Skip to content
ContentLora

    Tip: press / anywhere to search.

    ● Developing story

    Next-gen computing hardware tracker: 2024-2026 milestones

    This tracker follows the move from conventional chip scaling to new transistors, power delivery, packaging, optics and compute paradigms. As of October 2026, GAA transistors with backside power are shipping in Intel's 18A, TSMC's N2 and Samsung's SF2 are in volume production, HBM4 memory is shipping, and co-packaged optical switches are in production.[1][2][3][4][5]

    Editor reviewedUpdated Next-gen computing hardwareSemiconductorsComputing

    What we know

    • TSMC's N2, its first nanosheet process, entered volume production in 4Q 2025.[2]
    • Intel 18A, with RibbonFET and PowerVia backside power, ships in Core Ultra Series 3 since January 2026.[6]
    • NVIDIA's Spectrum-X Ethernet Photonics CPO switches were in production by May 2026.[5]
    • UCIe 3.0 raised chiplet link rates to 64 GT/s in August 2025.[26]
    • The largest neuromorphic system, Intel's Hala Point, is a research prototype.[11]
    • CoWoS packaging demand was still expected to exceed supply at the end of 2026.[22]
    • Samsung's 2nm SF2 GAA process is in mass production in the Exynos 2600.[3]
    • HBM4 is in mass production at SK hynix and Samsung; Samsung began shipping in February 2026.[4]
    • Intel plans 14A risk production in 2H 2027 and a volume ramp in 2028, with no external customer announced.[7]
    • NVIDIA plans optical NVLink links between racks for Rubin Ultra NVL576 and Kyber NVL1152.[10]

    What we don't know yet

    • Whether TSMC's A16, with backside power, was production-ready in 2H 2026 as scheduled, and which customers use it first.
    • Whether any external customer commits to Intel 14A, as Intel expected from 2H 2026.
    • When co-packaged optics reach NVLink scale-up in volume, after reports that NVL576 and a CPO NVSwitch may slip.
    • Whether TSMC's N2P met its second-half-2026 volume production schedule.
    • How Intel 18A yields compare with TSMC N2 in independent data.
    • Independent, at-scale measurements of co-packaged optics power and reliability versus pluggables.
    • Whether any neuromorphic or analog in-memory chip will run a production-scale AI model commercially.
    • Whether packaging capacity catches up with AI demand in 2027.
    Story status

    Story status

    State
    developing
    Started
    2022-06-30
    Timeline entries
    27
    Last entry

    Timeline data (JSON)

    Where things stand

    As of October 2026, the incremental technologies have crossed into volume production. TSMC’s N2 has been in volume production since the fourth quarter of 2025, and Intel’s 18A, combining gate-all-around transistors with backside power, has shipped in laptops since January 2026.[2][6] Samsung’s Exynos 2600 brought gate-all-around transistors to 2nm, and Intel now plans its next node, 14A, for a 2028 volume ramp using High-NA EUV.[3][7][8] Memory moved too: SK hynix and Samsung began HBM4 mass production, with Samsung shipping from February 2026.[9][4] Co-packaged optical switches moved into production in 2026, and NVIDIA has said optical links will extend its NVLink scale-up domain across racks.[5][10] Neuromorphic and analog in-memory computing remain at the research and prototype stage.[11][12]

    What to watch next

    • TSMC N2P and A16. TSMC scheduled N2P volume production for the second half of 2026 and says A16, which adds a backside power rail, will be production-ready in the same period; A14 follows in 2028.[13][14][15]
    • Intel 14A customers. Intel said external customers would start making 14A decisions in the second half of 2026; none had been announced by its July update.[16][7]
    • NVLink optics. Whether optical scale-up arrives with Rubin Ultra NVL576, or slips as a July 2026 report suggested.[10][17]
    • HBM4E. Samsung planned HBM4E samples for the second half of 2026.[18]
    • Optics moving toward the GPU. Watch for optical I/O on processors rather than only in switches, building on optical chiplets and interposers shown in 2025, and for Broadcom’s fourth-generation CPO.[19][20][21]
    • Packaging capacity. Whether CoWoS supply catches up with demand by the end of 2026, as some investors expect, and whether TSMC’s wafer-scale packaging arrives on its 2027 timeline.[22][23]
    • After nanosheets. imec’s forksheet and CFET results at VLSI and IEDM conferences signal how the transistor roadmap is holding up.[24]
    • Neuromorphic and analog at scale. Any successor to Hala Point or a commercial analog accelerator running a large model would be a major milestone.[11][25]

