Twitter: There’s Gold in Them Thar Tweets
SEE LAST PAGE OF THIS REPORT Paul Sagawa / Artur Pylak
FOR IMPORTANT DISCLOSURES 203.901.1633 /.1634
sagawa@ / firstname.lastname@example.org
October 23, 2013
Twitter: There’s Gold in Them Thar Tweets
- TWTR is losing money today, but not for long. Stripping out stock compensation, TWTR’s COGS (35.5% of sales), R&D spending (34.5%), and SG&A (40.7%) currently yield (-9.7%) non-GAAP operating margins, compared to FB at 41%, GOOG at 32% and LNKD at 15%. A big piece of this is TWTR’s late start on monetization – it only began selling ads in mid 2010, 4 years after its launch and 6 years after FB’s first ad sales. With sales more than doubling YoY, TWTR’s structural cost advantages vs. its cloud-based peers should become more apparent – it does not have to support massive searchable archives containing hundreds of billions of image and video files like GOOG and FB, and it does not have to maintain an enterprise sales and support organization like LNKD. With stock compensation expenses also likely to ebb after the IPO, we expect GAAP earnings to turn positive before the end of 2014 and GAAP operating margins to top 35% by 2017, while non-GAAP margins approach a sustainable level north of 43%, better than GOOG, FB or LNKD.
- TWTR margins are below water and comps, due to late start on monetization. TWTR is losing money, a condition once common for TMT IPO candidates, but less so in recent years. 58% of TWTR losses are related to stock compensation, which is likely at an unusual high point ahead of the IPO. Even stripping these out, operating margins of (-9.7%) are far below FB’s 41% and LNKD’s 14.7% calculated on the same basis. TWTR has only been ramping revenues for 3 years, since it began selling ads in 2010, while it has been building infrastructure and usage for more nearly 7 years. Arguably, it has established a cost basis that anticipates much larger future revenues.
- Scale is an important driver of profitability. For online services like TWTR, COGS are almost entirely data center costs, which are responsive to usage volumes and subject to considerable economies of scale. Moreover, revenues are only partly driven by usage, with additional key drivers advertising density and pricing unrelated to volume. Operating expenses are more driven by ambition than current volume – investments in engineering new services or expanding markets – and should also decline as a percentage of sales with increasing scale. Given, its late start on monetization, we believe TWTR can and will grow its sales considerably faster than its infrastructure costs or its operating expenses, allowing a fairly quick path to profit and an asymptotic approach to sustainable margins to rival or surpass its obvious peers.
- TWTR’s mature gross margins should be best in class. LNKD’s current GM (ex stock compensation) are 79%, FB’s are 75%, GOOG’s are 65% (ex MOT) and TWTR’s are 64%. For these companies, the predominant costs of revenue are the data processing infrastructure and related personnel, but contrasting business models dictate very different sustainable margins. GOOG and FB’s infrastructure costs will be the highest – indexing the entire web and maintaining a searchable archive of 100’s of billions of images both have extraordinary storage and processing implications. LNKD may have the least intensive infrastructure requirements, but carries significant non-tech costs in its COGS to support its enterprise HR business. TWTR’s near-real-time use case minimizes expensive search and storage costs, but creates a “need for speed” that could be greater than either GOOG or FB. Balancing these factors, we would expect TWTR’s margins at scale to be far closer to LNKD at the upper end, than to either FB or GOOG at the lower.
- TWTR R&D/Sales should trend toward GOOG. GOOG’s R&D, which includes funding for “moon shots” and for computer science research that leads the industry, has run between 12 and 14% of sales for the past 6 years. FB runs even leaner at 9-11% over the past 2. LNKD product development costs have been 21-23% of sales over 3 years, a rate the company believes will decline in the future, and which may include product management activities to support its enterprise products which may not have an analog at FB or TWTR. While we expect TWTR to be aggressive in developing new business opportunities, we expect the same of FB, and see 10-12% (ex stock compensation) as a reasonable sustainable level in the future.
- Media deals could help cut marketing expense. TWTR SG&A, at 40.7% of sales, is dramatically higher than GOOG (20%), or FB (24%), but slightly below LNKD (43.1). However, TWTR’s advertising-based business model is much more analogous to the first two than it is to LNKD, which must maintain a substantial sales force to sell its enterprise HR solutions. Over time, we would expect both FB and LNKD SG&A to trend toward GOOG’s levels, with TWTR, perhaps, having a slight advantage due to the prominence of its co-promotion activity with media partners.
