Opinion: AMD goes for broke

Indianapolis (IN) - AMD’s actions in recent quarters have revealed a trend which can only be described in one way : All in. It’s a poker term made generally popular by a recent television show, but it basically means AMD bets the house on one hand of cards. More often than not, at least on the show, the bet does not pay off. But when it does, it can be life giving.

Vision

Since this article is an opinion piece, I can admit that I am a big fan of AMD’s public vision. I have been for a long time. I see their desired goals and product ideas as forward thinking and fresh in this industry. To me, where they’d like to go makes much sense. It’s also obvious that their business goals align with their customer’s needs. I see them listening to customers and actually giving them what they want and need. While that reality may, in truth, be the only competitive edge they have due to their relative position against Intel, they have done it consistently throughout their history in one form or another. And I’ve seen it almost exclusively since August 10, 1999 when they formally announced AMD64 specs.

I think AMD’s stated vision is on track with a good plan. It’s on focus and they are intent on working toward those goals of bringing forth the best possible products to all of the marketplaces they’re targeting. But one overriding fact shadows everything I see at AMD. I just don’t see any possible way they can make it without dramatic adjustments to their business.

On the one hand, I look at AMD, I see their goals, their stated desires and the plans they have to get there and think "Yes ! That is exactly what we need." Then, on the other hand I look at their current technology issues, their consistently weakening financial position, and the rate at which Intel is accelerating their competing product lines and I think "No ! Why can’t people see there is value in AMD’s vision. AMD’s light illuminates the path ahead. It’s an investment in a brighter future. Buying AMD today means the real work they have in mind will actually come to fruition one day."

Mounting debt

AMD is looking right now at repaying debt of about $5.1 billion in 2012 ($3.7 billion plus 5.75% interest compounded annually). This will repay only their recently received cash injection from the sales of senior convertible notes which are due to be paid back 4.5 years from now. And considering what AMD has planned expenditure-wise between now and then, that makes the gap of comfort very small.

AMD has plans by then for several new products and technologies. They’re expecting to see 45 nm by the end of next year, and 32nm in either late 2009 or early 2010. They are developing two new lines of processors, called Bobcat and Bulldozer, and these will have new technologies and extensions added which are still being developed. These new CPUs will target new markets where AMD has no current presence. That means additional expansion and capital requirements. And just where is AMD going to get enough money to make it happen ?

AMD’s entire annual revenue for 2006 was $7.6 billion ($5.3 billion excluding ATI). The bottom line was a $47 million loss. If we go back to 2005, their revenue was $3.9 billion with $543 million in net profit. In 2012, AMD will be looking to pay back at least $5.1 billion in hard cash from an unknown annual growth rate beginning at the current $7.6 billion in 2006.

Let’s be optimistic : If we consider a best-case scenario where AMD gains 10% per year, that equates to $14 billion in total gross revenue in 2012. Attaining that degree of increase, revenue will not come cheap. Much of it will be spent in manufacturing costs and expanding capacity, for without additional capacity AMD will not be able to grow at those rates. And even with those numbers, AMD would be paying back $36 out of every $100 they make for the whole year. Of course, we are assuming here that AMD isn’t starting saving for paying back this debt in preceding years.

AMD’s historical increases in gross revenue and profits have been very erratic so it’s difficult to know what the future will hold. If revenue grows at an average annual rate of 5%, that means 2012 revenues of $10.7 billion. In that case every other dollar (48%) of total revenue goes to repay the debt. And if we see a neutral growth, then $65 out of every $100 will be spent repaying it.

I am forced to question whether AMD will be able to do this, primarily because of what we’re seeing at Intel. The current price war has brought the average sales price (ASP) of processors down significantly. AMD doesn’t have the volume or capacity Intel does to compete with much lower margins. The past three quarters have seen financial losses in excess of $500 million dollars per quarter. So, how does AMD find a path back out of financial loss ?

AMD has made it clear they realize the negative extent of their financial position. The issuance of senior convertible notes twice in one year, while bringing in a much needed $3.7 billion cash infusion at a point and time when they desperately needed it, may have only delayed the inevitable. Intel is not sitting idly by waiting for Core 2 to run its course. They have a 2-year tick tock cycle in effect which, by the time it runs through its first complete cycle, will have taken AMD through the paces. And with Intel’s continued manufacturing advancements and on-time delivery of new products, AMD’s position continues to grow more and more tenuous.

Fabrication

Intel will release over a dozen 45nm products next year. Their 65nm technology today is very mature and 45nm is ramping without any reported issue. We were told by Dirk Meyer, AMD’s CTO, during the recent Technology Analyst Day, that the migration to 65 nm has brought some unexpected challenges. He assured us that AMD is working through them. But they do exist and this explains why Barcelona will not be seen above 2.0 GHz initially.

What we’re seeing hamper AMD here is not a lack of technical ability, because Barcelona obviously works. There is a limitation of manufacturing ability, and it seems very likely that the hampering is due to a lack of cash. More money means more parallel research which means faster adoption of the correct tools and techniques.

