feature article
Subscribe Now

A Sale of Thirty-Two Bitties

AMD’s Multi-pronged Strategy Is Certainly Different, But Is It Better?

It was the best of times, it was the worst of times; it was the age of wisdom, it was the age of foolishness… it was the spring of hope, it was the winter of despair… we were all going direct to Heaven, we were all going direct the other way…”   Charles Dickens, A Tale of Two Cities

Is AMD the best microprocessor maker in the world, or the worst? Is the company more competitive than it’s ever been, or about to hit bottom? Are they blazing a trail to success, or on the road to perdition? 

Or the real question: Are you planning to buy AMD chips or aren’t you?

Advanced Micro Devices has new management, a new logo, new technology, a new focus on embedded systems, and a new product mix. What they don’t have is a lot of money, and money equals time. There isn’t a lot of runway left for Big Green to get airborne again.

On the plus side, AMD is one of the oldest and best-known microprocessor makers in the world, and one of only two credible x86 makers. If you want x86 chips – and you know who you are – then AMD and Intel are your only choices. AMD’s chips are typically cheaper than Intel’s, generally more power-efficient, and often better suited to embedded systems. While Intel continues to dominate the obscenely lucrative server and high-end markets, AMD aims a bit more toward the mainstream.

But, unlike Intel, AMD doesn’t have all its eggs in the x86 basket. The company is also a bona fide ARM licensee, and it has been cranking away on a number of ARM-based chips. That makes AMD the world’s only ARM/x86 switch-hitter. Like Pablo Sandoval, AMD can bat either left- or right-handed, depending on how the game is progressing. Want x86 compatibility? Yeah, we got that. Prefer ARM’s architecture? We’ve got solid hits there, too.

Heck, AMD even has top-notch graphics technology, thanks to its acquisition of ATI. So it’s a triple threat: x86, ARM, and GPUs! It’s like a one-stop processor emporium. What could be better?

Almost anything, judging from the financials. Wall Street doesn’t think AMD is worth spit, and customers are leaving the company’s products sitting on the shelves. The chipmaker is burning through its cash at an alarming rate as it sees its traditional x86 sales plummet even as it spends on R&D for new ARM-based chips. Even the iconic HQ building in Sunnyvale was sold to generate some ready cash. AMD now rents back its own office space.

What went wrong? Well, the PC business pretty much disappeared, for starters. AMD was always a distant second to Intel even when the PC business was booming. But back then, even a small share of a huge business was enough to keep a big corporation like AMD afloat, with enough profit left over to sink into advanced R&D. The company invested in new CPU architectures, including the first 64-bit x86 designs, and in its own semiconductor-manufacturing technology, back when AMD fabricated its own chips.

But as the PC market shrank, AMD’s share of it declined to the point of diminishing returns. The company spun off its manufacturing capacity as GlobalFoundries and cut way back on CPU designs. The writing was on the wall: the PC business wasn’t coming back, so AMD had to either find new markets or perish.

The obvious solution was to target its x86 chips at embedded systems, and this it did. After all, AMD’s chips were usually a bit slower, cheaper, and more efficient than Intel’s equivalents; what better way to pivot than to sell those chips to embedded developers?

The trouble with that strategy is that embedded sales don’t count the same as PC sales, at least not financially. Traditional x86 markets like PCs and servers command a big price premium, making sales into that channel unnaturally profitable. Even if AMD could sell just as many chips as before (and that’s a big if), it couldn’t charge as much for them, so revenue would sink anyway. Just as much silicon; just as much work; not as much money. Not a happy equation.

Moreover, embedded markets require domain-specific knowledge. Whereas all PCs and servers are pretty much the same (“Does it run Windows?” “Yes.” “Okay, I’ll take a million.”), every embedded market is different. If you’re working on mission-critical aerospace systems, do you want to buy microprocessors from a company that doesn’t know the first thing about your needs? Combine the lower average selling prices with the fragmented and unfamiliar market niches and you get expensive and frustrating corporate tail-chasing.

