NVIDIA FY 2015 Q3 Financial Results: Record Revenue For Quarter And Year To Date

25 Comments

View All Comments

• chizow - Thursday, November 6, 2014 - link

  I'd agree their Tegra revenue is lower than they hoped, however, Q2 results captured most of their Shield ramp depending on when they recognize revenue. So Q3 would most likely capture a significant portion of their Nexus 9 ramp even though it launched in their Q4. I think they will need strong sales of these 2 devices along with another design ramp or two to eclipse the ~$160M they are seeing so far.
  
  Overall its not doing too poorly imo, its still ~15% of their business mix, and while they certainly need to make break-even a priority (I believe they stated yearly Tegra cost is ~$700M), they will need a strong Q4 to hit that mark.
  
  I don't think they are targeting the commodity device market though, I think they are going to remain focused on vertical device integration for high-end devices and use this as a vehicle for GPU licensing. I think K1 has gone a long way in achieving that and given Nvidia the kind of mobile credibility they need to get the attention of vertical integrators and SoC makers. Jensen said in his earning calls commentary that their licensing talks were very active and had entered some important stages, so maybe we will see some announcements on that front soon.

• jjj - Thursday, November 6, 2014 - link

  Q2 had no Shield revenue at all and Nexus 9 is nothing, and Tegra is doing extremely poorly.
  They claim not to target what they can't get and they can't get into phones without integration and even their quad A9 with integrated baseband failed for w/e reason.
  Nobody recognizes preorders as revenue, Shield was all Q3 and as i said mind the ASP. 100k units would be 33-35 mil while a last gen chip itself is 25-30$ at best depending on SKU and volume. Nexus 7 first gen sold almost 2 mil units per quarter, second gen was slightly bulky and costed more and sold a lot less. Nexus 9 has a high price and 4:3 AR and no way in hell it sells much. Even if they do 500k units per quarter that's 15 mil $ only at 30$ ASP so less than 50k Shield units. Tegra revenue right now is mostly devices ( the 2 shield and the Note) ,old console revenue and auto and that masks the very low volume.
  As for break even it was stated some years ago that it's 1B but before Icera and break even might be ok from a financial point of view but that's next to nothing share and from a marketing perspective is not sustainable and you can't afford to offer proper assistance to device makers to be able to actually compete. And with more compute units going on the SoC and more and better custom cores, cost would go up.
  The licensing quotes mean nothing, it's what companies say in general as long as it can't be proven as a lie,they can lie all they want. Not saying it is a lie ,just that it could be and it's not reliable info without actual facts. Licensing is also pretty small revenue even if you get huge volume.
  And their claims that they don't target some segments are just empty, sour grapes. Again something most companies would claim to justify their failures. They targeted that market with the quad A9 and it failed to get wins and they will target it again when they'll have what with but remains to be seen when that is and when they can make something sellabe with integrated Icera. There is no non commodity market in mobile devices anymore anyway. 600-700$ phones are a dieing breed. Nvidia either competes or they are out and right now they are out. Even Marvell has a lot more units then them at this point.
  TK1 seems great but they need to actually sell it and for w/e reason that's not the case for now.
  And the so called commodity market is not bad. Mediatek has 49%+ gross margins and that's solid for a semi company ,lower than Nvidia's great margins at 55.5% but solid.
  You got to be realistic about it, you can't just go positive on everything.
  Maybe they surprise us with quad Denver and Maxwell on 20nm with integrated modem soon and they actually gain some traction or maybe they'lll be late again and Qualcomm will have their custom core on 16/14nm FF before Nvidia can integrate.My guess is that they are trying to get their ducks in a row and once there they'll push hard but we just don't know when that will be or if they can get it done.

• chizow - Friday, November 7, 2014 - link

  How is Tegra doing poorly when it is still showing high growth compared to last year, almost nearing 2012's peak numbers while completely dominating the Android SoC market in terms of performance by nearly double? Performance and success is not all about revenue and profitability in the short-term when you have set your goals for long-term gains and traction in the SoC market. Obviously the end goal is to make money, but that only happens in rare cases immediately, Nvidia has staked their future on mobile computing and Tegra is their avenue for accomplishing this.
  
