Microsoft’s move to outsource hollow‑core fiber (HCF) production to established glass and fiber manufacturers marks a key inflection point for a technology that has long promised dramatic gains in latency, bandwidth and energy efficiency for cloud and AI workloads. Over the last week Microsoft publicly announced strategic manufacturing collaborations with Corning and Heraeus to scale its HCF output, building on the company’s 2022 acquisition of Lumenisity and a string of laboratory breakthroughs that have pushed HCF attenuation below the historic limits of silica fiber. These partnerships are intended to accelerate the industrialization of Microsoft’s proprietary Double Nested Antiresonant Nodeless Fiber (DNANF) design and to seed a multinational supply chain capable of supplying Azure’s global network needs.
Microsoft and its research partners have been explicit about the potential gains: the company’s published materials and technical briefings point to roughly 47% faster effective light propagation through HCF compared to standard silica glass and up to about a 30–33% reduction in propagation latency per kilometer, depending on the specific fiber geometry and operating wavelength window. Those headline numbers have been cited in Microsoft’s public blogs and partner statements.
Most recently, Microsoft confirmed it will partner with Corning to produce HCF at Corning’s U.S. facilities, and with Heraeus Covantics to produce HCF at European and U.S. sites. Those partnerships are explicitly aimed at scaling manufacturing capacity and building a resilient, global supply chain for Azure’s HCF requirements. Microsoft says Azure engineers will work alongside Corning and Heraeus to transfer manufacturing IP, implement training, and drive the yield and metrology improvements required for larger‑scale production.
Independent science and engineering outlets noted not only the low loss but also significant reductions in chromatic dispersion—claims that, if borne out across production lengths, could ease transceiver DSP complexity and power draw in high‑capacity systems. The DNANF family uses concentric nested thin glass membranes (the “double‑nested” elements) to produce anti‑resonant reflection that confines light in the air core while suppressing leakage and higher‑order modes. Precise wall thickness, capillary geometry and contamination control are all critical to achieving the lab numbers.
Corning will leverage existing fiber and cable facilities (U.S. operations were highlighted), while Heraeus will supply preforms, fused silica raw materials and produce fiber at European and U.S. sites—building out a multinational manufacturing footprint intended to serve Azure’s global production needs. Microsoft frames these relationships as IP‑protected process transfers where Azure engineers will work on yield improvement and quality controls with partner operations.
At the same time, there are genuine technical and commercial obstacles. Achieving lab‑grade attenuation over many thousands of kilometers requires mastering contamination, geometry tolerances and cableization practices that are new to the industry. The economics must also improve before HCF will displace commodity SMF in general purpose networks. Microsoft’s approach—retaining R&D leadership while outsourcing production to proven manufacturers—mitigates manufacturing risk but does not eliminate it. The industry will watch yield curves, cost per km, field performance and industry standards activity closely over the next several quarters.
For data center and network architects, the prudent posture is to plan for HCF as a strategic option for latency‑sensitive and high‑capacity links, while continuing to rely on SMF as the workhorse medium for general routing. The Microsoft‑Corning‑Heraeus axis accelerates the timeline by years compared with single‑vendor ramp approaches, but mainstream renewal of the global fiber plant will be incremental, measured and driven by cost, operational maturity and standards convergence.
Microsoft’s latest move signals that HCF has left purely academic territory and is now a materials‑ and supply‑chain problem being solved at scale—an encouraging development for cloud operators and anyone designing networks for an increasingly latency‑hungry AI era. The next meaningful milestones to watch: production yield improvements reported by Corning and Heraeus, independent validation of long‑length attenuation on production reels, and the emergence of industry standards for HCF interconnection and testing. If those checkboxes are met, HCF could rewrite the economics of latency‑sensitive connectivity; if not, the technology will find narrower—but still valuable—use cases where its unique properties matter most.