The second part of AMD’s strategy was to buy into the ARM ecosystem. As a full-fledged ARM licensee, AMD could design its own embedded processors, a clever strategic move that arch-rival Intel would never make. For once, AMD could level a powerful strategic weapon against Intel, backing the giant into a corner of its own making. Well played, AMD. Game over – right?

Nah. Being an ARM licensee only boils AMD in a different stew pot. In x86-land, AMD had one competitor. Now it has a hundred. Every chip company that can still fog a mirror makes ARM chips; how is AMD any different? And how will selling (relatively) low-cost ARM-based devices help shore up its sagging profit margins? An ARM product line means a whole new R&D effort, plus all the fragmentation problems inherent in the embedded marketplace. And that helps how?

Maybe the third leg of AMD’s technology stool provides the answer. Ever since AMD acquired ATI, the company has had world-beating graphics technology (“visual computing” is the preferred phrase), which was becoming increasingly important in consumer electronics and elsewhere. Again, a clever strategic move that its (now two) major rivals likely wouldn’t duplicate. Intel has its own in-house GPU architecture to protect, as does ARM with its Mali technology. That leaves nVidia as AMD’s only standalone competitor in the GPU space. By most accounts, ATI and nVidia are neck-and-neck in terms of GPU performance, so AMD has a real star in its lineup. And that’s on top of x86 and ARM, the world’s two most valuable CPU franchises. How can you lose with that roster? 

Slowly, that’s how.

Donning our green eyeshades and glancing at the financial pages, we see that nVidia just finished selling $1.225 billion worth of chips in the most recent fiscal quarter, which puts it on track for about $5 billion in sales for the year. Of that, $173 million, or about 14 percent of revenue, was profit. The sales were overwhelmingly graphics chips ($991 million), not nVidia’s relatively new Tegra processors ($168 million), although Tegra sales are growing nicely.

Compare those results to AMD, which did $1.429 billion in revenue last quarter but saw only $17 million – just over one percent – in profit. That’s just one-tenth the amount of cash that nVidia netted, even with slightly higher top-line sales. And AMD’s meager earnings follow two straight quarters of losses totaling $56 million, more than absorbing the recent gains. AMD is running at a net loss so far in 2014. In fact, the company hasn’t turned a full-year profit since 2011, which means that nVidia makes more money in a good month than AMD has made in the past few years. 

What little profit the Sunnyvale firm does occasionally eke out goes to paying off its corporate credit card bills. AMD took on a lot of debt when it acquired ATI, and the “debt servicing” eats up a significant portion of its monthly budget. Just like many consumers consolidate their debts and refinance them into new loans, AMD has pushed out many of its debt obligations, but that just forestalls the inevitable. The ATI debt needs to be paid eventually, and AMD isn’t in any position to knock down the bill.

Wall Street is not forgiving. AMD’s market capitalization (the overall “price” of the company if you were to buy it) is $2.14 billion, while nVidia’s is over $10 billion. (For comparison, Intel, IBM, and Qualcomm all have market caps well above $100 billion). In other words, AMD is worth only as much as 4–5 months of its own sales. Most companies in this business are valued at several years’ worth of revenue (the “multiple”), so AMD’s low market cap reflects a deep mistrust of the company’s prospects. Wall Street has put AMD on life support.

The company does have almost $1 billion in the bank, which sounds like a lot. But that’s less cash than it’s had at any time in nearly 15 years, and a billion doesn’t go as far as it used to. This is a company that burns through tens of millions every month – millions per day – so its cash reserves won’t last very long at this rate. Economics 101 teaches us that the purpose of a business is to generate cash, not to expend it. AMD has experienced waves of layoffs and is on its fourth CEO in as many years. The company’s dwindling cash position means it has had to cut back on R&D, which can’t be good news for products in the pipeline. Repurposing x86 chips from PCs to embedded applications is comparatively cheap, but designing new ARM-based devices from scratch probably ain’t. And a rumored x86/ARM hybrid CPU architecture would be very expensive to develop, indeed. The quickest way to cut costs is to lay off the most expensive (read: talented) staff.