  While its true you don't recognize revenue on prepaid revenue, for contract manufacturing they may very well be recognizing revenue once they ship their contribution (the SoC) to the contract manufacturer. In the case they are just providing the SoC to an OEM, revenue is recognized when the SoC is shipped, so it depends if Nvidia is just selling the chip, whether they are the ODM, or they are the OEM, it all matters. For example, when Nvidia ship their GPU and memory packages to their retailers, they recognize the revenue even if the GPUs don't sell for another 4-8 weeks while AIBs bring the cards to market. Same if they strictly sold the SoC to an OEM or Tier 1 integrator, like Samsung, or HTC/Google for the Nexus 9, they recognize revenue as soon as the SoC is shipped. Its unclear how and when they recognize revenue for Shield since they are technically ODM/OEM, but I am sure they are recognizing some of it, there is nothing else to sustain that $160M number in Q2 and Tegra K1 was specifically cited there as a ramping product.
  
  As for breakeven and cost, I don't think they are worried about it, investors aren't worried about it while Nvidia is still making money overall because they believe in Jensen and his vision to help Nvidia break into emerging markets. I think many including myself are actually very impressed with Nvidia's ability to fight and create a niche in the mobile computing market while many others have fallen by the wayside (TI) or sat on the sidelines (AMD) and even industry giants with deep pockets (Intel) are having trouble breaking into this market. Reality of it is, Nvidia would probably be taking the R&D expense hit portion anyways as they don't like to layoff their talent.
  
  The phone market is something they are not interested in being a commodity player in. Simply put, graphics performance is not the #1 driver for smartphone SoC sales, its battery life and overall CPU responsiveness. Battery life is hugely dependent on integrated LTE for all the power states, and it doesn't seem like Nvidia is interested in producing 2 flagship SoCs to compete. You can read the interview here, Jensen admitted they made a mistake pursuing Tegra 4i and changed course:
  
  http://www.cnet.com/news/nvidia-ceo-sees-future-in...
  
  Again, you can see Nvidia is not positioning themselves to be a commodity SoC dealer, Mediatek is specifically cited as a reason they left that market. The licensing quotes do mean something actually, on a quarterly earnings call with investors and analysts, making misleading or misrepresentative comments can quickly get you in trouble, just ask AMD's leadership team regarding their comments a few years ago regarding Llano demand.
  
  Finally, Nvidia is so far ahead the rest of their Android competitors, they know if they just keep iterating and improving, they will eventually stand out and win the designs they want to win. Their biggest goal is growing and creating the markets they have started to nurture like automotive, Android gaming. Right now their biggest challenge is to convince devs and hardware makers that Android is a worthy gaming platform. This can only happen if they put the hardware out there, which they have in Tegra K1 and their Shield devices and they will push this advantage out even further next year with Erista which will leverage both 20nm and Maxwell. Certainly huge improvements expected there.
  
  Once Nvidia establishes clear tech leadership that others simply can't provide, that should bring the big vertical integrators and OEMs to the table to talk licensing or to buy Nvidia's SoCs outright, but we shall see!

• TheJian - Friday, November 7, 2014 - link

  You don't seem to understand all the revs up to this point are just delay tactics to get to discrete gpus in socs. You will start to see a shift from here forward especially with maxwell (look at the desktops for proof of this) and that is before the shrink which should allow integration of a modem if desired easily (or just use qcom's) as maxwell+20nm will drop power massively. The GPU will take over this decade, Qcom's ride is coming to an end (as modem drops importance, even anand said this about the modem). How fast do I need to hit my data cap? LOL.
  
  Maxwell is the first from the ground up made for mobile tech. I'm sure they wished they had more success (of course) with the first 4 revs (remains to be seen how good K1 is as more devices come), but the gpu will start to rule the high end devices soon and now they've put out their first custom core (before Qcom, they messed up 64bit). S810 isn't in house, and any K1 device on 32bit can just drop in the 64bit (pin compatible). K1 will start to improve the numbers for tegra (already is) and M1 is only 6 months away on 20nm. Good luck to cherry trail, S810 etc trying to stop M1 etc going forward. The ducks are, as you say, getting into their row ;)
  
  As the 64bit OS and it's software mature NV or someone else will put out a PC box competitor (with all the PC trimmings, including a full pc gpu if desired), and won't need Intel or AMD or Windows inside. NVlink removes the need for PCIE, Hypertransport and Infiniband. Unless Intel buys them, I wouldn't want Intel stock over the next few years at least not until they announce fabbing for a LOT more stuff from others that is or drop mobile which adds $4B+ to their income (instead of the losses of 4B+). NV will own a portion of the "new" PC market they are all about to create (by all I mean google with 64bit, Denver, better gaming, apps etc, ecosystem just getting going). The question is how much of the market can they take or can Intel buy them before they steal it? Will Jen even sell or does he just want revenge for their chipsets being killed (hence moving to cpus)?
  