Source: Data Center Dynamics Microsoft ramps up hollow core fiber production with Corning, Heraeus partnerships
Background
What is hollow‑core fiber (HCF) and why it matters
Hollow‑core fiber routes light through an air‑filled central channel surrounded by a microstructured ring of thin glass tubes. By guiding light in air rather than in solid silica, HCF reduces the effective refractive index and therefore the propagation delay, while also dramatically reducing nonlinear effects and chromatic dispersion that limit conventional single‑mode fiber (SMF). In practical terms, the result is lower latency per kilometer, higher permissible launch power, and—when the losses come down—longer unamplified spans and a wider usable optical spectrum.Microsoft and its research partners have been explicit about the potential gains: the company’s published materials and technical briefings point to roughly 47% faster effective light propagation through HCF compared to standard silica glass and up to about a 30–33% reduction in propagation latency per kilometer, depending on the specific fiber geometry and operating wavelength window. Those headline numbers have been cited in Microsoft’s public blogs and partner statements.
Microsoft’s HCF strategy so far
Microsoft acquired the UK‑based HCF specialist Lumenisity in December 2022, inheriting the company’s NANF‑derived products and a new 40,000 sq ft manufacturing plant in Romsey, UK. Since the acquisition Microsoft has invested in research, production upgrades and test deployments across Azure regions, turning a lab curiosity into a fieldable asset. The firm has publicly documented test deployments and pilot links and has described operational rollouts in multiple regions since 2023.Most recently, Microsoft confirmed it will partner with Corning to produce HCF at Corning’s U.S. facilities, and with Heraeus Covantics to produce HCF at European and U.S. sites. Those partnerships are explicitly aimed at scaling manufacturing capacity and building a resilient, global supply chain for Azure’s HCF requirements. Microsoft says Azure engineers will work alongside Corning and Heraeus to transfer manufacturing IP, implement training, and drive the yield and metrology improvements required for larger‑scale production.
The technical leap: DNANF and record‑low loss
The breakthrough in context
A string of 2024–2025 research milestones culminated in an industry‑level breakthrough: a DNANF design demonstrated attenuation below the long‑standing ~0.14 dB/km floor associated with the best silica fibers. Laboratory and test‑bed work by teams that include University of Southampton researchers and Microsoft/Azure fiber researchers reported attenuation as low as 0.091 dB/km at 1,550 nm, with sustained sub‑0.2 dB/km performance across exceptionally wide spectral bands and long test spools. That performance, if reproducible in long reels and commercial production, flips the conventional wisdom that hollow‑core fibers must accept worse loss in exchange for lower latency.Independent science and engineering outlets noted not only the low loss but also significant reductions in chromatic dispersion—claims that, if borne out across production lengths, could ease transceiver DSP complexity and power draw in high‑capacity systems. The DNANF family uses concentric nested thin glass membranes (the “double‑nested” elements) to produce anti‑resonant reflection that confines light in the air core while suppressing leakage and higher‑order modes. Precise wall thickness, capillary geometry and contamination control are all critical to achieving the lab numbers.
What the numbers mean in practice
- 0.091 dB/km at 1,550 nm: If reproducible at scale, this loss figure is lower than the historic minimum for silica SMF and implies longer unamplified spans, fewer repeaters or amplifiers, and reduced energy consumption for long‑haul links.
- ~47% faster propagation: This is not a magic doubling of throughput, but a reduction in propagation delay per kilometer arising from the lower effective refractive index of air versus glass—the key advantage for latency‑sensitive workloads like financial trading and real‑time AI inference.
- Broad spectral window: DNANF designs report sub‑0.2 dB/km across tens of THz, opening the possibility of using wavelengths beyond standard telecom windows for capacity expansion.
Manufacturing and supply‑chain realities
Why Microsoft is outsourcing production
HCF’s manufacturing is materially different from conventional silica fiber draws. DNANF geometries need extremely tight control of capillary thicknesses and nested tube concentricity at sub‑micron tolerances. Preform and tube fabrication, precision drawing, gas handling to limit core contamination, and new metrology to verify uniformity over kilometers all demand specialized equipment and process know‑how. Microsoft’s choice to partner with established fiber industry leaders—Corning, a global leader in glass and optical communications, and Heraeus Covantics, a preform and high‑purity fused silica provider—reflects the recognition that scaling HCF will require industrial‑grade manufacturing capability beyond what a single startup or in‑house facility can economically deliver.Corning will leverage existing fiber and cable facilities (U.S. operations were highlighted), while Heraeus will supply preforms, fused silica raw materials and produce fiber at European and U.S. sites—building out a multinational manufacturing footprint intended to serve Azure’s global production needs. Microsoft frames these relationships as IP‑protected process transfers where Azure engineers will work on yield improvement and quality controls with partner operations.