- Excessive stock compensation expense likely IPO related. We expect the current high level of stock option recognition to continue for several quarters, as the IPO, and the subsequent expiration of holding periods will trigger vesting and exercise. TWTR’s stock compensation expenses have nearly doubled YoY, averaging 18.7% of sales YTD. This is consistent with FB, which saw an even bigger jump – from 5.8% to 30.9% – around its IPO in 2012 and have since fallen back to 12.3%YTD. LNKD was less generous to its employees, with a modest IPO bump from 5.7% to 8.9% in 2011. While there are examples of mature companies with egregious employee option policies, GOOG, which has tempered stock compensation expense to 5-5.5% of sales over the past few years, could be an appropriate long term model.
- Sustainable operating margins, including stock compensation, could top 35%. Assuming GM between 75-80%, R&D at 10-12%/sales, SG&A at 20-23%, and employee stock compensation at 6-10%, long term operating margins at the mid-point would be 36.5%, or 44.5% on a non-GAAP basis. At these levels, we would expect TWTR’s mature margins to be somewhat better than LNKD, GOOG or FB. We note that infrastructure investment, expenses, and compensation are subject to management discretion, and we believe that spending above these levels in the near term is justified in pursuit of the significant further opportunities that are available to TWTR. Still we believe revenue growth will be sufficient to bring TWTR to profitability before year end 2014, and to reach our target sustainable cost levels by 2018.
The Cost of Tweet
TWTR is losing money. This is not a sign that a second internet apocalypse is close upon us, nor is it an indication that TWTR’s business model is inherently inferior to those of the “social” internet companies that came before it – FB, LNKD and GOOG – all of which have proven to be successful at making money. It is also a fact that is not likely to remain true for very much longer.
TWTR only began selling advertising just three years ago, and even then, we have argued, the company has been extraordinarily cautious about the ad density on user timelines. For a company with costs largely fixed relative to sales, a late start on monetization has had the obvious effects on profitability. However, advertising sales are growing with a vengeance, expected to be up more than 120% for 2013, and likely to stay strong as user growth and engagement expand, timelines finally get a denser stream of advertising and high-priced video ads from TWTR’s extensive media partnerships begin to kick in.
Assuming sales growth decelerates to 40% by 2017, we believe TWTR can be more profitable than its peers. Stripping out stock compensation, to be addressed separately, its gross margins are at 64% YTD. By comparison, LNKD’s were at 79%, FB’s 75%, and GOOG’s (ex MOT) were 65%. For all of these companies, data center costs are the primary element of COGS. TWTR’s near real-time business model does not require it to maintain massive searchable archives of storage intensive image or video files like either FB or GOOG, and thus, its infrastructure should be considerably less costly in the longer term. We peg TWTR sustainable gross margins at 75-80%.
GOOG, home of the “moon shot”, runs R&D at 12-14% of sales. FB runs leaner, at 9-11%. There is no reason a mature TWTR, currently spending 32.5% of sales on R&D, should spend more freely than FB. Likewise, TWTR’s SG&A, now nearly 41% of sales, is far higher than either GOOG (20%/sales) or FB (24%), and only slightly behind LNKD, which runs a substantial direct sales and support organization for its enterprise HR business. We believe that both TWTR and FB will scale to approach GOOG’s SG&A levels. Finally, stock compensation is up to 18.7% of sales YTD. This is typical of an IPO year – FB’s employee stock spending jumped 7 fold in 2012 to 30.9% before dropping to 12.3% so far in 2013. While the market for engineering talent remains frothy, LNKD managed to make do paying its employees 8.9% of revenues and GOOG has been even more frugal, with stock compensation expenses running between 5-5.5% of sales for the past few years.
All in, we think TWTR can get to better than 35% GAAP operating margins by 2017, as monetization catch up to usage, as economies of scale in data center infrastructure kick in, and as expenses level out to sustainable levels. We expect TWTR to be profitable by the end of 2014, as revenues nearly double again, and post-IPO employee stock compensation ebbs back to reasonable levels. We expect that many sell side models will stay overly conservative on the cost side, leaving the company with relatively easy expectations ahead of it. We continue to believe that a better than $30B IPO valuation can be justified on conservative assumptions.
