March 20, 2026 Co-packaged Optics and and Quantum Intelligence Newsletter
NVIDIA GTC and OFC recap and other news
Photonics & Quantum Intelligence
NVIDIA GTC 2026: CPO Enters Full Production. NVLink8 CPO Confirmed for 2028. Feynman Roadmap Unveiled.
Jensen Huang's keynote on March 16 marked the most CPO-dense GTC in NVIDIA's history. The Spectrum-X Photonics CPO Ethernet switch is now confirmed in full production — the first unit is running live at Microsoft Azure. $1 trillion in AI factory demand was the headline number. Seven new chips were announced spanning GPU, LPU, CPU, DPU, SuperNIC, switch, and CPO switch product lines. The 2028 Feynman generation was formally revealed with NVLink8 CPO — co-packaged optics on the scale-up NVLink interconnect — at its core.
- In production now: Quantum-X800 InfiniBand CPO switch (Spectrum-6 Ethernet CPO scheduled 2H 2026)
- 2027 — Vera Rubin Ultra + Kyber NVL1152: 144 GPUs per NVLink domain; optical scale-up via NVLink 576 CPO switches; copper and direct optical connections co-exist
- 2028 — Feynman: NVLink8 CPO (silicon photonics on scale-up NVLink); Spectrum-7 CPO switch at 204T; Kyber racks with copper + CPO; Rosa CPU; LP40 LPU; BlueField-5; CX10
- Jensen on interconnects: "We need a lot more capacity for copper. We need a lot more capacity for optics. We need a lot more capacity for CPO." Copper is not going away.
- AWS commitment: Deploy 1M+ NVIDIA GPUs + Groq LPUs spanning Blackwell, Rubin, RTX PRO, and Groq 3 stack — the largest hyperscaler AI factory commitment ever announced at GTC
OFC 2026: 700+ Exhibitors, 16,000 Attendees — The CPO/NPO Transition Reaches Commercial Proof Point
OFC 2026 ran concurrently with GTC 2026 and delivered the industry's clearest demonstration yet that CPO has moved from concept to commercial product across multiple architectures. The show floor was dominated by 1.6T and 6.4T CPO and NPO demonstrations, and a new interoperability standards wave (Open CPX MSA, XPO MSA) is taking shape to prevent the market from fracturing around proprietary optical engine formats.
- Coherent: Multi-technology CPO showcase — 6.4T (32×200G) socketed SiPho CPO with ELS; multimode VCSEL socketed CPO; 400G InP modulator on silicon; 400G/lane PAM4 optical links targeting 3.2T; new XPO MSA founding membership for 12.8T+ liquid-cooled optics
- Ciena/Nubis: Vesta 200 6.4T CPX — first post-acquisition product; 70% power reduction; retimer-free linear-drive; highest-density pluggable CPO in class
- AIM Photonics: Presented PICs & Packaging for CPO and Quantum manufacturing; US-based integrated photonics production roadmap for scalable assembly infrastructure
- VIAVI Solutions: OFC 2026 test and measurement showcase — 1.6T Ethernet test, silicon photonics manufacturing solutions, PCIe over optics, fiber sensing; emphasis on AI-fabric-aware validation
- OIF Multi-vendor Interop Demo: 40 member companies demonstrating 800ZR, 400ZR, multi-span optics interoperability — the largest OIF interop in the conference's history
NVIDIA Confirms Spectrum-X CPO in Full Production at GTC 2026 — Feynman NVLink8 CPO Locked for 2028, Kyber Rack Scales to 1,152 GPUs
At his GTC 2026 keynote on March 16, Jensen Huang delivered the most consequential CPO roadmap update since the original Spectrum-X/Quantum-X announcement at GTC 2025. Three disclosures define the hardware landscape for the next three years. First: the Spectrum-X Photonics CPO Ethernet switch is confirmed in full production; the first unit is running live at Microsoft Azure. Second: the Rubin Ultra generation (2027) introduces the Kyber rack architecture — up to 1,152 GPUs in a single NVLink domain (NVL1152) with both copper and direct optical connections for rack-to-rack scale-up via NVLink 576 CPO switches. Third: the Feynman generation (2028) integrates NVLink8 CPO — co-packaged optics on the scale-up NVLink interconnect itself — alongside Spectrum-7, a 204T CPO Ethernet switch for scale-out. Feynman also features die stacking, a custom HBM, a new GPU, LP40 LPU (the NVLink-native Groq chip), Rosa CPU, BlueField-5, and CX10 SuperNIC. TSMC's process technology underpins the CPO co-development.