  Either way Wintel will get hurt by the android/arm juggernaut that is building before our eyes as a gaming platform (and surely at some point a 64bit app platform too, app devs will follow units sold and soon pc like power, memory etc). Pascal should be ready in time for a far more mature 64bit android OS and lots of games taking full advantage of PC gpu tech but NOT on Wintel/DirectX. I really hope AMD gets in soon, but it may be too late for them (cutting even more employees and no doubt R&D again next year, you can't win like this). These two are the only companies who have a 20yr track record of great gpus and drivers to run them for gaming and ALL devs have experience with their new gpus for every rev (that now move into mobile). Everyone else has to play catch up now. On top, they'll have to pay lic fees just like Intel to compete as NV is now starting to sue over their stolen patents ;) It will take another 2yrs to see that shake out, but this will bring substantial money just like Qcom modem tech. Even anandtech called it the wild west of theft in socs. I think they just waited for gaming to get good enough and the market to get big enough to be worth suing. Nobody will escape NV/AMD, though AMD (if they have any patents worth lawsuits) will wait for NV to make the case and win before using that verdict in their own case. AMD has no money to make a suit currently, while NV has the cash on the balance sheet to wait for years to win here. They would be wise to let NV make the case first. Note NV left out apple on purpose (and smaller guys), the first case will be used on them at some point for some sort of agreement or suit (why fight the big 140B dog, attack smaller major players first then use verdict on apple and watch all others fall in line).
  
  You have to think big picture and longer term, not what is happening today. Also licensing isn't small if you have the right tech/IP. IE, MS makes $5 for every unit sold with android on it. I'm sure they laugh daily about this. Not sure this will always be the case, but to date google hasn't been able to get out of this.
  
  http://www.cbronline.com/news/mobile-and-tablets/m...
  ROFL. That's just samsung.
  
  http://www.infoworld.com/article/2622012/android/m...
  This has been going on for years and growing with each unit of android sold.
  http://www.zdnet.com/microsoft-is-making-2bn-a-yea...
  
  You'll either end up using NV gpus or socs (to avoid suits etc), or pay some fee just like Qcom/MSFT get for modem/OS crap. Cuda didn't make money in the first year either (but 7+yrs later it is the reason for 85%+ workstation gpu sales). Be it lic fees, socs, grid streaming to devices, or full on PC like machines soon, NV is set to make some money (not to mention cars etc which will use even more tech as they go self-driving, already 4 socs per car etc). You can't really make a gpu for anything today without stepping on NV/AMD patents probably and AMD just keeps getting weaker here. NV has 7000 gpu patents and only cited 7 for the suit, probably the kiss principle at work here for now...LOL. They can bleed on tegra until the 64bit ecosystem and games get up to snuff requiring NV (or AMD) like hardware to play (or you'll be left behind). They can wait out the lawsuits also. Intel lost and so will Qcom/samsung etc etc etc. It's just a matter of who loses what, how much etc. Of course there are multiple companies that might try to buy NV (or AMD, surely NV bid first as they're #1 gpu) at some point too. IE, Google, Microsoft, samsung, Intel, apple. Any of these could afford the ~$30B+ sticker price (probably far more), though it's probably tougher for Intel with the hate between them.
  
  They were a far weaker company in 2007 (balance sheet, revenue, assets, portfolio etc) with less opportunity for growth than now and had a share price that was ~double today with ~800mil profit. It doesn't take a rocket scientist to do the math here. You'll do well in NV stock for a while and now you get a dividend while waiting ;) I could go on but people should get the point when compared to their 2007 positioning. There are many ways to win here owning it (long-term, I'm not a day trader - usually...LOL).
  