    How this tracker is maintained

    Entries are added from official feeds of labs, standards bodies and journals listed for this field, and each entry links to its claims. Status labels follow the site’s rules: confirmed entries rest on primary sources or multiple reports, reported entries on a single credible outlet.[5] For the wider supply picture, see AI chip supply concentration.

    Timeline

    19 confirmed8 reported

    1. reported

      SK hynix reported in volume with 16-layer HBM4[45]

      A supply-chain tracker reported 48GB, 16-layer stacks in mass production from the third quarter; Micron had been shipping 12-layer HBM4 for Rubin since March.

    2. reported

      Intel says 14A defect density beats its target curve[44]

    3. reported

      Intel commits to a 14A volume ramp in 2028[7]

      Risk production for Intel's own products is planned for the second half of 2027.

    4. reported

      Report says NVIDIA's Kyber rack and CPO-linked NVL576 may slip[17]

      Tom's Hardware, citing SemiAnalysis, reported a delay to 2028; NVIDIA said only that its roadmap was intact.

    5. reported

      CoWoS supply gap reported narrowing as TSMC expands capacity[43][22]

    6. confirmed

      NVIDIA says Spectrum-X Ethernet Photonics is in production[5][42]

    7. confirmed

      NVIDIA outlines optical NVLink scale-up for Rubin Ultra NVL576 and Kyber NVL1152[10][41]

      Eight 72-GPU racks in one NVLink domain using copper and direct optical connections; Spectrum-6 switches carry 200 Gb/s co-packaged optics.

    8. confirmed

      imec updates CFET roadmap for the A7 node and beyond[24][40]

    9. confirmed

      Samsung begins HBM4 mass production and commercial shipments[4][18]

      11.7Gbps per pin, up to 13Gbps, on 1c DRAM with a 4nm logic base die (Samsung figures). HBM4E sampling planned for 2H 2026.

    10. reported

      Marvell completes acquisition of Celestial AI[38][39]

    11. confirmed

      Intel confirms High-NA EUV for 14A, has no committed external customer yet[16][8]

    12. confirmed

      Intel launches Core Ultra Series 3 on Intel 18A[6]

    13. confirmed

      TSMC N2 enters volume production in the fourth quarter[2]

      TSMC confirms volume production started in 4Q 2025; the exact start date was not announced.

    14. confirmed

      Samsung lists the 2nm Exynos 2600 in mass production[3]

      Samsung calls SF2 the industry's first 2nm GAA process; a June 2026 teardown confirmed it in Galaxy S26+ phones.

    15. confirmed

      Intel details Panther Lake, first product on 18A with RibbonFET and PowerVia[1][37]

    16. reported

      Broadcom samples Tomahawk 6-Davisson, a 102.4 Tb/s CPO switch[35][36]

    17. confirmed

      SK hynix completes HBM4 development and readies mass production[9][34]

      2,048 I/O connections double bandwidth over HBM3E; speeds above 10Gbps against the 8Gbps JEDEC standard (SK hynix figures).

    18. confirmed

      Nature Computational Science paper reports analog in-memory attention for LLMs[33][12]

      Up to four orders of magnitude lower attention energy than GPUs, with gain-cell arrays limited to 64x64 and an adaptation algorithm to map pre-trained models.