The manufacturing hurdles
The DNANF structure is precise and unforgiving:- Extremely tight wall‑thickness tolerances on nested capillaries.
- Strict control on gas composition inside the core and the presence of micro‑contaminants.
- Metrology that can verify sub‑micron geometry across kilometer‑length reels.
- Cableization and cabling jacketing methods that preserve HCF geometry and low bending loss in the field.
Deployment, interoperability and the use cases that matter
Early deployments and corporate targets
Microsoft has said it has been deploying HCF in Azure regions since 2023 and public reporting indicates pilot and production links have been installed. Press coverage cites that pilot programs and installed runs exceed hundreds of kilometers; some outlets report Microsoft installations totaling around 1,200 km of live HCF with corporate presentations and executive comments referencing a target to deploy up to 15,000 km across the Azure network in the coming years. Those figures appear in multiple trade reports and company talks, but the 15,000 km number should be treated as a corporate deployment target rather than independently audited route‑by‑route measurements.Interoperability with SMF
Practical network operations will almost certainly use HCF alongside conventional SMF for the foreseeable future. Splicing and interconnection strategies already exist—Lumenisity and Microsoft engineering have demonstrated techniques to fuse HCF to SMF and to use hybrid links—but full replacement of SMF is neither necessary nor likely in the medium term. HCF is most valuable where propagation delay or very wide bandwidth per fiber is a gating constraint: cross‑DC links, latency‑sensitive metro routes, AI fabric interconnects, and select transoceanic or long‑haul spans where amplifier count is critical.Primary use cases
- AI and real‑time inference fabrics: Lower propagation delay improves end‑to‑end inference latency for distributed AI models.
- Financial trading and high‑frequency markets: Microseconds saved per kilometer translate directly into business value.
- Hyperscaler backbone links and long‑haul routing: Lower loss and broader spectrum could reduce repeaters and increase capacity per fiber pair.
- Specialized sensing, laser delivery and quantum links: The air core and broader spectrum unlock niche scientific and defense applications.
Competitive landscape and industry momentum
Microsoft is not alone in betting on HCF. Other startups and incumbents are moving: cable and fiber manufacturers (e.g., Prysmian), new entrants like Relativity Networks, and materials specialists are forming partnerships to scale HCF production and cabling for hyperscalers and telcos. These parallel efforts illustrate a broader industry recognition that HCF could be a foundational technology for next‑generation networks if production and cost hurdles are solved. The new Corning and Heraeus collaborations put Microsoft in a better position to influence standards and interoperability because they embed HCF manufacturing capacity inside companies that already supply global carriers and data centers.Risks, unknowns and open engineering questions
No technology is without caveats, and HCF’s path from lab to ubiquitous network layer is littered with concrete engineering and commercial risks.- Scaling laboratory results to production reels: DNANF low‑loss metrics were demonstrated in controlled settings and pilot‑scale runs. Producing thousands of kilometers with the same attenuation across many spools demands extraordinary process control and may reveal new failure modes (microbending sensitivity, contamination during cableization, reel‑to‑reel uniformity). The industry has noted that manufacturing impurity control was a key enabler for the low‑loss window.
- Cost per kilometer and economics: Until HCF reaches comparable per‑km costs to SMF (including installation, amplifier savings and lifecycle maintenance), carriers will use it selectively. Economies of scale and manufacturing yield improvements are required to make HCF broadly price‑competitive for commodity routes.
- Field robustness and cabling practices: HCF’s nested thin‑wall structure is mechanically different from monolithic glass core fiber. Cable jacketing, buffer tubes, ducts and handling procedures must be optimized to prevent microbending and preserve mode confinement. Existing fiber installers and splice teams will need new training and tools.