Exh 1: Twitter Revenue and Cost Forecast
Starting From the Red
Twitter heads to its IPO with impressive sales growth built on user growth, increasing engagement, rising ad density and buoyant ad prices, but an unmistakable red number at the bottom of its income statement. Twitter has been losing money, and according to the amended S-1, that loss is moving in the wrong direction, with this year’s net loss likely to be at least double last year’s $79M. Part of that loss is driven by employee stock compensation, which inevitably balloons around an IPO – Facebook’s stock compensation to its employees increased more than seven-fold in 2012 to more than 30% of sales before coming back to a more reasonable 12% this year. Twitter’s largess to its workers is a bit less egregious for shareholders – we expect payments this year to hit about 18% of sales – and will almost certainly wane once insider lock-ups and vesting pass in 2014 (Exhibit 2). We will address our expectations for stock compensation later in this piece.
Exh 2: Stock-based Compensation Expense, as % of Revenue
Exh 3: Twitter Ad Revenue and Monthly Active Users by Region, 2010 – Q3 2013
The other main reason for Twitter’s red ink is its nonchalance in monetizing its rapidly expanding platform. Twitter operated for four years on venture money and by selling licenses to its user data before it deigned to sell its first ad in 2010 (Exhibit 3). By then, the service was already reaching nearly 50 million monthly active users around the world. Since, ad sales have ramped quickly, rising nearly 250% in 2012 and on pace for nearly 125% growth this year, but until very recently, user timelines have remained noticeably sparse with advertising. Indeed, management noted in the S-1 that “decisions that we make regarding optimizing user experience and satisfying advertiser demand” is a key driver of the “advertising revenues per 1,000 timeline views” metric that it uses to measure monetization, and that the algorithms generated by those decisions were regularly refined to maximize “the value of our platform to users and advertisers”.
Exh 4: Internet ad CPM Ranges by Media, 2013
Exh 5: Twitter Amplify Partners
We have noted a marked increase in the amount of advertising on our own timelines over the last few months and can only believe that Twitter has refined its algorithms to weigh the “satisfying advertiser demand” side of the equation more heavily. Noting also that video ads, which typically carry CPMs 10-15 times higher than display ads, are growing in the mix as a consequence of the many Amplify partnership agreements signed with major media players over the past quarter, Twitter ad pricing should also be improving (Exhibit 4-5). The improvements in both ad density and pricing are evident in the acceleration of the “Ad Revenues per 1,000 Timeline Visits” metric, particularly in the US, where the YoY improvement has gone from 9% in the March quarter to 50% in the recently revealed September results (Exhibit 6). Given the nature of the Twitter timeline – i.e. a chronological stream of 140 character posts – and the context by which its users visit – i.e. seeking timely and topical information – we believe that Twitter’s well targeted ads are particularly well suited to its service and that the timelines can absorb even greater advertising density without damaging the user experience.
Exh 6: US Ad Revenue per 1,000 Timeline Views, Q1 2012 – Q3 2013
As Twitter’s advertising density and pricing increase, and as its user base grows and becomes more engaged, we expect the company’s revenues to grow at a better than 60% annual growth rate over the next 4 years, forecasting roughly $4.5B in sales for 2017. Given that Twitter’s costs are primarily driven by its usage rather than its revenues, raising the level of monetization will go a long way toward improving the bottom line. We plan to delve more deeply into the reasoning behind our sales growth projections in a future piece.
The COGS Must Be Crazy
For cloud-based services, data center costs represent the large majority of the costs of goods sold. These costs – the amortization of related capital spending, lease payments, data center personnel, and related services expenses – are semi-variable as management makes decisions to invest in capacity and functionality in response to the usage of the service. This is true of Twitter, and of comparable companies like Facebook, Google and LinkedIn. Note that the investment decisions, and thus, the resulting COGS, are related to the usage of the service rather than its monetization, so improvements in the density and/or price of advertising should directly drive gross margins higher.
Stripping out employee stock compensation expenses from COGS, Twitter’s current gross margins are around 64% (Exhibit 7). On the same basis, Google’s gross margins (excluding Motorola) are 65%, Facebook’s are 74% and LinkedIn, including the bulk of its separately reported depreciation and amortization expenses, are roughly 79%. The differences amongst these comparable companies’ margins reflect differences in the amount and the sophistication of the data center operations necessary to support their various businesses.