    19. confirmed

      UCIe 3.0 chiplet standard doubles link speed to 64 GT/s[26]

    20. confirmed

      Lightmatter and Ayar Labs unveil photonic interposer and UCIe optical chiplet[20][19]

    21. confirmed

      NVIDIA announces Spectrum-X and Quantum-X Photonics co-packaged optics switches[31][32]

    22. confirmed

      IBM details three studies on analog in-memory computing for transformers[25]

    23. confirmed

      Nature publishes roadmap for neuromorphic computing at scale[30]

    24. confirmed

      IBM reports NorthPole language-model inference results[29]

      Under 1 ms per token on a 3-billion-parameter model, with 72.7x the energy efficiency of a comparable GPU (IBM figures).

    25. reported

      TSMC outlines wafer-scale packaging plans for 2027[23]

    26. confirmed

      Intel unveils Hala Point, a 1.15-billion-neuron neuromorphic system[28][11]

      Built from 1,152 Loihi 2 chips and deployed at Sandia National Laboratories as a research prototype.

    27. confirmed

      Samsung starts 3nm production with gate-all-around transistors[27]

      The first production GAA process, marking the start of the post-FinFET era.

    Sources

    Each numbered claim is a statement we checked against the sources listed with it. Status shows how well established it is.

    1. [1]

      Intel 18A, used for Panther Lake, combines RibbonFET transistors with PowerVia backside power delivery. confirmedas of 2025-10-09

    2. [2]

      TSMC's 2nm (N2) process started volume production in the fourth quarter of 2025, as planned. confirmedas of 2026-10-10

      • 2nm Technology · TSMC · 2nm Technology section (retrieved 2026-10-10)
    3. [3]

      Samsung's Exynos 2600 smartphone processor, which Samsung lists as built on the industry's first 2nm gate-all-around process, was listed in mass production in December 2025. confirmedas of 2025-12-19

    4. [4]

      On February 12, 2026 Samsung said it had begun mass production of HBM4 and shipped commercial products, running at 11.7Gbps (up to 13Gbps) with its sixth-generation 10nm-class (1c) DRAM and a 4nm logic base die. confirmedas of 2026-02-12

    5. [5]

      By May 31, 2026, NVIDIA said its Spectrum-X Ethernet Photonics co-packaged-optics switches with 200Gb/s SerDes were in production. confirmedas of 2026-05-31

    6. [6]

      Intel launched Core Ultra Series 3 (Panther Lake), the first compute platform built on Intel 18A, at CES on January 5, 2026, saying systems would be available that month. confirmedas of 2026-01-05

    7. [7]

      In July 2026 Intel said it remained on track for 14A risk production of its own products in the second half of 2027 and had decided to fully commit to a high-volume 14A ramp in 2028. reportedas of 2026-07-24· forecast

    8. [8]

      Intel confirmed in January 2026 that High-NA EUV lithography is targeted at its 14A process. reportedas of 2026-01-23

    9. [9]

      On September 12, 2025 SK hynix said it had completed HBM4 development and prepared for mass production, doubling bandwidth with 2,048 I/O connections, improving power efficiency by more than 40% and exceeding the JEDEC 8Gbps standard speed with more than 10Gbps. confirmedas of 2025-09-12

    10. [10]

      In March 2026 NVIDIA said Vera Rubin Ultra NVL576 will join eight 72-GPU racks into one 576-GPU NVLink domain using copper and direct optical connections, and that its later Kyber racks will scale to NVL1152 with similar direct optical rack-to-rack links. confirmedas of 2026-03-16· forecast

    11. [11]

      Hala Point was initially deployed at Sandia National Laboratories and is a research prototype rather than a commercial product. confirmedas of 2024-04-17

    12. [12]

      The analog attention design could not run pre-trained models directly because of gain-cell non-idealities, so it needed an adaptation algorithm, and its gain-cell arrays were limited to 64x64 to contain voltage (IR) drop. confirmedas of 2025-09-08

    13. [13]

      TSMC schedules volume production of its enhanced N2P process for the second half of 2026. reportedas of 2026-10-10· forecast

    14. [14]

      TSMC says A16 will offer 8-10% higher speed at the same voltage or 15-20% lower power at the same speed than N2P, with up to 1.10x chip density, and will be production-ready in the second half of 2026. confirmedas of 2026-10-10· forecast

