe3P-based GPUs will power "Crescent Island" data center and workstation GPUs, similar to the current Arc Pro series for professional users. This leaves Arc gamers without a clear indication of when, or even if, they will receive a new gaming Arc GPU based on the Xe3 or Xe4 IP in the coming months.

Intel recently announced the Arc Pro B70 and Arc Pro B65 graphics cards, which maximize the BMG-G31 "Battlemage" Xe2 silicon but are intended for professional users only. A recent driver update added the ability to play games on these cards, but they are still primarily designed for AI workloads and professional visualization. Gamers are still seeking clarity about future updates, and the lack of recent rumors regarding an additional Arc gaming discrete GPU is concerning. Even the higher-end Arc B770, which has been rumored for some time, seems to be on hold. Upcoming events like Computex might offer Intel a chance to at least provide a teaser of what the desktop Arc GPU family will look like or if there will be any updates soon.

Intel Foundry has been attracting significant attention from the industry lately, with some of the world's largest companies expected to announce new foundry commitments this fall, according to UBS Group. The investment bank has reported that Intel Foundry is on the brink of securing many new contracts scheduled for this fall, citing the release of version 1.0 of the 14A node process design kit (PDK) as a key catalyst. There have been numerous rumors that Apple, AMD, NVIDIA, Google, and Broadcom are considering utilizing Intel's silicon manufacturing, including the 18A, 18A-P, and 18A-PT nodes, as well as the upcoming 14A node. Apple is rumored to manufacture some of its M-Series "Apple Silicon" laptop processors in 2027 using the 18A-P node, while Google might leverage Intel's EMIB and Foveros 3D advanced packaging technology for some of its TPU designs.

Companies regularly evaluate leading industry nodes but often choose TSMC's manufacturing due to its reliability, high volume capacity, and advanced packaging, which has produced numerous high-performance and low-power designs at scale. However, Intel Foundry has been investing significant resources and logistics into attracting external clients, leading UBS to anticipate multiple foundry commitments being announced this fall. Late last year, we learned that Apple was waiting for Intel to release the 18A-P PDK version 1.0 or 1.1, scheduled for Q1 and Q2 of 2026, respectively. As we are now in Q2, we are awaiting further confirmation to see if Apple proceeded with the deal, but UBS expects it did.

NVIDIA today announced the world's first family of open source quantum AI models, NVIDIA Ising, designed to help researchers and enterprises build quantum processors capable of running useful applications. To achieve useful quantum applications at scale, significant breakthroughs are needed in quantum processor calibration and quantum error correction. AI is key for turning today's quantum processors into large-scale, reliable computers. Open models empower developers to build high-performance AI while maintaining total control over their data and infrastructure.

Named after a landmark mathematical model that dramatically simplified the understanding of complex physical systems, the NVIDIA Ising family provides high-performance, scalable AI tools for quantum error correction and calibration — two of the most critical challenges in building hybrid-quantum classical systems. Ising models run the world's best quantum processor calibration and enable researchers to tackle much larger, more complex problems with quantum computers by delivering up to 2.5x faster performance and 3x higher accuracy for the decoding process needed for quantum error correction.

Valve is implementing a number of new changes to its Steam platform, all aimed at enhancing the quality of the user experience for gamers. One of the latest updates involves adding a 30-day price history to all listed games. This feature will allow buyers to see if a game's price has changed recently, promoting greater transparency. Previously, gamers had to rely on third-party tools like SteamDB to track price fluctuations over time. Now, the Steam platform itself is set to include price tracking, integrating this functionality directly. This update will also highlight when a game is on discount, the percentage drop compared to its launch price, and how it compares to its price over the last 30 days. Gamers will be able to use these metrics to decide whether to wait for a Steam Sale or purchase the game immediately if it's a good deal.

Additionally, Valve is reportedly developing the SteamGPT AI system to assist with customer support queries, covering issues such as refunds, platform problems, and payment processing, among others. There is also talk of a new "Frame Estimator" tool that can predict your PC's performance before you buy a game. While Valve hasn't confirmed these features or provided a timeline for their release, these developments suggest preparations for future updates, which could be unveiled soon. Below is a sample price tracker from SteamDB for NBA 2K26 game, as we could see something similar on the main Steam platform soon.

