Bandwidth is the lifeblood of the internet’s arteries. Though hidden from the eye, an ever-growing network of very high-capacity subsea fibre-optic cables forms the essential conduit for linking billions of internet users with each other and their desired content hosted by Meta, Netflix et al. Bandwidth on these cables routes over 99 per cent of internet traffic; new satellites such as SpaceX, play a minor role.
Until the last decade, most of this cable bandwidth was owned by groups of publicly regulated telephone companies and leased to businesses, including internet providers, much like a utility service. Today, however, TeleGeography, which charts subsea bandwidth, estimates that over 75 per cent of ‘lit’ (i.e. active as opposed to potential) cable capacity is privately owned by just four big tech companies (competition policy notwithstanding) which use it for cloud platform and content delivery services.
The global take-up of generative AI services, such as ChatGPT and Gemini, has only led big tech to double down on bandwidth. A new cable construction boom is underway to connect AI-centric data centres with billions of expected AI users across the world.
The bandwidth of new cables, such as Meta’s planned 31,000 mile long globe-circling Waterworth project, is staggering. With a design capacity of 480 trillion bits per second – enough for 4.3 billion simultaneous Zoom calls or perhaps 10 billion competing AI prompts – it neatly complements the trillion dollar ‘shock and awe’ investments announced by AI companies.
How did we get here?
To date, most countries (including, notably, the EU’s coastal states, but excluding China, which is a special case) have granted big tech the necessary licences and permits to install their private fibre-optic cables across nationally owned coastal waters in return for fairly modest fees and limited national security conditions (e.g. mandating police access to cable landing stations; requiring in-country storage of local user data). A liberal regime of reciprocal landing rights was an article of faith in the clubby old world of telco cables.
Should the owner of a private subsea cable have any responsibility to limit the use of AI bandwidth for services which fail local public safety standards?
But that world is now dead. The ‘digital conquistadors’ of big tech, to borrow from Giuliano da Empoli,1 have birthed a new international empire, one that has scant respect for the laws and rules of the past age. That empire, as noted earlier, was delivered by big tech’s build-out of an astonishing, vertically integrated and continent-spanning digital infrastructure knit together by trans-oceanic cables.
Facing this unprecedented (and lighting quick, by telco standards) campaign, most authorities in the old world of cable permitting have lowered the drawbridge. Landing rights have been quietly granted (sometimes in concert with a local partner) to open overseas markets to tech’s new conquerors.
And yet, what has been surrendered at the shoreline has often sparked resistance in the metropolis as higher authorities grapple with the consequences of the landing parties’ sustained digital campaigns. Notably, the cable-based export of US social media by big tech has led to a steady string of legal actions by various authorities in the EU and Pacific Basin to combat pressing local concerns with online child safety, harmful content and cybercrime.
Likewise, the efforts of big tech to curb competition in the markets for internet searches, advertising products and other services has been challenged in the courts by various authorities in the EU and also, of course, in the US. More recently, the training of generative AI products has triggered a spate of additional anti-piracy cases against big tech companies and their partners in various national venues.
Beyond that, the delivery of AI models, which rely on massive clusters of chips, is being challenged at the local level as the construction of large new data centres upends traditional land use polices and drives up electricity costs. Notably, Ireland now has a de facto moratorium on grid connections for new datacentres due to the strain existing facilities have placed on electricity supply.
What happens next?
Let’s start with some obvious questions: should a company which has abused its market position or pirated other people’s works or hurt children online be granted additional bandwidth to deliver its services? If so, on what terms?
Along similar lines, should the owner of a private subsea cable have any responsibility to limit the use of AI bandwidth for services which fail local public safety standards (e.g. by lacking proven guardrails or withholding training data)? In addition, should cable owners with market power be required to offer some cost-based capacity to competing services (e.g. for open source AI models)?
In a world where a handful of large companies own the majority of the world’s internet highways and manage the bulk of the bit streams, what type of new rules are needed to keep internet services competitive and protect consumers?
Some answers may be provided next year in a US FCC docket, which seeks to prioritise landing rights for subsea cables that export preferred AI models.2
In other nations, where cable landing permits have been the exclusive domain of maritime or communications regulators, the new AI-driven cable boom is likely to bring trade, competition, digital rights and economic development officials into the fray too.
A longer version of this article will be included in the March 26 issue of Intermedia.
and then