Exh 7: Gross Margins – GOOG, FB, LNKD, and TWTR
For example, Google’s search business maintains an index of more than 50B web pages, visiting more than 20B pages each day as it crawls the web for updates. Google users execute more than 3 billion searches every day, with the most popular content cached on Google’s own servers to speed response times. Google’s YouTube subsidiary streams more than 6 billion hours of video every month, from a searchable archive that contains more than 250 million hours of video. The Google Maps database requires more than 21 petabytes of storage – e.g. 21 million gigabytes. The epic scale of these business units, along with a number of less epic businesses, requires the world’s largest and most sophisticated data processing infrastructure, staffed by the industry’s most accomplished data center organization. The investment required accounts for Google’s gross margins being somewhat lower than many other web-based businesses.
Facebook’s infrastructure needs may be less imposing than Google’s, but they are epic in their own way. Facebook manages the world’s largest archive of image files, with more than 240B user uploaded photographs contained in its data centers and instantly available to its users, adding more than 350M new photos every day. This archive requires more than an Exabyte – i.e. 1,000 petabytes or a billion gigabytes – of storage. Facebook’s 1.2 billion active users spend 20 billion minutes a day on the site (Exhibit 8). The costs of the infrastructure necessary to support this storage and processing intensive business model are inherent in Facebook’s 75% gross margins.
Exh 8: Datacenter profiles of GOOG, FB, LNKD, and TWTR
LinkedIn’s infrastructure needs are modest compared to Google and Facebook. LinkedIn serves more than 200M users, but generates more than half of its revenue from H.R. services that it sells to corporate clients, giving it exceptionally strong revenue per user for its relative size. Moreover, as a service dedicated to supporting professional networking, LinkedIn does not have to maintain massive archives of personal photos, map data, or videos, giving it much less intensive data center needs. Accordingly, LinkedIn’s gross margins, at 79%, are best in class and could go higher as the company gains more scale.
Twitter’s infrastructure needs look more like LinkedIn than either Google or Facebook. While the service is generating a lot of daily activity, with 500 million tweets per day distributed to more than 230 million users who collectively visit the site 1.7 billion times a day, Twitter’s use case is almost entirely real time. Users very rarely look at tweets older than a day and archived tweets are available only by specific user. The very large majority of tweets contain no more than 140 characters of text and perhaps a URL link, requiring de minimus capacity to store. Most of the tweets containing embedded rich media, like photos and video, are sponsored tweets, which are not archived in a searchable database. Even the 500K or so daily video vines tweeted represent only 833 hours of video given their short form of 6 seconds or less. Storage required for vines is also de minimus as a single vine is under 1MB in size and current volumes would need under 500GB of additional storage a day. Twitter simply does NOT have storage requirements of the same magnitude as either Facebook or Google. Twitter DOES have a strong commitment to the nearly instant delivery of content, which carriers its own implications for infrastructure investment, but to a lesser extent, this premium on currency is also shared by Google and Facebook.
We believe that Twitter’s mature gross margins should be better than Facebook’s, and perhaps, equal to or better than LinkedIn’s. In our simple model, we are assuming that Twitter’s gross margins will reach 78% by 2017 and believe that this is a conservative projection (Exhibit 9).
Exh 9: Twitter projected revenue segments and cost/expense items as % of sales
Where We’re Going, We Won’t Need Roads
Twitter is spending a ton on R&D, even after stripping out the stock compensation for the engineers. At an estimated $222 million for 2013, Twitter is spending more than a quarter as much as Facebook on revenues that are a twelfth as high. In part, this is driven by a bit of a silicon valley feeding frenzy over a supposedly dwindling supply of top-notch engineers that resulted in Twitter making a number of acqui-hire deals for talent.
Assuming Twitter has been stockpiling talent for the future, it will grow into its R&D spending, and if the perception of an engineering shortage is accurate, it would find it hard to hire as fast as it is growing sales anyway. Ultimately, we don’t see a reason why Twitter would have need to keep R&D as a percentage of sales any higher than Google or Facebook. Google, famous for its commitment to “moonshot” R&D projects with potential payoffs long in the future, like self-driving cars, balloon-based wireless networks, and Google Glass, has spent between 9 and 11% of sales on R&D for each of the last 8 years. Facebook has historically spent even less, although its spending has jumped into the upper end of Google’s range in the past 6 months. It is not clear why a mature Twitter would need to spend any more than that. LinkedIn spends more, but includes development costs for its enterprise H.R. systems solutions in its R&D spending and may not be an appropriate comparable (Exhibit 9).