      • 2nm Technology · TSMC · A16 Technology section (retrieved 2026-10-10)
      • 2nm Technology · TSMC · A16 Technology section (retrieved 2026-10-10)
    15. [15]

      TSMC schedules its A14 process, a full-node step achieved through dimensional scaling, for volume production in 2028, promising 10-15% more speed or 25-30% less power than N2 and more than 20% higher logic density. confirmedas of 2026-10-10· forecast

      • 2nm Technology · TSMC · A14 Technology section (retrieved 2026-10-10)
      • 2nm Technology · TSMC · A14 Technology section (retrieved 2026-10-10)
    16. [16]

      In January 2026 Intel said it had no committed external customer for its 14A process yet, but that customers were evaluating its 0.5 process design kit and would begin making firm supplier decisions from the second half of 2026 into the first half of 2027. confirmedas of 2026-01-23

    17. [17]

      In July 2026 Tom's Hardware, citing SemiAnalysis, reported that NVIDIA's Kyber rack had slipped to 2028, that the eight-rack NVL576 linked by co-packaged optics was likely to slip or ship in low volume, and that a production-ready co-packaged-optics NVSwitch was not expected before the Feynman generation; NVIDIA responded only that its roadmap was intact. reportedas of 2026-07-06

    18. [18]

      Samsung said it expected its HBM sales to more than triple in 2026 and planned HBM4E sampling in the second half of 2026, with custom HBM samples in 2027. confirmedas of 2026-02-12· forecast

    19. [19]

      Ayar Labs unveiled on March 31, 2025 an 8 Tbps TeraPHY optical I/O chiplet with a UCIe electrical interface, powered by a 16-wavelength light source. confirmedas of 2025-03-31

    20. [20]

      Lightmatter unveiled the Passage M1000 on March 31, 2025, a photonic interposer of more than 4,000 square millimetres offering 114 Tbps of optical bandwidth, built on GlobalFoundries' Fotonix platform. confirmedas of 2025-03-31

    21. [21]

      Broadcom says a fourth-generation co-packaged-optics platform targeting 400 Gbps per channel is in development. reportedas of 2025-10-08· forecast

    22. [22]

      Institutional investors cited by Taiwan's Economic Daily News expected the CoWoS supply-demand gap to narrow from about 20% to about 10% by the end of 2026. reportedas of 2026-06-15· forecast

    23. [23]

      In 2024 TSMC projected a 2027 wafer-scale System-on-Wafer package with more than 40 reticles' worth of silicon and room for more than 60 HBM stacks. reportedas of 2024-04-30· forecast

    24. [24]

      imec's logic roadmap has the forksheet transistor extending nanosheets to the A10 node and monolithic CFET arriving at the A7 node. reportedas of 2026-02-26· forecast

    25. [25]

      In January 2025 IBM Research described three studies on analog in-memory computing for transformer models - a 3D mixture-of-experts architecture featured in Nature Computational Science, a phase-change-memory edge accelerator presented at IEDM, and a transformer run on an analog chip, published in Nature Machine Intelligence. confirmedas of 2025-01-29

    26. [26]

      The UCIe 3.0 specification, released on August 5, 2025, raised chiplet link data rates from 32 GT/s to 48 GT/s and 64 GT/s. confirmedas of 2025-08-05

    27. [27]

      Samsung announced on June 30, 2022 that it had begun production on a 3nm process using gate-all-around transistors, calling it the world's first 3nm process with its MBCFET design. confirmedas of 2022-06-30

    28. [28]

      Intel's Hala Point, announced April 17, 2024, contains 1.15 billion neurons on 1,152 Loihi 2 processors made on the Intel 4 process. confirmedas of 2024-04-17

    29. [29]

      IBM reported in September 2024 that NorthPole ran a 3-billion-parameter language model at under 1 millisecond per token with 72.7 times the energy efficiency of the next-lowest-latency GPU. confirmedas of 2024-09-26