Elon Musk announced today on X that Tesla has completed the design of a new generation of its AI chips for Full Self-Driving (FSD). The CEO also mentioned earlier that the AI5 chip will offer performance comparable to NVIDIA's "Hopper" architecture, with two AI5 units matching the power of a single "Blackwell" processor. In late 2025, reports indicated that Samsung had achieved a significant win for its previously struggling foundry business, as Tesla decided to split the manufacturing of its new AI5 accelerator between Samsung and TSMC. The chips will be produced at Samsung's plant in Taylor, Texas, and TSMC's facility in Arizona. This decision is part of a strategy to keep the supply chain diversified and maintain chip supply under control for any demand scenario.

Samsung and TSMC are not the only partners in this project. Tesla is also sourcing DRAM chips from SK hynix, which appear to be LPDDR5X memory integrated into the package. On both the left and right sides, there are two rows of SK hynix LPDDR5X memory modules, each with three modules. This totals 12 LPDDR5X memory modules per AI5 chip. With 16 GB per module, this results in 192 GB of LPDDR5X memory per single AI5 SoC.

In the latest Q1 2026 quarterly figures, ASML announced plans to ship over 60 EUV units this year, including both High-NA and Low-NA EUV lithography scanners. This is a significant increase from the 48 units shipped in 2025, but it doesn't break any records compared to the company's expectations for 2027. ASML predicts shipping about 80 EUV systems in 2027, driven by massive demand from memory makers. As memory has become the most sought-after commodity in AI data centers, there is simply not enough supply to meet demand. In Q1 2026, ASML reported that nearly half—45% to be precise—of its revenue came from South Korea.

While ASML hasn't provided specific data per manufacturer, we know that South Korean semiconductor makers primarily focus on memory, storage, and to a lesser extent, logic production. More than half of the quarterly shipments, at 51%, are dedicated to memory production. Companies like SK hynix and Samsung are purchasing many EUV systems to support their growing memory and storage businesses, which include the latest technologies like GDDR6, HBM3, HBM4, and DDR4/DDR5, all consumed by AI data centers worldwide. SK hynix plans to install 20 Low-NA EUV units in the next two years, all designed for HBM memory and advanced storage solutions, making SK hynix one of ASML's largest customers.

AMD's China store on JD has just started accepting pre-orders for the upcoming Ryzen 9 9950X3D2 Dual Edition desktop processor. Priced at an MSRP of $899, this model is $200 more expensive than the launch MSRP of the Ryzen 9 9950X3D. In China, AMD's store requires pre-orders with a deposit of 100 RMB, which counts as 300 RMB towards the final payment due to the "early bird" pre-order benefit. The final payment will be made on April 22, the planned retail launch date. Additionally, at launch, the processor will be available both as retail PIB (processor in box) packages and as part of pre-built gaming desktops by popular OEMs and system integrators. Currently, no Western stores offer pre-orders for this CPU, either in the United States or Europe. Most likely, these stores will start accepting orders on April 22. Some, like B&H, have a placeholder to notify customers when it becomes available.

The Ryzen 9 9950X3D2 Dual Edition is a 16-core, 32-thread processor built on the "Zen 5" microarchitecture and features 3D V-Cache technology, with stacked 3D V-Cache on both of its 8-core chiplets. This configuration results in an on-package L3 cache of 192 MB and a total cache (L2+L3) of 208 MB. In comparison, the 9950X3D has 3D V-Cache on only one of its two CCDs and relies on software-based OS scheduler guidance to ensure gaming workloads are directed to the CCD with 3D V-Cache. As a result, this new processor is more suited for productivity and workstation tasks, while the 9950X3D remains focused on gaming. Aside from the cache differences, the 9950X3D2 Dual Edition is mostly similar to the 9950X3D, except its maximum boost frequency is 5.60 GHz, which is 100 MHz lower than the 9950X3D, and it has a higher TDP of 200 W compared to the 170 W of the 9950X3D. While technically compatible with any Socket AM5 motherboard, support requires UEFI firmware updates from motherboard vendors.