According to some of the diaspora, AMD operates as a “used-to-be-big company,” meaning it retains its large-scale bureaucracy, but without the large-scale headcount needed to operate it. Departments that have been hollowed-out by layoffs struggle to manage the old procedures. Progress can be slow and inefficient. Internally, AMD retains its big-company atmosphere, even though it’s no longer a big company. The old operating system has accumulated too much cruft. What the firm needs is an organizational reboot.

If you were AMD’s CEO for a day – and who knows, you may get your chance soon – how would you manage the company? Barring any major catastrophes, the PC market isn’t going to suddenly revive itself, so we can forget any deus ex machina resurrections. Granted, Intel’s top-end server processors can be ludicrously expensive, and that creates a pricing umbrella under which you can operate. But only for a while. AMD has underpriced Intel in the past, but that window always closes whenever it suits Intel to do so. It’s nice business when you can get it, but you probably shouldn’t depend on it for the long term.

That leaves the embedded market(s). Do you attack them with x86, with ARM, or with both? 

Relatively few embedded systems need or want x86 compatibility, and many can’t tolerate the heat or power requirements. AMD can price its chips any way it wants to, so that’s not an issue. Making those processors application-specific will take time and resources, however, neither of which you’ll have in much abundance. So you’ll likely start out with some modestly tweaked PC processors and market them as embedded chips. Precisely what AMD has done up to this point, in other words.

What do you do with the ARM half of the house? ARM-based micro-servers are all the rage these days, and AMD has the chops to pull off a server play, after years of peddling Athlon. The company has no particular SoC expertise, nor does it have any unique networking IP. But it does know the server market (at least, as it existed a few years ago). So there’s that.

How about ARM/x86 hybrids or combos? A-ha, this is one area where you’d have the market to yourself. Neither Intel nor any of the other ARM vendors can switch-hit like you can. You’re uniquely ambidextrous.

But ask an ambidextrous person how useful that skill is, and you’ll get something along the lines of, “not very.” Having two dissimilar processors in the same chip might be a fantastic technical challenge and (if it works) a technical triumph, but how useful is it, really? What practical applications can you think of that require both ARM and x86 compatibility?

And if those applications don’t already exist, imagine trying to convince someone that they need to design a new system around a chip with both. Changing processors is already akin to a religious conversion for many developers; now try forcing two processors on them simultaneously. It’d be like changing religion and changing language on the same day. Like flying cars or VCR/DVD combo players, hybrid processors seem like an interesting technical solution in search of an actual problem. Given your company’s limited technical and financial resources, will you sink precious time and money into a risky and expensive ambidextrous CPU if there’s no obvious payoff?

How about aiming for the low end? You could make AMD the low-price leader, producing budget x86 chips alongside inexpensive ARM-based processors for a range of cost-sensitive embedded applications. You’d get by on thin margins but make it up in volume.

Volume is indeed the Number One driver of semiconductor profitability, but only if you own and operate your own fab, and AMD doesn’t do that anymore. AMD sends its chips out to third-party foundries (sometimes GlobalFoundries, sometimes not), just like everyone else. Filling the fab lines is no longer your concern, so volume, per se, isn’t a panacea.

Besides, the low-cost niches seem to already be filled. Any embedded segment worth anything already has an incumbent supplier or two. Unless you’re going to discover a brand new multibillion-dollar market that everyone else has somehow overlooked – oops! – you’re going to have to compete with the established players. On their own turf. And you with no particular expertise in that area. Against vendors that often have government backing, lower overhead expenses, and no compunction about price dumping. Good luck with that.

