DE CIX, which claims to be the world’s largest Internet, cloud, and AI exchange operator, surveyed 400 of the largest IT and telecom infrastructures in the U.S. and the U.K. and reached some troubling conclusions. While 96% of those surveyed felt comfortable about the reliability of their own network, the average response was that network operators expect to spend an average of 11.5 hours per week over the next two years resolving network connectivity issues. That’s more than a workday every week diagnosing connectivity issues. One-tenth of respondents spend between 20 and 41 hours per week dealing with connectivity issues.

Just under half of the IT teams surveyed already spend more than ten hours per week troubleshooting cloud connectivity issues. The most typical problems reported include downtime or reliability issues (26%), latency or slow performance (28%), and security vulnerabilities/DDoS attacks (27%).

When asked to list their biggest concerns over the next year, the IT teams listed: latency (28%); security vulnerabilities (27%); downtime and reliability (26%); cost of connectivity (24%); data sovereignty (23%); staff expertise (22%); and visibility into data flows (20%).

The cost of network outages can be staggering. According to a study released in July by Gartner and the Ponemon Institute/Emerson, the losses for an outage for various industries is:

• Enterprise & Data Centers: $300,000 to $540,000 per hour (with major incidents exceeding $1 million).
• Finance & Banking: $1 million to $5 million+ per hour due to high-frequency trading and continuous transactions.
• Auto Manufacturing & Industrial IoT: 2.3 million per hour due to assembly line freezes.
• Retail & E-Commerce: $1 million to $2 million per hour in direct lost checkout revenue.
• Healthcare & Hospitals: $474,000 to over $3.2 million per hour.
• Small-to-Medium Businesses: $8,000 to $25,000 per hour.

DE CIX says these findings are particularly troubling in a world where many businesses and applications are increasingly reliant on AI, which needs low-latency uninterrupted connections. It seems likely that as AI becomes more prevalent, the impact of outages will be more widespread and cost more than estimated above.

The big Internet network carriers all brag about their reliability, but in a world where large multi-hour outages are becoming routine, there is nobody around who is delivering 5 nines (99.999%) reliability – the long-time standard for a high-quality network connection. I still see backbone contracts that guarantee major uptime, but every such contract now comes with wiggle room language that says that the carrier selling you a circuit is not responsible for outages caused by somebody else.

I’ve written before about the increase in big network outages. Verizon has already had two major outages this year, one in January and one in August. We’ve seen the negative results of major outages. Airports shut down. Numerous businesses stop functioning.

As good as our backbone networks are, they are complex, and network connections must pass through multiple nodes to cross the country. I think it would be ironic if the biggest Achilles’ heel of AI turns out to be the long-haul fiber networks and the Internet hub system in the country.