During a media tour at Lexar HQ in China, Digital Foundry spoke with Lexar's Europe General Manager, Grace Su, who explained what gamers are buying and where they are willing to compromise. Interestingly, Lexar found that, despite the pricing of DRAM and NAND Flash, users could purchase lower-capacity RAM kits without issues, while SSDs with capacities lower than 512 GB weren't selling well. Although this is not a direct comparison, when building a new PC system, you need both a RAM kit and an SSD. Based on their budget, enthusiasts often have to decide between a larger SSD or a bigger RAM kit. However, Lexar reports that users are so committed to having larger SSD space that they are even willing to revert to HDDs rather than buy an SSD smaller than 512 GB. This is understandable, considering that modern games can take up hundreds of gigabytes of space.

Currently, high RAM prices are driving PC gamers to invest considerably more in their gaming rigs to play smoothly. As 16 GB is somewhat considered the minimum for a Windows 11 system, some games and tasks are so demanding that PC DIY enthusiasts need to invest in 32 GB kits at prices that were unimaginable just a year ago. For example, we reported on the DDR4 spot market pricing, which saw the cost of a single 16 Gb module increase by about 2,200% over the past year, with only a minor 5% correction in March. We can only hope for more corrections soon so that PC gamers can finally see signs of relief, but with AI demand projected to continue rising, that seems far from reality.

Yesterday, SemiAccurate posted a rumor that NVIDIA was planning to acquire a large PC maker, which would have reshaped its position within the PC space. However, it turns out the rumor is not true. NVIDIA issued a statement to CNBC saying, "The media report is false; NVIDIA is not engaged in discussions to acquire any PC maker." This brief statement effectively shuts down the rumors, suggesting that NVIDIA does not plan to go through with such an acquisition, even if it might have been considered in the past. Large corporations, especially with NVIDIA's $4.6 trillion market capitalization, are always looking to expand their business in areas where they are dominant and enter new markets where they could capture a significant share. NVIDIA holds a strong position in the dedicated GPU market, with about 94% market share among AIB GPU shipments.

The company is also looking to enter the laptop space with the upcoming N1/N1X SoCs, which would allow NVIDIA to tap into a completely new segment among PC users. In this context, acquiring a PC company would be a logical move, but it would present significant regulatory hurdles, as the company might not get approval from market regulators in the United States, European Union, and other regions. Now that we know the rumor is not true, we are waiting to see if any more rumors emerge about which company might have been the potential target. For now, we are refraining from making definitive claims or pointing fingers at any specific company.

Intel's upcoming Core Ultra 400 Series "Nova Lake-S" desktop processors are set to launch in the second half of this year, marking what seems to be the company's most ambitious lineup yet. Recently, new rumors have emerged suggesting that Intel is also working on a "Nova Lake-S" desktop APU featuring 12 Xe3P graphics cores. This configuration is expected to deliver graphics performance similar to the Arc B390 integrated GPU found in "Panther Lake," albeit with a slight twist in the overall GPU core design. We've already covered a recent leak detailing the entire Core Ultra 400 Series "Nova Lake-S" desktop lineup, but it seems Intel is developing and testing more in its labs than initially anticipated. Readers may recall brief rumors about a variant called "Nova Lake-AX," which was rumored to feature a single compute tile with eight "Coyote Cove" P-cores and 16 "Arctic Wolf" E-cores, along with a four-core LPE island, totaling 28 cores. The iGPU configuration for this variant was said to include 48 Xe3 cores, potentially making it one of the most powerful APUs ever.

However, since the "Nova Lake-AX" appears to be on hold, Intel is preparing other solutions with less powerful GPU configurations. According to a post by Jaykihn on X, the new desktop APU would feature four "Coyote Cove" P-cores, eight "Arctic Wolf" E-cores, and four LPE-cores. This 4+8+4 configuration would be paired with 12 Xe3P cores, which are slightly different from the regular Xe3 cores used in the standard "Nova Lake-S" desktop processors. Intel's lineup is becoming somewhat confusing, but it seems the main lineup will consist of standard desktop series split into several SKUs with high core counts and bLLC, or the newly rumored SKU with more Xe3P cores and fewer CPU cores, similar to AMD's Ryzen G-Series of desktop APUs. Do note that the image below is just an illustration, and not the actual render.