Hey, it could happen. The 8088 processor wasn’t especially popular when IBM created its first Model 5150, and look how that turned out.

Here you are, with three major weapons in your arsenal, but no clear target to aim at. You’ve got ARM, you’ve got x86, and you’ve got GPUs. And a fourth weapon, if you count AMD’s 45-year history and brand recognition. (Or maybe that’s a liability…) The x86 architecture is the most valuable, in the sense that it’s the hardest to duplicate. But it is paradoxically perhaps also the least-valuable asset moving forward because it’s not clear which customers are going to want x86 compatibility.

ARM compatibility is also valuable – all the cool kids are doing it now – but devalued precisely because of that. AMD is just another ARM vendor, and not an especially well-established one at that. How do you differentiate from all the other ARM vendors, most of whom have carved out some sort of niche identity for themselves?

On the GPU front, you’re head-to-head against nVidia, where you’re doing fine, as long as you keep investing and innovating. That’s a market that never holds still. You’ve also got a technical advantage over both Intel and ARM, which nurture their own in-house designs. And you’ve successfully integrated the GPU and a processor on a single chip, proving that heterogeneous computing can work, even if it did prove trickier than you expected.

You’ve got almost a billion dollars in the bank, and a clock with a few years left to run on it. What’s your next move?

“There’s no need to fear. Underdog is here!”

Remember the Saturday morning cartoons with Underdog and his sidekick/girlfriend Sweet Polly Purebred? As mild-mannered Shoeshine Boy, undercover superhero Underdog kept a low profile. Nobody paid him much attention, but when the bad guys threatened, he sprang into action to save the day. AMD is definitely an underdog, and it does a pretty good Shoeshine Boy impression. Maybe the company’s dual personality really covers a hidden superpower under that humble exterior. Let’s hope it springs into action soon. 

3 thoughts on “A Sale of Thirty-Two Bitties”

  1. I’ve done a couple server designs around Intel Xeon products, and while general documentation is easy to access with NDA, unless there are serious sales numbers behind the design, it’s difficult to get additional support or docs.

    AMD has been even a bigger black hole to design with for similar reasons.

    Given that, it’s much easier to sell a design with Intel product inside, which makes the design/sourcing decisions very lopsided.

  2. How about a series of low-end x86 chips for embedded systems? Would that fly?

    Back in Ye Olden Tymes, AMD and Intel used to make a bunch of embedded x86 processors, but they both canceled them after a few years. That kind of burned a lot of embedded designers, who where then gun-shy about adopting x86 chips again.

    Could AMD stake out a territory for itself by committing to x86 in embedded?

Leave a Reply

featured blogs
Aug 18, 2018
Once upon a time, the Santa Clara Valley was called the Valley of Heart'€™s Delight; the main industry was growing prunes; and there were orchards filled with apricot and cherry trees all over the place. Then in 1955, a future Nobel Prize winner named William Shockley moved...
Aug 17, 2018
Samtec’s growing portfolio of high-performance Silicon-to-Silicon'„¢ Applications Solutions answer the design challenges of routing 56 Gbps signals through a system. However, finding the ideal solution in a single-click probably is an obstacle. Samtec last updated the...
Aug 17, 2018
If you read my post Who Put the Silicon in Silicon Valley? then you know my conclusion: Let's go with Shockley. He invented the transistor, came here, hired a bunch of young PhDs, and sent them out (by accident, not design) to create the companies, that created the compa...
Aug 16, 2018
All of the little details were squared up when the check-plots came out for "final" review. Those same preliminary files were shared with the fab and assembly units and, of course, the vendors have c...
Jul 30, 2018
As discussed in part 1 of this blog post, each instance of an Achronix Speedcore eFPGA in your ASIC or SoC design must be configured after the system powers up because Speedcore eFPGAs employ nonvolatile SRAM technology to store its configuration bits. The time required to pr...