Over the course of the past six months or so, a debate has swirled around DISH Network (DISH). The company's co-founder and board chairman Charlie Ergen has suggested the organization could build the infrastructure necessary to power an entire wireless telecom network for a modest $10 billion. The key, he explained, would be the utilization of software rather than hardware, which lowers upfront costs between 40% and 60%. Meanwhile, MoffettNathanson's senior analyst Craig Moffett unofficially led a chorus of critics suggesting that estimate was alarmingly low. In his words, "The idea that DISH might spend $10 billion (their own estimate on previous conference calls) and then somehow be finished is, well, just silly." Moffett underscored his assessment by adding: "Verizon spends $15 billion annually to maintain a network that they've already built."

The financial truth is likely somewhere in between. While $10 billion might underestimate DISH's total cost of getting into the wireless business should Sprint and T-Mobile finally consummate their long-awaited union, virtualized networks rather than dedicated hardware with a relatively short lifespan probably does represent some degree of cost-savings.

The problem is that roughly two-thirds of AT&T's (T -0.64%) wireless network is now virtualized, but the move hasn't measurably lowered the telco's total capital spending or operating costs.

Photograph of woman at desj using a calculator.

Image source: Getty Images.

AT&T's software-driven shift hasn't driven savings

Given the choice, network operators like the idea of a software-driven blank slate over hardware-dependent infrastructure. Upgrades and repairs to a purely physical system often require workers climbing up towers with new technology, whereas virtual networks can be updated and integrated remotely with just a few keystrokes. That's where Ergen says DISH can save money. Others are skeptical.

It's not as if investors and industry insiders can only guess as to the benefits of virtualization, though. AT&T has been virtualizing its wireless network since 2016, with 65% of it being software-defined as of the end of last year. If there are any meaningful net savings to the shift, they're not evident in any of the company's (relative) numbers.

That's not necessarily an idea easy to ferret out with just a cursory glance. AT&T isn't just mobile phones. It's also landlines, and television, and (as of 2018) it's also HBO and Time Warner. The company doesn't break out capital spending by division. Of all its arms, though, mobility is still the single biggest -- accounting for about one-third of its total revenue -- and easily the most cost-intensive in terms of capital requirements. Of the $20 billion earmarked for capital investments in 2020, the bulk of it is intended to expand its 5G service as well as build the underlying fiber-optic network that 5G and the Internet of Things will ultimately necessitate.

Any noteworthy savings just aren't there ... not yet, anyway. As a percentage of revenue, the third quarter's 11.5% worth of spending on property and equipment is more or less in line with the past four years, each of which saw AT&T wade deeper into virtualization water. The proportion may have lowered, relatively, in 2018. That's only because of the addition of Time Warner into the mix, though, which didn't require a great deal of additional capital spending.

Graphic of AT&T revenue versus capital and equipment spending.

Data source: Thomson Reuters. Chart by author.

The shift hasn't dramatically lowered operating expenses for AT&T's mobility division either. The company spent 56.2% of Q3's wireless revenue just to keep its network up and running, but that's in line with operating expense levels we've seen since 2016.

Graphic of AT&T Mobility Revenue and Mobility Operating Expenses.

Data source: Thomson Reuters. Chart by Author.

Ergen has also noted that existing wireless providers are still servicing legacy 3G and even 2G connections, and he's right. Nevertheless, with three-fourths of its network now virtualized, one would have expected AT&T to at least show some sort of measurable savings by now.

Virtualization isn't a panacea

If software-defined networks are truly cheaper to plug in, then where's the money going? Aside from the modest expense linked to its lingering 3G and 2G connections, AT&T's financials point to a concern voiced by Ovum's principal analyst Daryl Schoolar last year when Ergen's idea moved into the spotlight. Schoolar explains: "There's a sense also that it probably costs you more money on the network services and integration side. You have to spend more money there integrating more disparate vendors together." ACG Research's lead analyst Chris Nicoll echoed the idea, warning: "They better find an excellent systems integrator to help them along."

In that vein, AT&T has done a great deal to ensure its software-driven network works as well as it needs to. It created its very own digital engine, called ECOMP (enhanced control, orchestration, management & policy), specifically aimed at making the company's virtualization ambitions a reality. In 2017, the company released its home-grown ONAP (open networking automation platform) open-source coding to help hardware vendors better meet AT&T's unique needs. Even then, the company concedes that the closer it gets to its 2022 goal of 100% virtualization, the more difficult it's become to continue moving forward.

It would be short-sighted to think DISH Network won't bump into a similar, unexpected headwind.

It's possible one could look at AT&T's CapEx and mobile operational spending trend and say there's a tiny amount of progress being made. One could also argue all of the prospective savings associated with virtualization lie ahead. After four years of steady progress down the virtualization road for AT&T, though, something demonstrable should have fallen into place by now. It's just not there. DISH, as well as AT&T investors, should be asking why.

The good news in all this? Carriers can do a whole lot more for consumers with software-minded networks than they could with hardware-driven networks.