China’s 1,007 km Hollow-Core Fiber Test: 64 Tb/s Each Way and the Path to Commercial Networks

For enterprise infrastructure buyers, the most useful comparison is not a national leaderboard. Microsoft has already described hollow-core deployment in Azure and a manufacturing expansion involving Corning and Heraeus. Together, these developments put the spotlight on the less glamorous work of industrialisation: delivering qualified cable, installing it consistently and restoring service when something breaks.

What China’s distance test actually demonstrated​

The Wuhan East Lake High-Tech Development Zone’s November 9, 2026, account describes an extended C+L-band transmission system operating at 800 Gb/s per wavelength, using PCS-16QAM modulation. Its headline capacity is 128 Tb/s combined across both directions. That is not 128 Tb/s available to traffic travelling one way.

The distinction matters because capacity belongs to the complete optical system—not simply to the fibre. The reported result combines the transmission medium with wavelength allocation, modulation and supporting equipment. It should therefore be read as a demonstrated configuration, not a universal bandwidth rating for hollow-core cable.

These boundaries follow from the configuration described in the Wuhan account and the transmission-verification description carried by the Chinese Embassy in Laos. Neither short account supplies a complete technical acceptance report.

The embassy’s account also repeats a claim that the technology doubles capacity and extends transmission distance by 60% compared with the best-performing solid-core alternatives. That comparison needs to remain attributed: the published summary does not identify a matched reference system or enough methodology to establish a general ranking.

The takeaway: the combined result deserves attention without being stretched into claims the published evidence cannot support.

Why hollow-core fibre can reduce delay​

Hollow-core fibre guides most of the optical signal through a hollow region rather than a solid silica core. A technical account published on Beijing’s international science and technology innovation platform says the Chinese team used high-purity argon to address absorption associated with residual gases. It also reports link attenuation below 0.1 dB/km at 1,550 nanometres. These remain attributed project disclosures, rather than independently reproduced measurements.

The physical attraction is straightforward: signals can propagate faster through the hollow core. Microsoft separately describes its design as providing up to 47% faster transmission and approximately 33% lower latency than conventional single-mode fibre. Those percentages describe related effects, not two benefits that can be added together.

For illustration, increasing propagation speed by 47% reduces travel time over a fixed distance to approximately 68% of its previous value—a reduction of about 32%. That is a reciprocal calculation, not an additional performance claim.

China’s reported 3.43 ms also needs a defined measurement boundary before it can be compared with a cloud request or financial transaction. The public summaries do not specify precisely where timing started and stopped, which processing stages were included, or the measurement uncertainty. A fast optical path is not automatically an equally fast application.

Azure shows why manufacturing matters​

Microsoft’s September 23, 2025, announcement provides an important commercial comparison. The company said it had deployed hollow-core fibre across multiple Azure regions since 2023 and was expanding production through collaborations with Corning and Heraeus. Its engineers were working with the manufacturers on process implementation, training, yield, metrology and reliability.

That list is more revealing than another superlative. Manufacturing yield determines how much output becomes usable product; qualification determines whether that product can become accepted infrastructure. Microsoft’s own programme makes clear that scaling requires more than reproducing an impressive optical result.

There is also a supply-chain distinction worth preserving. Microsoft described production through Corning’s US facilities and Heraeus sites in Europe and the United States, while operationalising Microsoft manufacturing-process intellectual property. Multiple manufacturing locations do not necessarily mean independent technology alternatives. That is a procurement inference from the disclosed arrangement, not evidence that the arrangement is inherently insecure.

Other industrial evidence addresses different parts of deployment. Prysmian’s July 2026 announcement reported installation tests with Relativity Networks and Dura-Line, plus manufacture of a 24-hollow-core-fibre cable measuring 10 mm in diameter at its Claremont, North Carolina, plant. Those are useful cable and installation milestones—not a complete lifecycle-cost comparison.

AI and financial networks need workload evidence​

Chinese reporting identifies data-centre interconnection and financial connectivity as deployment opportunities, including a reported commercial cross-border financial connection between Shenzhen and Hong Kong. It also combines commercial and pilot installations in a total exceeding ten projects. That mixed count should not be mistaken for ten equivalent, mature production services.

For buyers, the recommended test is narrower: does improving this particular connection improve the outcome being purchased?

Consider an illustrative AI workload in which exposed waiting on the affected link accounts for 10% of completion time. Cutting that component by 30%, with everything else unchanged, reduces total completion time by only 3%. This is arithmetic under explicit assumptions—not an observed AI benchmark.

A sensible evaluation should therefore measure:

• Completion time for an unchanged workload.
• Useful throughput within the required response-time limit.
• Whether communication savings remain visible after other bottlenecks.
• Whether the same benefit could be obtained more economically elsewhere.
• Performance when traffic moves to the alternative route.

For financial connectivity, the equivalent discipline is to measure the relevant transaction path and customer value rather than infer revenue from a transmission headline.

National choices through November 2031​

The next five years are better treated as a qualification programme than an adoption forecast. The following are proposed priorities, not claims that governments have adopted hollow-core deployment mandates.

The regulatory backdrop reinforces these questions without answering the technology choice. The European Commission describes NIS2 as a cybersecurity framework that includes supply-chain security. The UK’s telecommunications security framework likewise addresses network security and supplier dependencies; its revised code discusses procurement and total ownership costs. These are reasons to examine control and continuity—not evidence of a requirement to buy hollow-core fibre.

Buy the service, not the headline​

Before committing a critical route, buyers should request a complete configuration, defined timing endpoints, channel-loading and error results, sustained observation records, installed-link acceptance criteria, repair procedures and a full cost comparison. These are proposed procurement requirements designed to close the gaps between a reported demonstration and a dependable service.

The strongest evidence of progress through November 2031 would be repeat orders supported by sustained field operation, predictable restoration and measurable application gains—not simply another capacity record.

China’s distance test advances the engineering argument. Azure’s production programme and the wider cable ecosystem show why the industrial argument is separate. The eventual winner need not own the loudest announcement; it needs to deliver a connection that customers can justify, operators can maintain and procurement teams can replace when necessary.