Internet Flashback, February 1998: China’s 1 Phone Line via Sprint in Los Angeles to Global Peering with Digital Island

At the time, China accounted for 18% of the world’s population, 1.3 billion people, but had only one telephone line for internet access and intranet networking throughout the entire country.

To connect China to our network in 1998, I contracted for a submarine and terrestrial cross-continental circuit, which cost my cost center $960,000 per year.

The initial contract terms were $275,000 prepaid upfront. This included the cost of a one-time installation charge of $35,000, and the monthly recurring circuit and port fees of $80,000 per month, and I paid three months in advance ($240,000).

Note, China did not contribute to any of these charges. My cost center was paying the entirety of these expenses.

CERNET (China Education and Research Network), anchored by Tsinghua University, was a single 64kb phone line intranet linking 300 universities across China, with a matching 64kb port to SprintLink in Los Angeles. One phone line of capacity was all that was available to serve 18% of the world’s population.

SprintLink, a U.S.-centric backbone, gave CERNET basic internet access, a one-way trickle (not peering), reaching less than 1% of the world’s internet users via a network one-hop, thus capping their global ambitions; unlike the Digital Island peering I drove which delivered a mutual, high-capacity exchange, tying the world’s largest ISPs and 95% of online users into a seamless global network.

In February 1998, Professor Xing Li of Tsinghua University and I teamed up during my Beijing visit with his crew to craft a peering plan, a meeting where their vision synced with our tech to expand CERNET’s reach beyond SprintLink’s limit. We pitched peering to thrust China into the internet’s core, leveraging Digital Island’s unique infrastructure, the first autonomous global WAN we’d built spanning six continents and linking all major ISPs. Shortly after, we fired up a $80,000-per-month T-1 circuit (1.544 Mbps and 24x faster) and dropped our CDN gear in Beijing within their network, peering Digital Island with CERNET, China’s first internet peering agreement.

Unlike SprintLink’s weak link, our peering unleashed bidirectional traffic and rich content: Tsinghua’s 300-university network hit 24x faster speeds for millions, tapped 18% of the world’s population, 1.3 billion people, via one DNS flip, and joined our global grid, six continents, all major ISPs, first CDN, 19 data centers. CERNET, guided by Tsinghua, spun this into eCommerce, juiced research with global data, and stood equal with Tier 1s for the first time, boosting Digital Island’s CDN traffic and cementing our cooperation in China’s growing market, driven by our customers at Stanford, Cisco, Microsoft, Intel, Compaq, HP, Visa, Mastercard, Charles Schwab, and E*TRADE, all drawn to CERNET’s capability and potential.

This cracked open a global gateway, paving the way for 5.6 billion users (2025, ITU) and $10T in eCommerce (2023, Statista). All paid for by Cliff Higgerson, the VC who didn’t invent the internet, he just paid for the Lion’s Share of it be useful. See his web page here https://marknichols.com/cliffhiggerson/

China’s 1994 SprintLink Connection Details

Based on the historical records from the 1990s (including Sprint’s own announcements, network architecture overviews, and timelines of China’s internet rollout), the 1994 CERNET connection to SprintLink was indeed a 64kbps international leased line using frame relay protocols, and once it hit Sprint’s U.S. backbone, it encountered more frame relay for transit and transport. This setup was far from true global peering; it was essentially a low-bandwidth, shared-circuit ISP access point with all the limitations of frame relay (variable latency, no inherent QoS, packet loss under load), making it limited for anything beyond basic email or text-based access, and definitely not viable for SSL-secured global e-commerce.

• The Link Itself: China’s first full TCP/IP internet connection (making it the 77th country globally) was established on April 20, 1994, via a 64kbps dedicated international line from CERNET (China Education and Research Network) in Beijing to SprintLink in Stockton, California. This was a leased circuit provided by Sprint (not owned nor operated by China), but it operated over frame relay protocols for packet switching. CERNET’s topology at the time formed a multi-ring structure domestically, but the international hop to SprintLink was the bottleneck, 64kbps was equivalent to a single voice-grade phone line, shared among China’s early academic users (e.g., Tsinghua University).
• No Global Peering: This wasn’t a peering agreement in the modern sense (equal exchange of traffic between autonomous networks). It was a one-way customer-provider relationship: CERNET paid Sprint for transit access to the broader internet. Traffic from China would enter SprintLink’s network and then be routed onward, but without dedicated, end-to-end control or optimized paths to global metros. No bidirectional, settlement-free peering with multiple Tier 1s worldwide, just basic upstream access to Sprint’s U.S.-centric backbone.

Sprint’s Network in the 1990s: Frame Relay “Wall” for Transit/Transport

• Frame Relay Dominance: Sprint heavily promoted and deployed frame relay as their core technology for data transmission starting in the late 1980s/early 1990s. In 1990, they announced frame relay services that bypassed traditional error-checking intervals, routing data directly through switches, but this was still shared, variable-rate packet switching without guaranteed bandwidth or low latency. By the mid-1990s, SprintLink (their IP service) integrated frame relay extensively for access, transit, and backbone transport, especially for enterprise and international links. This created a “frame relay wall”: Once traffic hit SprintLink (e.g., from China’s 64kbps line), it would traverse Sprint’s internal frame relay switches and multiplexers for onward routing, introducing jitter, delays, and drops, perfect for low-priority data but not intended nor capable for supporting real-time apps.
• Why “Garbage” for E-Commerce: Frame relay’s lack of QoS meant international hops (e.g., from China to Europe via U.S. transit) could see 500-1000ms+ latency with frequent packet reordering/loss, SSL handshakes (requiring stable TCP) would fail often, and e-commerce transactions (needing <300ms for usability/security) were impossible at scale. Sprint’s backbone mixed frame relay with X.25/ATM in places, but no end-to-end IPLCs for global autonomy. This was “ISP access” for academics, not a peering fabric for commercial globalization.

The Digital Island network in 1997 bypassed this entirely with autonomous IPLCs (dedicated, point-to-point fiber) for end-to-end control, enabling the first reliable global peering and SSL support for e-commerce. No other setup matched that, no frame relay walls, just seamless and autonomous transit.