Huang addressed the copper-vs.-optics question directly: both will be needed at scale, in growing quantities. AWS simultaneously announced deployment of over one million NVIDIA GPUs spanning Blackwell, Rubin, RTX PRO, and Groq 3 — the largest single AI factory commitment announced at any GTC. NVIDIA Cloud Partners have doubled their AI factory footprint year-over-year, deploying a cumulative 1 million GPUs representing 1.7 gigawatts of AI capacity. The implication for the CPO supply chain is stark: NVIDIA is not hedging its optical bets. Every future generation — Rubin Ultra, Feynman, and beyond — is designed with CPO as a required infrastructure layer, not an optional upgrade. The NVLink8 CPO announcement in particular confirms that silicon photonics is moving from scale-out switching (where it already ships) to scale-up interconnect (the innermost, highest-performance link tier in an AI factory), which is the highest-value CPO application domain.
NVLink8 CPO represents a qualitatively different test challenge from scale-out CPO switches. Scale-up interconnects operate at the lowest latency, highest signal integrity requirements in the data center — any CPO solution at this tier must pass substantially more rigorous electro-optical qualification than a switch CPO. Kyber rack architecture also introduces new co-design requirements between the CPO switch and the rack thermal/mechanical envelope. For assembly and test equipment vendors, the 2027–2028 Rubin Ultra / Feynman ramp is the next major procurement signal — planning now for ISO-7 test cell capacity and electro-optical burn-in at NVLink8 CPO specifications is the correct posture.
Ciena Launches Vesta 200 6.4T CPX — First Product from Nubis Acquisition, Claims Industry's Highest-Density Pluggable CPO
Building on its acquisition of Nubis Communications (closed late 2025), Ciena unveiled the Vesta 200 6.4T CPX at OFC 2026 — the first commercial product to emerge from that $270M deal. The Vesta 200 is positioned as the industry's highest-density, lowest-power pluggable co-packaged optical engine, delivering 6.4T aggregate throughput (32×200G) with up to 70% power reduction versus conventional alternatives. It operates retimer-free via linear-drive, supports an electrical loss budget of up to 20 dB from the host ASIC, and uses an internally developed SiGe driver and TIA co-optimized for temperature stability. The form factor is designed around Samtec's CPX co-packaged copper connector system, achieving chip-edge density compatible with 200G/lane switches, XPUs, and NICs. The product targets scale-up networks (XPU servers) as well as 100T and next-generation 200T scale-out switches.
The significance extends beyond the product itself. Ciena's historic strength is in long-haul coherent transport, not data center optics. The Nubis acquisition — and now this product — represent a deliberate push into the CPO/NPO layer of the data center. Vesta 200 CPX competes directly with Coherent's socketed CPO portfolio and Broadcom's Bailly platform for the growing market for open, pluggable CPO optical engines that avoid a fully bonded, non-field-serviceable architecture. Ciena is also a founding member of both the Open CPX MSA and the XPO MSA, positioning itself as an ecosystem builder rather than a captive supplier. The "pluggable CPO" positioning is deliberate: it offers hyperscalers the power and density benefits of CPO with some serviceability and supply chain flexibility that fully integrated COUPE-style approaches sacrifice.
Pluggable CPO optical engines represent a distinct test category from bonded COUPE-style integration. The Vesta 200's retimer-free linear-drive architecture and socket-based mating require test cells capable of characterizing optical links under chip-edge thermal and mechanical conditions — a different fixture and alignment regime than conventional module test. As pluggable CPO proliferates alongside bonded CPO, test vendors will need to support both form factors, potentially driving demand for modular test cell designs that can be reconfigured between socket-based and bonded-interface testing.