We are modeling Twitter R&D at 12% of sales for 2017, and again, believe that this is a conservative projection.
Exh 10: R&D Spend as % of Revenue – GOOG, FB, LNKD, and TWTR
Twitter spends 29.5% of sales on sales and marketing expenses. This is on a par with LinkedIn, which maintains a substantial direct sales force and customer support organization to back its enterprise H.R. solutions business. Twitter has no such excuse. Google and Facebook, with consumer driven cloud-based business models much more analogous to Twitter, each spent a little over 11% of revenues on S&M in 2012, with Facebook taking that even lower thus far in 2013. It is not clear why Twitter sales and marketing spending ought to be significantly higher than Facebook’s, but we are modeling S&M spending at 14% of sales for 2017. We believe that the actual spending ought to be considerably lower than that (Exhibit 11).
General and administrative costs have bounced around a lot for the on-line services cohort. Google’s spending appears to have settled down between 6 and 6.5% of sales in the last few years, but had ranged between 4 and 8% prior to 2009. Facebook has been even more volatile, popping from 7% to nearly 11% in its IPO year before falling back to 7.5% this year. It is likely that the legal and administrative costs of an IPO weigh heavily on G&A, although the bump from 12.8% to 13.1% during LinkedIn’s IPO is obviously less extreme. It is possible that Twitter’s next quarter or two will see a temporary bump in G&A due to the legal and advisory costs of the IPO (Exhibit 12). Longer term, we expect Twitter’s G&A to settle below the 8% that we have modeled for 2017.
Exh 12: S&M Spend as % of Revenue – GOOG, FB, LNKD, and TWTR
Exh 13: G&A Spend as % of Revenue – GOOG, FB, LNKD, and TWTR
Pay Me Now AND Pay Me Later
Stock-based compensation is a sore subject with most investors, and it is not clear how the tech industry has been able to exclude their generous employee option programs from the earnings numbers most commonly used by investors. Still, stock options are a particular distraction during an IPO year, as many employees will have awards, vesting and lock-up dates tied to the offering. Twitter believes that the value of its unrecognized pre-2013 employee awards is about $199M, most of which will be recognized by year end 2014. Additional employee awards, totaling almost $700M will be parsed out over the next 4 years, almost certainly augmented by additional incentive awards (Exhibit 14). Still, future awards are unlikely to be as generous as these pre-IPO allotments, and stock compensation as a percentage of sales will come down.
Exh 14: Twitter’s Projected Stock-based Compensation, Net of Forfeitures
Google’s 2003 IPO resulted in a substantial two year bump in its stock compensation, but since, the expense has settled in at 4.5-5.5% of sales. Facebook is likely still digesting its IPO-related awards, posting a whopping 31% of sales number in 2012 and 12.3% YTD. LinkedIn came public a year earlier, but saw its employee stock compensation number bump up in each of the last three years to the current 11% level (Exhibit 14). We believe that stock compensation for all of these companies will settle into the mid to upper single digits relative to sales. We are modeling 8% for Twitter in 2017 and hope that the company will look to bring its stock compensation number considerably lower in the years beyond that.
Exh 14: Recent Tech IPO Stock Based Comp as % of Revenue
Exh 15: Twitter Revenue and Cost Forecast, YoY Growth Assumptions
What’s it Worth to Ya?
Our current model projects Twitter to generate $4.45B in sales for 2017 and $1.65B in operating earnings for 2017 (Exhibit 15). Applying a current day Facebook or LinkedIn sales multiple – roughly 21x trailing sales – would suggest a better than $90B valuation at year end 2017. Even applying an extraordinary discount rate of 40% to account for risk in achieving our sales target, Twitter would be worth more than $33B today. At a 25% discount rate, justifiably lower given Twitter’s well established trajectory, the company could be valued at better than $45B.
Comparing Twitter more closely to Facebook and LinkedIn, we believe that by 2017, Twitter will have gained sufficient scale to generate GAAP operating margins close to its mature, sustainable level. As we have argued in this piece, we believe those margins should be as strong, or stronger than Facebook today with top line growth that will also be as good or better than present day Facebook. In that context, we believe that Facebook is a reasonable comp and applying its current trailing sales multiple Twitter’s projected 2017 sales is a useful tool for evaluating the value of Twitter at the upcoming IPO.