"The Dog ate my Homework"

On the face of it, it is not drop-dead simple why Dropbox, the storage specialist for smaller and mid-size firms, necessarily benefits if we are entering a new era of computing some would logically call the “AI era,” and superseding the “mobile computing” or “cloud computing” era. 

Dropbox is most often used by companies with ten to 50 employees and $1 million to $10 in revenue, according to Enlyft. 

Nor is it clear why reducing Dropbox headcount by 16 percent, though perhaps an understandable response to slowing revenue growth rates, necessarily is connected to core tasks the company has to undertake to thrive in the next era of computing. 

To be sure, use of large language models and using generative AI presupposes processing of huge amounts of data and huge data sets, which of course requires lots of storage. 

Some note that a single LLM training operation can cost millions of dollars.

But training of LLMs is not something most smaller firms are going to be able to afford, so it is unlikely that Dropbox would see much revenue upside from serving that function. 

Likewise, training and inference operations will require lots of new processing cycles. Perhaps Dropbox believes it has an opportunity to provide such intensive compute support for its current customer base. 

Perhaps something else is partly at work. Retailers, for example, often explain poorer financial results by pointing to bad weather that kept shoppers away, or unusual weather that depressed demand for products with a seasonal purchase pattern. 

As true as that might be, it also is a convenient excuse. 

Some argue a wave of layoffs in the technology business is as much about doing what competitors are doing as much as anything else, aside from the argument that firms over-hired in the wake of the Covid pandemic. 

Business customers often do the same thing, arguing they must spend to “digitally transform,” become more agile, innovate faster or apply technology to reduce costs or remake product values. 

There still is a rational argument to be made that if firms expect lower revenues in the near future, then cost cutting makes sense. Cost cutting to preserve profit margins also makes sense. 

On the other hand, the stated rationale for any set of actions, strategies and tactics also includes a healthy dose of “because that is what people expect us to say or do” behavior. 

The “we need to get ready to invest in AI” argument is fairly new, though.

Generative AI and Large Language Models Will Not Affect Edge Computing Demand All That Much

Generative AI and large language models, as applied to many, if not most applications, will boost needs for advanced processors, electricity and cooling solutions. 

But it might not affect demand for edge computing in any significant way. Nor will it have measurable effect on data center or computing architectures, as AI training and inference operations still make most sense at centralized, hyperscale locations. 

Nvidia CEO Jensen Huang is quite positive about what routine use of large language models will mean for sales of advanced processors, citing an acceleration in demand for Nvidia processors. Some analysts believe generative AI could add as much as $6 billion in revenue for Nvidia within three years.  

Demand driver	Nvidia products that will benefit	Degree of impact
Increased demand for compute power	Datacenter GPUs, AI accelerators	High
Increased demand for data storage	DGX systems, storage solutions	Medium
Increased demand for bandwidth	Networking solutions	Medium
New security risks	Security solutions	Low

Others should benefit as well. Some argue infrastructure to support generative AI could reach $50 billion by 2028, for example. Some note that a single LLM training operation can cost millions of dollars. Those costs are going to be borne by app providers of all sorts. 

Data center CxOs might also tend to agree about changes generative AI processing will create. 

We already know that generative AI requires lots of computing cycles, which we might quantify as floating point operations per second. The actual impact from a single request depends on the size of the dataset that is interrogated, the complexity of the question or task or the type of query. 

Text responses, perhaps ironically, might require an order of magnitude more processing than generating an image. And generation of an image might require an order of magnitude more processing operations than generating a text-to-speech use case. 

Model Type	FLOPS Requirements
Large Language Model (LLM)	100+ petaFLOPS
Image Generator	10+ petaFLOPS
Speech Generator	1+ petaFLOPS

So capital investment budgets will likely be altered to add more-more processing power than we presently tend to see. As a background issue, there will be some additional demand for higher connectivity between data centers, data centers and peering points and domains to domains, though it remains unclear how much incremental demand might be generated. It is a non-zero number, but it is hard to quantify at the moment.

Change	Investment response	Needs driving investment
Increased demand for compute power	Investment in more powerful hardware	Generative AI workloads are computationally intensive
Increased demand for data storage	Investment in more storage capacity	Large datasets that are required for AI training and inference operations
Increased demand for bandwidth	Investment in more network capacity	To support the high-speed data transfer that is required for AI applications
New security risks	Investment in security measures	To protect data centers from the growing threat of AI-powered cyberattacks

It also seems likely there could be architectural impact as well. Edge computing makes sense to support lower-latency use cases and applications. It is unclear so far whether the training of large data sets or end user operations will actually drive edge computing very much. 

Inference operations might still need to be conducted at large hyperscale centers, not at the edge. For that reason, use of large language models might not actually cause data center architecture to shift to greater reliance on edge computing, for example. 

Higher energy requirements will likely lead to newer approaches to cooling, though. 

Architecture or design change	Possible actions by data center operators	Magnitude of capex impact
Increased demand for compute and storage resources	Deploy more powerful servers and storage systems	High
Increased need for cooling and power	Upgrade cooling and power infrastructure	High
New security and compliance requirements	Implement new security and compliance measures	Medium
Changes in data center layout and design	Reconfigure data center layout and design	Low

More Virtual Interconnection is Coming, Especially Driven by Edge Computing

One hallmark of markets that are deregulating, driven heavily by technology innovation or able to use new forms of virtualization is that the barriers that keep competitors from entering a market fall. The colloquial way of expressing this is the statement “firms outside our industry now are entering it.” 

Consider one example, namely the revenue Equinix earns from interconnection services. 

 source: Equinix 

In the fourth quarter, Equinix reported 447,600 interconnections. Most of those were traditional cross connects inside Equinix facilities. But 11 percent of those interconnections were of the wide area network sort that traditionally have been sold by communications service providers. 

In fact, the notion of “networking” arguably supersedes earlier language about “colocation” or “rack space”as the primary business function. 

source: Equinix 

Over time, more such interconnections might be virtualized. In the pre-internet era, data charges were based on both capacity required and the distance data has to travel. These days, charges are set based on capacity per port or connection, on a distance-insensitive basis. 

Logically, if interconnection is based on capacity and distance, then whether two domains are interconnected by a local area network or a WAN connection should not matter. Products sold still make a distinction between cross connects in a building and interconnections across a WAN. 

That might change in the future, as more interconnections between domains increasingly are a matter of virtual functions between locations rather than use of dedicated cables between servers. Edge computing, for example, should hasten that trend.