- Standards and interoperability: For widespread adoption, standards bodies and industry consortia will need to define mechanical, optical and connectorization norms for HCF to ensure multi‑vendor compatibility. Microsoft’s partnerships position it to shape de‑facto standards, but the broader industry will push for open specifications.
- Verification of deployment claims: Public figures such as “15,000 km planned” are corporate targets; independent confirmation (route‑level, audit or regulatory filings) is often absent in early stages. Analysts recommend treating such numbers as indicative of intent rather than completed installations until carrier filings or public route maps back them.
What this means for data centers and enterprise networking
For data center operators and enterprise network architects, HCF presents both an opportunity and a set of design choices:- Opportunities: Lower latency links between critical clusters, the potential to reduce active amplification in long‑haul links, and expanded usable optical spectrum for higher per‑fiber capacity. For operators building AI fabrics that are latency‑sensitive, strategic HCF insertion could materially improve service quality and lower power per bit.
- Practical considerations: Mixed HCF/SMF topologies will require careful fiber plant audits, new splicing best practices, and updated maintenance playbooks. Enterprises should include HCF options in their medium‑ and long‑term network roadmaps but avoid wholesale replacement of proven SMF plants until cost and operational maturity are demonstrable.
Roadmap to mainstream adoption: realistic timelines and milestones
- Short term (12–24 months): Pilot scale‑ups, yield improvement cycles with Corning and Heraeus, limited regional deployments for latency‑critical routes. Expect vendor‑driven hybrid HCF/SMF offerings and specialized products for hyperscalers and financial services.
- Medium term (2–5 years): Commercialization of improved DNANF draws at higher yields, gradual price reductions, emergence of HCF‑aware transceivers and connectors, and initial standards work. Wider commercial links in metro and inter‑DC backbones become practical as costs decline.
- Long term (5+ years): If manufacturing scale and cost converge with SMF economics, HCF could become a mainstream option for latency‑sensitive and high‑capacity routes. Even then, SMF will remain dominant for the bulk of mass‑market fiber due to entrenched infrastructure and cost advantages unless new HCF manufacturing paradigms dramatically reduce unit costs.
Final analysis: transformative, but not an immediate replacement for SMF
Microsoft’s partnerships with Corning and Heraeus are a clear, pragmatic step toward industrializing HCF. By pairing the DNANF research and the Romsey pilot plant with global glass and fiber manufacturing leaders, Microsoft is addressing the hardest part of any materials and photonics breakthrough: volume production with reliable yields. If Corning and Heraeus can replicate laboratory DNANF performance in production reels at acceptable cost, HCF will shift from niche use to a mainstream tool for latency‑critical and high‑capacity network segments.At the same time, there are genuine technical and commercial obstacles. Achieving lab‑grade attenuation over many thousands of kilometers requires mastering contamination, geometry tolerances and cableization practices that are new to the industry. The economics must also improve before HCF will displace commodity SMF in general purpose networks. Microsoft’s approach—retaining R&D leadership while outsourcing production to proven manufacturers—mitigates manufacturing risk but does not eliminate it. The industry will watch yield curves, cost per km, field performance and industry standards activity closely over the next several quarters.
For data center and network architects, the prudent posture is to plan for HCF as a strategic option for latency‑sensitive and high‑capacity links, while continuing to rely on SMF as the workhorse medium for general routing. The Microsoft‑Corning‑Heraeus axis accelerates the timeline by years compared with single‑vendor ramp approaches, but mainstream renewal of the global fiber plant will be incremental, measured and driven by cost, operational maturity and standards convergence.
Microsoft’s latest move signals that HCF has left purely academic territory and is now a materials‑ and supply‑chain problem being solved at scale—an encouraging development for cloud operators and anyone designing networks for an increasingly latency‑hungry AI era. The next meaningful milestones to watch: production yield improvements reported by Corning and Heraeus, independent validation of long‑length attenuation on production reels, and the emergence of industry standards for HCF interconnection and testing. If those checkboxes are met, HCF could rewrite the economics of latency‑sensitive connectivity; if not, the technology will find narrower—but still valuable—use cases where its unique properties matter most.
Source: Data Center Dynamics Microsoft ramps up hollow core fiber production with Corning, Heraeus partnerships