Apple has informed its supply chain that the company now aims to produce a total of 10 million first-generation MacBook Neo laptops, as consumer demand has been phenomenal. Initially, with the MacBook Neo launch, Apple expected consumers to purchase between 5 and 8 million units throughout the lifecycle of the first generation. However, since demand has exceeded initial expectations, Apple is significantly increasing production to meet this demand. As the Cupertino-based company has access to a wide network of manufacturing partners, ramping up production is straightforward, provided the main component—the A18 Pro SoC—is consistently supplied by TSMC.

Additionally, Apple is already planning a second-generation MacBook Neo with major upgrades to the overall system, primarily due to the new A19 Pro SoC, which will come with 12 GB of RAM. The current MacBook Neo features the mobile A18 Pro chip and is limited to 8 GB of RAM. According to recent rumors, Apple may upgrade the MacBook Neo's internals in 2027, equipping it with an A19 Pro, the same SoC found in the latest iPhone 17 Pro smartphones. Inside the MacBook Neo, Apple has opted to reuse the iPhone 16 Pro's chip, which is produced by TSMC and includes 8 GB of LPDDR5X memory. This memory is directly attached above the A18 Pro SoC using Integrated Fan-Out Package on Package (InFO-PoP) technology, creating a 3D wafer-level fan-out package.

Epic Games launched its Epic Games Store in 2018 to challenge the dominance of Valve's Steam platform in the PC gaming space. Known for its promotions and free games, the storefront has attracted a large gamer base that quickly adopted the platform. However, two former Epic Games employees claim that the Epic Games Store saw only an initial spike in active users during the free game promotions, with gamers eventually returning to Steam. Many gamers find switching between multiple storefronts to be a cumbersome experience. While free game deals are an excellent way to attract customers initially, the Epic Games Store needs to make further improvements to retain those gamers.

Meanwhile, Valve's Steam platform continues to break records for concurrent gamers. Just a few weeks ago, the platform reached 42,318,602 players on Sunday, March 22 of this year. This set a new record, surpassing the previous impressive 42 million concurrent players on January 11 earlier this year. At the time of writing, Valve's Steam platform has 37,053,591 active players, indicating that a large majority keep returning, with no signs of slowing down.

Valve's Linux developer, Natalie Vock, has introduced a new method for handling memory management on GPUs with 8 GB or less VRAM. Since a large part of the gaming community uses systems with limited resources, improving performance and VRAM allocation is an excellent way to maximize frames per second. According to Vock's solution, GPUs with 8 GB of VRAM or less on Linux will now benefit from better background VRAM management, allowing more free space for games. Several Linux kernel patches have been submitted to guide how Linux manages resources when a game is loaded. Previously, when a game used too much VRAM on a low-VRAM GPU, the Linux Graphics Translation Table (GTT) would move data to system RAM to prevent the game from crashing. Since Linux doesn't prioritize programs, it could sometimes move the actual game from VRAM to system RAM to make space for something like a browser or another third-party app.

In Vock's testing before these new patches, running Cyberpunk 2077 resulted in the system using only 6 GB of the 8 GB of available GPU VRAM, with about 1.37 GB spilling over to GTT and being allocated in system RAM. This could lead to game stuttering and frame-pacing issues. Now, Vock has developed a real-time dmemcg-booster, which stands for Device Memory Control Groups, to inform Linux when a program running on the GPU VRAM needs to stay there without being moved to GTT and then to system RAM. With the new patch, running Cyberpunk 2077 now uses almost the entire available system VRAM, reaching 7.4 GB, while GTT allocation has been reduced to 650 MB, cutting usage by more than half. In the picture below, you can check the "GameThread" behavior before and after the patch is applied.

DDR4 pricing is finally showing signs of easing, as DDR4 16 Gb spot prices experienced a 5% correction in March after a more than 2,200% increase over the past year. The massive demand for AI data center expansion absorbed all available DDR production, making DDR4 and DDR5 pricing widely discussed topics among enthusiast communities. This has impacted PC DIY enthusiasts, laptop makers, PC OEMs, and everyone in between. Over the past year, the rapid price increase has effectively priced out many consumers from upgrading or installing new systems with larger RAM capacity. However, there are signs that the situation might improve as spot pricing has started to decrease.