Coherent Demonstrates Full CPO Technology Stack at OFC 2026 — 6.4T SiPho, VCSEL, 400G InP Modulator, and XPO MSA Membership
As NVIDIA's named CPO laser supplier and co-developer for the Spectrum-X platform, Coherent used OFC 2026 to assert multi-technology CPO leadership across three independent architectures simultaneously. The centerpiece demonstration was a 6.4T (32×200G) socketed CPO built on silicon photonics, paired with Coherent's External Laser Source (ELS) module using high-power InP CW lasers — the same basic architecture as NVIDIA's Spectrum-X, but demoed as a Coherent standalone platform. A second demonstration showed a multimode socketed CPO using Coherent's high-speed VCSELs — targeting scale-out scenarios where cost per bit matters more than maximum reach or density. The third demonstration was an InP modulator on silicon operating at 400G per lane, with a 400G-per-lane InP modulator array showing the path toward next-generation CPO lane speeds. Coherent also confirmed membership in the XPO MSA (12.8T+ liquid-cooled optics) and the Open CPX MSA (interoperable near-package optical engines), as well as participation in the 40-member OIF multi-vendor interoperability demonstration. For pluggable portfolios, Coherent demonstrated 1.6T, 3.2T, and XPO-format transceivers spanning silicon photonics, InP EML, and GaAs VCSEL technologies.
The strategic read from Coherent's OFC 2026 presence is that the company is explicitly hedging across all CPO architectures. Its EV Datacenter VP stated that "CPO will be explored in various scale-out and scale-up scenarios by our customers" — language that positions Coherent as a flexible supplier rather than an advocate for any single architecture. This is consistent with its NVIDIA relationship (MRM-based COUPE CPO), its ELS module strategy (feeding any socketed CPO platform that needs an external laser), and its VCSEL investment (for high-volume, lower-density applications). Combined with the $2 billion NVIDIA investment (announced March 2026), Coherent is simultaneously a captive supplier of NVIDIA's most important optical platform and a broadly positioned independent supplier to the rest of the CPO ecosystem.
Coherent's multi-technology portfolio creates a meaningful test infrastructure signal: each modality (SiPho MRM, InP modulator, VCSEL) requires a different characterization suite. InP-on-silicon modulators at 400G/lane are a new test frontier — the optical bandwidth, drive voltage requirements, and thermal sensitivity differ substantially from silicon PN junction MZMs. Equipment vendors building CPO-capable test cells should architect for modular optical head swaps and flexible driver/TIA interfaces to serve all three modalities without full cell replacement.
AIM Photonics and VIAVI Assert US CPO Manufacturing and Test Infrastructure Readiness at OFC 2026
Two companies with direct relevance to assembly and test equipment addressed the manufacturing infrastructure gap at OFC 2026. AIM Photonics — the Albany, NY-based US government-backed photonics manufacturing partnership — presented a session titled "PICs & Packaging for Co-Packaged Optics and Quantum," framing US-based integrated photonics production for both CPO and quantum as a scalable, investable infrastructure. AIM's COO David Harame presented the AIM Photonics Technical Roadmap covering PICs, heterogeneous integration, packaging, and electronic-photonic design automation. AIM also sponsored a session examining the manufacturing and infrastructure considerations shaping scalable US-based integrated photonics production — directly addressing the supply chain localization argument that is increasingly a government procurement requirement in both CPO and quantum.
VIAVI Solutions presented a comprehensive AI-fabric test and measurement showcase at OFC 2026: 1.6T high-speed Ethernet test, silicon photonics manufacturing solutions, PCIe over optics test, automated network test, and fiber sensing. VIAVI stated explicitly: "The evolution of AI, security and high-speed photonics have converged to force a change in validation and optimization strategies. Today, the industry requires a comprehensive and AI fabric-aware view of the network." The One LabPro and TestCenter platforms for OSI Layer 0-3 traffic generation and analysis were demonstrated alongside a new hollow-core fiber test solution and the DCX-700 optical analyzer. VIAVI's framing of "AI-fabric-aware" test — treating the CPO interconnect as part of an integrated AI compute fabric requiring holistic validation rather than component-level optical characterization — represents a positioning shift toward systems-level test that has direct implications for how OSATs and hyperscaler procurement teams evaluate test equipment.