  Agree with most of chizow's comment and add to it, Jen is a LONG term thinker. I doubt he ever thought Tegra 1 would make money, or even shield r1 (grid and shield only cost 10mil to dev each!). It's about maturing the android gaming platform. It's about matching the PC ecosystem at some point (gaming on a 64bit android platform via appstores/tegrazone etc, apps added at some point using cuda etc, thus skipping Wintel entirely and selling cpus by then in them), and becoming defacto gpus for many other types of devices at some point that all start to require powerful gpus and soon their cpu too as an Intel replacement. If you don't get on their own gpu treadmill, you'll pay lic fees. IE, if you don't buy a windows phone, MS still gets $5 from android anyway. NV wins with the same logic here until someone can come up with gpus that are radically different.
  
  Think about a PC that NV sells a GPU in, probably a SOC (basically same as an Intel cpu/amd cpu) running it costing far more ($100-200 undercutting Intel on cpu side, running 4ghz, huge fan/heatsink etc pc like), and having no windows fee. A great gaming system/pc like app system to sell into developing markets that can't afford great Wintel machine costs but has the same gpu power and possibly the same cpu power. You could cut a powerful PC box by $280 if you took $150 off Haswell/Broadwell etc and $130 from Windows, but it would be running the same gpu/driver abilities. A $500 box is easier to sell than $780 in developing countries right (and here too no doubt, insert any priced machine)? It will run the same HD's/SSD's/Memory etc. They'll still be selling in phones/tablets/chromebooks too, but make no mistake the goal is a chunk of Intel's $60B+ revenue also. Look at cpu scores vs. Denver, and imagine that soc running 4ghz today with double the die size (~123mm^2 now, so ~250mm - pre die shrink to 20nm here) and charging $100-200 for it vs. $30-35 for K1 for a whole line like Intel/AMD product lines. Just 10-20% of Intel's market adds massively to NV's report. IF they gain enough share do they at some point tell Wintel our gpus no longer will be compatible with your crap? LOL. We're no longer making Windows drivers? Hmmm....
  
  Grid went from 100 companies testing it, to 200, now 600+. Clearly there is interest in that too, and at some point it starts paying as they move from testing to buying it. They are nearing the 800mil mark of 2007 and none of this stuff is really kicking in yet, while being a much stronger company already vs. then with many more ways to drive revenue now. Outstanding shares will be below 2007 shortly also. As a stock this is a no-brainer IMHO until someone shows a gpu type product that can't be nailed by patents but can replace NV is everything they're moving to (datacenters, mobile, pc's, android pc soon, cars etc now). The real story starts to happen here next year most likely as 64bit matures for ARM's side along with the first REAL mobile gpus for NV at 20nm, opening up all sorts of stuff for NV cpu's and more gpus or lics. Willful infringement will be costly no doubt on the lic side for many (samsung ignored NV for 2yrs! Not sure about Qcom etc). Discrete driver enhancements also slip over to mobile as NV has stated before (along with cuda, gameworks etc). The whole shebang is a much bigger game than most realize I think. NV's goal is to get their gpu (and now cpu/socs too I guess) into everything, like Intel inside ;) They have many helpers getting them there. IE google for OS, fabs, ARM IP as a base cpu constantly updated, device makers, app stores to cover software sales leading to more of their product sales.
  
  I'm still waiting for NV to announce entering the game market themselves (as in Nvidia MADE games). A mere 100mil/year could fund 10-25 top quality games from 4-10mil each specifically highlighting their hardware. You don't have to sell many copies of great games to break even on them and push your platform. IE @ $10ea you only need to sell 10mil total over those 10-25 games to break even and build a library of reasons to buy more NV products. You already know your own hardware inside out. Until you get enough chips out though just keep paying for porting (gameloft just ported 6 of their top games to NV) as fast as possible to achieve the same goal. Get SteamOS ported to your ARM cpu also for quick porting of their whole linux library at some point and non Intel Steambox's! I'm sure Valve/NV is working on this if not NV alone.
  