First, readers need to understand the difference between spot and contract pricing in DDR memory. Contract pricing is what memory makers like Samsung, SK hynix, and Micron agree on with large buyers, such as PC OEMs like Apple, HP, or Dell, or even server makers who purchase DRAM in large quantities. On the other hand, spot pricing is an "on-the-spot" deal that provides immediate pricing without any contract. This spot pricing is typically reserved for smaller DRAM volumes and is usually bought by DRAM distributors and supply chain logistics. In the spot pricing market, DDR4 has fallen about 5% in March compared to February pricing, and the same applies to DDR5 memory spot pricing.

Popular PC DIY enthusiast tools like CPU-Z and HWMonitor were briefly compromised during the night between April 9 and April 10. This left anyone who downloaded these tools from the CPUID website vulnerable to malware they thought was a normal software tool. What actually happened was that multiple users on Reddit noticed their downloads from CPUID for both CPU-Z and HWMonitor were flagged multiple times by Windows Defender for security concerns. During that night, users attempting to download these popular utilities found the website pointing to a different host where the modified .exe files were stored. After realizing what was happening, users started removing these compromised .exe files and waited for CPUID to issue a statement. One of the owners of CPUID published a statement on X, explaining that attackers had replaced the original download links with malware-infected file links, resulting in a widespread security vulnerability. Thankfully the website is now back up and safe to resume downloads.

Hi,
Investigations are still ongoing, but it appears that a secondary feature (basically a side API) was compromised for approximately six hours between April 9 and April 10, causing the main website to randomly display malicious links (our signed original files were not compromised). The breach was found and has since been fixed. Sorry for the inconvenience. I did my best to fix that mess as soon as possible :-/
Sam.

Microsoft's Copilot AI has generated significant negative sentiment among Windows 11 users, to the extent that even Microsoft has recognized the need to scale back some of the Copilot functionality it has long promoted. According to Windows Latest, Microsoft has updated core Windows applications such as Notepad and Snipping Tool. In March 2025, Microsoft added Copilot to Notepad, updating the note-taking application to integrate Copilot AI for rewrites, summaries, and other AI-related text features. However, as users grew weary of the persistent AI push, Microsoft put these integrations under review earlier this year. In response to strong backlash from the enthusiast community, Microsoft has removed the Copilot AI icon from the latest versions of Notepad and Snipping Tool, restoring them to their regular look. However, Microsoft now calls these Notepad features like rewrite and summarize as "Writing tools."

This change is a significant positive step by Microsoft. Additionally, the forced inclusion of the Copilot AI button in every application has been paused, as there has been minimal interest from users in utilizing these features. The TechPowerUp Forums have been a consistent source of criticism against Microsoft's forced AI integration, with a large community of PC enthusiasts opposing the "AI everywhere" approach. The Redmond-based company is also working on numerous performance, security, and other enhancements for Windows 11, and we are already beginning to see some of these changes. Considering that Microsoft has 80 Copilot apps and services, this is not a thing to take lightly.Below is the before and after comparison with the Copilot AI removed, courtesy of Windows Latest.

Intel Foundry has been delivering some interesting research lately, and the newest installment is the world's thinnest gallium nitride (GaN) chiplet, with the base silicon measuring just 19 micrometers in thickness. The company manufactured a 300-millimeter GaN-on-silicon wafer, which is one of the first combinations of GaN with traditional silicon logic on a single chip. In traditional manufacturing setups, GaN is reserved for power electronics and is not mixed with silicon-based computational logic. However, Intel has managed to combine GaN semiconductors for power delivery with silicon-based compute logic, meaning that power chips can now perform basic computations and actions on their own, without needing a separate chip to dictate behavior.

Intel manufactured the GaN wafer with silicon combined on the 30 nm process, which reportedly demonstrated excellent properties. This includes stable current carrying, very low power loss, and the ability to block voltages up to 78 V without leakage. What is fascinating is that GaN is usually used for wide bandgap applications, with high radio frequency performance exceeding 300 GHz. GaN also provides much better power delivery due to its superior material properties and operates more stably at higher temperatures. This is especially significant for workloads like electric vehicles or data center point-of-load delivery, where space is limited, temperatures can reach about 150°C, and stability is crucial.