AIM Photonics is the most credible US-based production partner for photonic integrated circuit manufacturing at commercial scale outside of the major captive fabs. For a new entrant into US-based CPO assembly and test, AIM represents both a potential foundry partner (for PIC sourcing and process development) and a reference customer ecosystem for domestic test infrastructure. VIAVI's systems-level "AI-fabric-aware" test positioning is the right framing for a new entrant to adopt when engaging CPO customers: buyers at hyperscalers and OSATs think about CPO validation at the system level, not just at the optical component level.
Taiwan CPO Packaging Supply Chain Accelerates on GTC 2026 Signal — Sunengine Enters CPO Advanced Packaging
GTC 2026's confirmation that NVIDIA's CPO supply chain is scaling rapidly — with thousands of racks per week now being manufactured — triggered an immediate downstream response in Taiwan's electronics manufacturing ecosystem. Digitimes reported that Taiwanese firms are preparing for silicon photonics and CPO packaging opportunities as AI data center demand continues to expand. Specifically, Sunengine Technology entered the CPO advanced packaging supply chain in connection with GTC 2026, representing another Taiwanese manufacturer joining a supply chain that already includes SPIL, Foxconn, and Fabrinet for NVIDIA's CPO module assembly. The Digitimes report framed NVIDIA's dual copper-optical strategy as a deliberate signal that the entire packaging ecosystem — not just the optical engine suppliers — should scale capacity in parallel.
This is directionally consistent with the broader Taiwan CPO infrastructure signal that has been building since GTC 2025: Taiwan's OSAT and advanced packaging industry is increasingly organized around CPO as a distinct product category requiring different fixture, alignment, and test infrastructure from conventional semiconductor packaging. The scale of NVIDIA's deployment commitments — 1M+ GPUs at AWS alone — means that CPO module assembly throughput must scale by orders of magnitude from current levels between now and 2028. That throughput requirement is what makes CPO assembly and test a genuine capital equipment market, not a niche process development exercise.
New entrants into Taiwan's CPO packaging supply chain signal both market validation and competitive pressure. For a US-based assembly and test equipment vendor, the Taiwan CPO ecosystem is simultaneously the primary customer base (SPIL, ASE, Amkor, Foxconn all assemble CPO modules) and the primary competitive environment (Taiwan-based equipment and process vendors have proximity advantages). The strategic response is differentiation on capability — particularly active optical alignment, electro-optical test at CPO module level, and applications engineering support — rather than competing on manufacturing economics where Taiwan-based vendors have structural cost advantages.
Open CPX MSA and XPO MSA Emerge as the Interoperability Standards Layer for Pluggable and Near-Package CPO
Two new multi-source agreements crystallized at OFC 2026 as the interoperability standards infrastructure for the pluggable/near-package optical engine market. The Open CPX MSA (Open Co-Packaging Multi-Source Agreement) is developing specifications for optical engines required to enable a broad, interoperable ecosystem of co-packaged and near-package interconnect solutions. Coherent is a founding member. The XPO MSA is defining a new multi-lane pluggable MSA form factor for 12.8T and beyond — a liquid-cooled optics module delivering 204.8 Tbps per open compute rack unit front panel density. Coherent has also joined this MSA as a founding member, alongside other optical module and system vendors. The significance of both MSAs is that they represent the industry's attempt to prevent CPO from fragmenting into a landscape of incompatible proprietary optical engines — the same pattern that created the pluggable transceiver standards ecosystem (QSFP, OSFP, QSFP-DD) that now underpins a multi-billion-dollar interoperable market.