  Alone NV wasn't worth much, but together with their ARM/android side partners and fabs they can easily get into Wintel's turf big time. Sheer unit numbers sold for arm already draws the same number of game devs as PC's do. I don't see anyone else (currently) on the map that can do the same without a gpu alternative that can't be sued. At $2bil+ units sold yearly soon (already? Gartner said 2.4B this year in july including PC's), think about just $1-2 in gpu lic fees per unit yearly and surely it would be more? I'm not sure what the fee would be per unit but think about Microsoft's $5 for the OS per unit (look at Qcom's modem royalties from Apple alone for an example far worse than MS). NV only gets ~4.4B in revenue yearly now. That's a HUGE gain no matter how that ends up and as in MS's case it's 90%+ margin there. You already do/did the work, you're just nailing everyone for using it. Interesting times ahead for sure.

• TheJian - Friday, November 7, 2014 - link

  oops, should be 75% workstation sales. But point is the same. 7yrs of groundwork with cuda lead to dominance. We're 5 models into tegra, so another 2 or 3 and...When they were doing T3, Jen said T4/5/6 were all being worked on at the same time, so clearly it was always a long game and certainly more people know who Nvidia is today than before tegra which is also good over time.

• jjj - Friday, November 7, 2014 - link

  You have any clue how much money ARM is making? How much of that is from licences and how much from royalties on how many chips?
  The GPU IP is very different from CPU and even more so from modem.Yes Qualcomm can make 8B a year from licenses but that's very different and might not last as it is.
  There are already others that license GPU and you also got Intel and AMD with IP.That means competition in something that is already at low prices.
  Nvidia might not even get the Intel cash soon anymore and even that deal is not one way. Nvidia pays Intel for patents and Intel pays Nvidia but Nvidia gets payed more because of Intel's much higher volume.
  As for mobile devices , the next step is glasses and yes we will need a lot of image processing but we also need very low power small chips and that means cheap ships. Yes in robots, in cloud they have opportunities but that wasn't the point. The point was that if they they want to be in mobile they need to do a lot more. Mobile is today, you can't have a tomorrow if you don't have a today. I said that in GPU they are doing great and felt no need to expand on that but in mobile they are not relevant because for w/e reason they are not getting wins.
  The modem is not at all less relevant, there is less and less competition and everybody is trying to catch up to Qualcomm. Connectivity is more and more important as the growth in connected devices is exponential.Nvidia didn't payed 367 mil for Icera because they thought the modem doesn't matter.
  Nvidia has messed up in Tegra and there ample evidence, first the financial numbers and device wins and then the exec changes in recent times for that division.You have to be delusional to claim they are doing well or pretend it was intentional to become irrelevant.When they'll be able to do better, i got no clue, can only wait and see.

• Pwnstar - Friday, November 7, 2014 - link

  It's not an SoC if it has a "discrete" GPU. Perhaps you meant SoCs that get close to discrete performance?

• przemo_li - Sunday, November 9, 2014 - link

  Yes, writing good driver is hard.
  
  But not that hard.
  
  Right now we have Mesa/Gallium to draw upon. That is proven path! (Intel on Lin/And, Freedreno made by one person on Lin that perform better than adreno.... etc.)
  
  AND gpgpu, mean more emphasis on open specs. Like full public ISA docs for CPUs right now.
  
  Intel have it, AMD have it (GPGPU team publish more docs, than GPU team).
  
  Nvidia do not have it.
  They started some limited cooperation regarding K1 and forward. As this gpu is served by desktop drivers for Nv. So Nouveau can be expected to move forward faster.
  
  But will they take it far enough?
  
  Also game development relay on gpu manufacturer support right now. This is limited supply possibility.
  
  Smaller teams will have easier to to get Intel attention. (Its just joining mesa IRQ, and they may even get their problems solved by 3rd party devs first!)
  On Nvidia? Black box driver, and reliance on guesswork.

• chizow - Thursday, November 6, 2014 - link

  Strong results from Nvidia, obviously due to the strength of their Maxwell GPU line-up. It will be interesting to see what they can do with a full quarter of Maxwell sales although inventory checks show stock levels are much better and they are (finally) facing more serious price competition from AMD.
  
  I would agree their Tegra revenue being nearly flat Q2Q is a bit of a concern, we'll see if the Nexus 9 ramp in Q4 has any impact on that. They will need a few more major design wins to improve on that number for next quarter.

• cpy - Friday, November 7, 2014 - link

  I bet those 900 cards will make them nice income boost, they are still sold out here and unavailab.e