NVIDIA's upcoming N1 SoC, designed for gaming laptops, has officially appeared for the first time alongside its motherboard intended for a laptop. Featuring 128 GB of LPDDR5X memory running at 8,533 MT/s, the SoC is paired with SK hynix modules. NVIDIA's OEM partners for this launch, including Dell and Lenovo, have confirmed that the new NVIDIA laptop platform will be available this year, and we have seen early benchmark submissions of the N1 SoC. As NVIDIA prepares to enter the gaming laptop market, the company initially released its GB10 SoC for the DGX Spark mini workstation. This chip allowed NVIDIA to test its collaboration with MediaTek on developing new low-power SoCs that offer unusually high-performance graphics for AI applications, as well as very high RAM capacity.

In the latest images that appeared on Goofish, we see a laptop motherboard equipped with the NVIDIA N1 SoC and 128 GB of LPDDR5X memory closely packed in a specialized PCB area. This area might be designated for the cooler, which could potentially also cool the LPDDR5X memory modules. NVIDIA plans to introduce its N1/N1X SoCs with Arm-based CPU cores around Computex 2026. These will offer consumers 20 CPU cores, consisting of 10 Cortex-X925 and 10 Cortex-A725 CPU cores based on the Armv9.2 ISA, along with a "Blackwell" GPU optimized for low power settings with up to 6,144 CUDA cores in the higher-end form. You can also see the port selection on the side of the PCB, which seems to be USB-C or Thunderbolt 4/5, USB-A, and an HDMI connection. Since we don't know which laptop model this is, we have to wait a bit more before getting final details.

In the latest series of leaks about Intel's upcoming "Nova Lake" chips, we learn that the desktop "Nova Lake-S" designs will feature a new socket mounting and independent loading mechanism (ILM). According to exclusive information from VideoCardz, the NVL-S desktop designs on the LGA-1954 socket will include a new 2L-ILM socket mechanism to improve contact with CPU coolers. Given that Intel's upcoming NVL-S desktop processors will have up to 52 high-frequency, power-hungry cores, a better mounting mechanism is definitely needed. This 2L-ILM is a two-level independent loading mechanism, using one lever on each side of the CPU to secure it into the LGA-1954 socket. Applying pressure on both sides of the socket ensures improved flatness while the locking mechanism does its job.

Intel categorizes its desktop sockets in various forms. One of the most basic is the Default-ILM, along with the RL-ILM. As seen with "Arrow Lake," Intel's motherboards divide the socket locking mechanism into the ILM and RL-ILM designs, applied depending on the sector. For the lower-end sector, the Default-ILM socket is used, while the RL-ILM is applied to higher-end overclocking motherboards, providing better pressure and ensuring a flatter surface contact for a cooler. Noctua and Cooler Master already differentiate these two in their CPU coolers, but the mounting hardware is generally the same for both. The RL-ILM is simply a better design for overclocking due to the improved surface contact it provides.

Apple's MacBook Neo, MacBook Air, and the latest MacBook Neo series of laptops are now susceptible to a significant performance drop if they remain powered on for 49.7 days without a reboot. Specifically, after exactly 49 days, 17 hours, 2 minutes, and 47.296 seconds of continuous operation, the entire TCP/IP networking stack shuts down and stops establishing new connections. This unusual bug stems from a kernel-level issue in Apple's XNU kernel involving tcp_now, a 32-bit unsigned integer that tracks elapsed milliseconds since boot for the TCP stack. When it reaches its maximum value of 4,294,967,295, the counter stalls instead of resetting, disrupting the arithmetic macOS uses to expire and discard closed connections.

These inactive connection sessions accumulate quickly, increasing CPU usage as the system struggles to handle an ever-growing backlog. Once the available port pool, typically around 16,384 ports, is fully used, no new connections can be opened. Applications that rely on network access begin to fail, although already-established sessions continue to function, and the machine still responds to pings, albeit much more slowly. This initially delayed the diagnosis of the bug. A reboot resets the counter and restores normal operation, but the issue will recur on the same schedule after another 49.7 days of continuous use. There are potential solutions to fix the bug by using larger 64-bit integers, but this would require more extensive code changes, performance testing, and additional work compared to a simple fix.