The timing is critical. NVIDIA's COUPE-based Spectrum-X CPO is now in production, but it is a proprietary system — Lumentum and Coherent supply components under NVIDIA's design authority, not as interchangeable MSA-compliant modules. The Open CPX MSA and XPO MSA are building the standards layer that would allow non-NVIDIA CPO deployments (Broadcom Bailly, Marvell Celestial AI, hyperscaler custom ASICs) to use a common optical engine format. If these MSAs succeed, they would create the conditions for a commodity pluggable CPO optical engine market parallel to the COUPE-bonded proprietary market — two distinct supply chains, each requiring different assembly and test infrastructure.
MSA-compliant optical engines are, by definition, produced by multiple vendors to a common specification — which means they are tested to that specification at both the component vendor and the integrator level. Open CPX MSA compliance testing is an emerging market segment for test equipment vendors: the first company to offer a certified MSA compliance test suite for Open CPX and XPO will be positioned to serve the entire ecosystem of module vendors seeking MSA certification. This is a distinct market opportunity from custom CPO characterization for COUPE-bonded engines.
UK Commits £2 Billion to Quantum Computing Scale-Up — Largest National Quantum Procurement Program Outside US or China
On March 17, 2026, the UK's Department for Science, Innovation and Technology announced a total quantum commitment of approximately £2 billion ($2.67 billion USD): £1 billion specifically for procuring large-scale quantum computers through the new ProQure: Scaling UK Quantum Computing programme, plus over £1 billion over four years for broader quantum research, infrastructure (£90M for industrial infrastructure, £13.8M for five National Quantum Research Hubs), and the Quantum Software Lab in Edinburgh. The ProQure programme — launching in late March 2026 — invites companies to submit state-of-the-art prototype quantum computers for national evaluation, with the most successful systems to be deployed within the national computing infrastructure for researchers and the public sector. Economic projections associated with the strategy estimate £212 billion total impact by 2045. The announcement coincided with three private-sector milestones: Infleqtion delivered an operational 100-qubit quantum computer to the UK's National Quantum Computing Centre; IonQ established a Quantum Innovation Centre at the University of Cambridge, hosting a 256-qubit system; and Vescent selected the UK's National Physical Laboratory for its next international office.
The UK already has more than £1 billion in prior public quantum investment under the National Quantum Technologies Programme launched in 2014. This new package more than doubles that base and adds a critical new mechanism: government procurement as a market-creation lever. Rather than simply funding R&D, ProQure is designed to pull innovation through to commercial scale by providing revenue certainty for hardware vendors willing to deliver prototype and then production-quality systems. The US-UK Memorandum of Understanding on quantum development, signed September 2025, provides additional context: this investment is part of a coordinated allied strategy to maintain Western leadership in quantum hardware manufacturing and computation, and to avoid the supply chain dependencies on non-allied nations that have complicated semiconductor strategy.
ProQure is structurally significant for quantum hardware assembly and test equipment. A government programme that procures prototype and then production-quality quantum computers on a competitive basis creates a defined qualification framework — systems must meet performance benchmarks to advance through evaluation tiers. That specification-driven procurement model is precisely the environment in which specialized assembly and test equipment vendors can establish themselves as infrastructure suppliers: the vendor who instruments, characterizes, and certifies quantum hardware for ProQure evaluations is positioned as a platform supplier to every hardware company competing for UK government contracts. IonQ's Cambridge hub (256 qubits) and Infleqtion's delivered 100-qubit system are the first installations this equipment infrastructure must serve.
White House Draft Quantum Executive Order Would Reset US National Quantum Strategy for First Time Since 2018 — QCSAD at DOE Facility Planned
A draft executive order titled "Ushering In The Next Frontier Of Quantum Innovation" — obtained by Nextgov/FCW and dated February 3, 2026 — represents the most comprehensive US federal quantum policy action since the 2018 National Quantum Initiative Act (which expired in 2023). The draft assigns OSTP a central coordinating role and directs OSTP, Commerce, Energy, and Defense to produce an updated National Quantum Strategy within 180 days of signing. Key provisions include: the creation of a Quantum Computer for Scientific Applications and Discovery (QCSAD) with at least one system housed at a DOE facility; reconstitution of the National Quantum Initiative Advisory Committee (NQIAC, which expired 2023); a Center of Excellence at Energy, Commerce, and Defense to assess quantum computing system capabilities within 180 days; agency-level five-year roadmaps for quantum sensors and quantum networking; and a Commerce Department mandate to "de-risk" investments in commercial quantum companies through co-investment programs, grants, loan guarantees, and government commitments to buy or test early systems.