    30. [30]

      A review by 23 researchers published in Nature on January 22, 2025 set out a roadmap for scaling neuromorphic computing. confirmedas of 2025-01-23

    31. [31]

      NVIDIA announced Quantum-X Photonics InfiniBand switches for availability in 2025 and Spectrum-X Photonics Ethernet switches for 2026. confirmedas of 2025-03-18

    32. [32]

      NVIDIA said in March 2025 that its silicon photonics switches use 4x fewer lasers and deliver 3.5x more power efficiency and 10x better resiliency than traditional methods. confirmedas of 2025-03-18

    33. [33]

      A September 2025 Nature Computational Science paper reported an analog in-memory attention design that cut attention latency and energy by up to two and four orders of magnitude, respectively, compared with GPUs. confirmedas of 2025-09-08

    34. [34]

      SK hynix said it used its market-proven Advanced MR-MUF stacking process for HBM4, which injects a liquid protective material between stacked chips and hardens it, to minimise mass-production risk. confirmedas of 2025-09-12

    35. [35]

      In October 2025 Broadcom began sampling Tomahawk 6-Davisson, its third-generation co-packaged-optics Ethernet switch with 102.4 Tb/s of capacity, to early-access customers. reportedas of 2025-10-08

    36. [36]

      Broadcom says Tomahawk 6-Davisson cuts optical interconnect power by about 70% compared with pluggable transceivers. reportedas of 2025-10-08

    37. [37]

      Intel says 18A delivers up to 15% better performance per watt and 30% better chip density than Intel 3, and calls it the first 2-nanometer-class node developed and manufactured in the United States. confirmedas of 2025-10-09

    38. [38]

      Marvell completed its roughly $3.25 billion acquisition of optical-interconnect startup Celestial AI on February 2, 2026. reportedas of 2026-02-02

    39. [39]

      Celestial AI claims its Photonic Fabric can raise off-package bandwidth between data center chips by up to 25 times while substantially reducing power. reportedas of 2026-02-02

    40. [40]

      imec's design study for monolithic CFETs, presented at IEDM 2025, found that giving n- and p-type devices different channel orientations can raise drive current by up to 20%, a booster it expects the A3 node to need. confirmedas of 2026-02-26

    41. [41]

      NVIDIA's Spectrum-6 switch offers 102.4 Tb/s with 512 lanes and 200 Gb/s co-packaged optics, replacing pluggable transceivers. confirmedas of 2026-03-16

    42. [42]

      NVIDIA claims Spectrum-X Ethernet Photonics delivers 5x better power efficiency and 5x longer AI uptime than networks using traditional transceivers. confirmedas of 2026-05-31

    43. [43]

      TSMC's monthly CoWoS packaging capacity is reported to reach 120,000-140,000 wafers in 2026, with partner packaging firms adding 50,000-60,000. reportedas of 2026-06-15

    44. [44]

      In August 2026 Intel's chief financial officer said 14A defect density was tracking better than Intel's target curve, though production remained about two years away. reportedas of 2026-08-28

    45. [45]

      Supply-chain tracker Silicon Analysts reported in September 2026 that SK hynix began mass-producing a 48GB, 16-layer HBM4 stack in the third quarter, while Micron had been in volume production of 12-layer HBM4 for NVIDIA's Rubin since March. reportedas of 2026-09-15

    Revision history (2)
    1. Page created.
    2. Brought the tracker up to October 2026 with Samsung SF2, Intel 14A, HBM4 and NVLink optics; fixed the analog attention paper date (8 September 2025) and corrected superseded claims.

    Created Oct 10, 2026. Last reviewed by an editor on Oct 10, 2026. Next scheduled review: Jan 10, 2027.

    Cite this page

    "Next-gen computing hardware tracker: 2024-2026 milestones." ContentLora, updated Oct 10, 2026. https://contentlora.com/events/computing-hardware-tracker

    Spotted an error? Suggest a correction or emailcorrections@contentlora.com.

    Keep exploring

    ☰ Contents