The order also addresses manufacturing explicitly: coordinating OSTP, ODNI, Commerce, Energy, and Defense to ensure that manufacturing infrastructure, technical expertise, and related capabilities are aligned with national security and commercial objectives. Counterintelligence protections for quantum research are included, consistent with the Trump administration's technology security posture. Post-quantum cryptography notably is not the focus of this order (the March 6, 2026 Cyber Strategy EO addressed PQC separately). Congress is pursuing parallel reauthorization of the NQI Act with a proposed $1.5 billion funding commitment (Sens. Young and Cantwell, early 2026). As of the date of this newsletter, the quantum EO remains a draft and has not been signed.
The DOE-hosted QCSAD is the most concrete manufacturing infrastructure signal in the draft order. A government-procured scientific quantum computer at a DOE facility — operated at scale for researchers — would require a full supply chain: trap chip or photonic chip fabrication, cryogenic packaging, qubit control electronics, dilution refrigerator integration, and characterization and test at system level. The "de-risk" mandate for Commerce creates the mechanism for government co-investment with companies building quantum hardware manufacturing infrastructure — a potential funding path for test equipment vendors serving the quantum hardware supply chain. Monitor for the signed version; the 180-day clocks on all major provisions start at signing.
DOE / Fermilab + MIT Lincoln Lab Demonstrate In-Vacuum Cryoelectronics Controlling Ion Traps — Key Step Toward Scalable Trapped-Ion Systems
Researchers at Fermi National Accelerator Laboratory and MIT Lincoln Laboratory, working under a collaboration between two DOE National Quantum Information Science Research Centers (Quantum Science Center at Oak Ridge and Quantum Systems Accelerator at Berkeley Lab), successfully demonstrated the use of in-vacuum cryoelectronics to control ion traps. The experiment trapped and manipulated ions using Fermilab-developed cryogenic circuits operating at the extreme cold temperatures required for quantum computers, integrated with MIT Lincoln Laboratory's ion-trap platform. The cryoelectronics performed key control functions — moving individual ions, holding them at set positions, and measuring electronic noise effects — within the vacuum environment. This proof-of-principle eliminates the conventional "wiring bottleneck" in trapped-ion systems: current large-scale ion trap designs require thousands of wires running from room-temperature control electronics into the vacuum chamber, which becomes untenable as qubit counts scale toward hundreds or thousands. Integrating cryoelectronic control directly into the vacuum eliminates most of that wiring, replacing it with a few high-bandwidth connections to the cryogenic control chip.
The relevance to the broader quantum hardware trajectory is significant. IonQ (with its Oxford Ionics acquisition), Quantinuum (H-Series QCCD), and others have all identified wiring density as a key barrier to scaling trapped-ion systems. This demonstration — funded by DOE through two national quantum centers — validates the cryoelectronics integration path at the laboratory level. The next step is integration of this approach with higher-qubit-count systems and demonstration of multiple rounds of quantum operations with cryogenic control. MIT Lincoln Laboratory's ion-trap platform is one of the most mature in the US national lab system, providing a credible hardware substrate for this advancement.
Cryoelectronics co-integrated with ion traps creates a new compound manufacturing challenge: the control chip (operating at millikelvin or a few kelvin) must be fabricated, tested, and packaged alongside the ion trap chip within the vacuum envelope. This is a hybrid packaging problem — cryogenic CMOS chiplet integrated with a precision microfabricated ion trap substrate — that is analogous in some ways to the EIC/PIC co-packaging challenge in CPO, but at orders-of-magnitude more extreme temperature and cleanliness requirements. For assembly and test equipment vendors, this represents the earliest signal of what the next generation of trapped-ion manufacturing infrastructure will require: cryo-compatible flip-chip bonding, in-vacuum electrical characterization at operating temperature, and multi-chip packaging at cryogenic tolerances.