We know AMD has 65nm issues as it is backed up by two sources. First, the fact that no 65 nm AMD64 products meet or exceed AMD’s high-end 90 nm products in terms of speeds or compute ability. They only win out in power savings. This is unprecedented in AMD’s history as each new process technology has always seen faster clock speeds. The other source is Dirk Meyer himself. He said flatly that there were issues and that AMD was working through them.

We’ve heard reports that AMD is expecting to launch 2.3 GHz limited edition CPUs by the end of this year. That is a far cry from the original 2.5 GHz initial release which was slated for months ago.

Right now, Intel is continuing to perfect its high-k/metal gate solution in production 45 nm parts which we’ll see later this year. According to Intel’s press releases on high-k/metal gates, the new technology has the potential to reduce gate leakage by an order of magnitude (10x) over previous implementations. Intel’s first-run A0 silicon (produced in January of 2007) was able to boot Windows without issue. They were able to take first-run, raw silicon chips, package them as a CPU, install them in a motherboard and boot Windows. They did this within 3 hours after receiving the raw silicon. With no significant issues it was quite an accomplishment. And Intel has had months to tweak their already working design before its launch later this year.

Intel has done this at 45 nm with a new process technology while AMD is still struggling with traditional 65 nm SOI process technology.

No matter how you turn it, AMD will have to answer questions soon whether it makes so much sense taking its CPUs to 45 nm in 2008 and whether the company will be able to do so.

Capacity and cash

If we look at fabrication capacity and cash on hand, then I personally feel we could just about end this article right now. AMD has no dedicated 45 nm fabrication facilities. All of their 45 nm test projects will happen at Fab36 in Dresden, Germany. It is also their only 300 mm fab and this will remain that way until Fab 30 is revamped and converted to Fab38. And all of that revamping will come with significant expense, something on the order of $2.2 billion.

Today, Intel’s D1D research fab in Hillsboro, Oregon, serves as their first 300 mm 45nm production fab. Three more will go online by the end of 2008. One each in Arizona, Israel and New Mexico. Intel uses a "copy exactly" process for fabrication construction which allows yields and equipment to be identical. When they get a particular process technology working with yields and performance at rated goals, they exactly duplicate that fab someplace else. They literally duplicate everything. Factors like humidity, temperature, wall paint, floor tiles, wiring, A/C ducts, electricity routing, everything is the same. It truly is a "copy exactly" process, which enables the company to avoid surprises in yield.

This is why Intel’s manufacturing process can mature so quickly and easily across all fabs. Any fabrication facility which finds a newer/better way to increase production or yields can immediately spread that knowledge to all other facilities. And because everything is the same everywhere else, the new process will implement and scale without unexpected events.

Still, the point here is that by the time AMD reaches production 45 nm, Intel will have three 45 nm fabs cranking out an estimated 75,000 300 mm wafer starts per month. If we see an average die size of 175mm^2, that’s about 23.5 million CPUs per month (at a 90% yield). These will end up binned for the various clock speeds and performance goals (and not based on their true abilities as we’ve seen extreme overclocking on even 65 nm chips).

AMD cannot compete with those numbers, even with supplemental manufacturing capacity by Chartered Semiconductor on their certified 65 nm SOI process. It cannot be done even within a few years, and even with significantly more cash than they have today or are likely to have in the future. The truth is, capacity and cash weigh in heavily on this equation, and they fall against AMD.

Overclocking

Let’s look at what we know of Core 2’s overclocking abilities. The maximum frequencies seen today in production are officially are 3.0 GHz. We see Core 2 Duos overclocking on air to speeds in excess of 4.0 GHz right now, today on popular overclocking sites. This means without any additional cooling, today’s Core 2 Duo CPUs have at least 33% headroom. When the overclockers move to extreme cooling solutions, like liquid nitrogen or dry ice, we’ve seen overclocks to 6.0 GHz. A 65nm Cedar Mill Pentium 4 Model 631 was recently overclocked to 8.2 GHz ! (Sidenote : The Netburst architecture inside the P4 uses a double-pumped ALU, meaning it is internally computing some data at 16.4 GHz at that speed !) Even the Core 2 Quads, which are a mating of two separate dual-core CPUs, have seen overclocks in the 5.5 GHz range with liquid nitrogen. This overclocking requires both CPUs inside the single package to overclock at those levels. And whereas single-die dual-core CPUs are seeing 6.0 GHz, the impressive part comes from the dual-die dual-core CPUs mating together and still able to achieve 5.5 GHz or more.

If we look at AMD’s overclocking numbers we find much more sporadic higher-clocking results. Typically, AMD chips can overclock about 50% on extreme cooling solutions. This means that it’s overclocking around half the amount Intel’s chips are. This indicates how close AMD’s manufacturing process is to the limit of the technology.

The point here is this : Intel’s chips are highly overclockable. There is much headroom to move forward and compete against any advantage AMD might be able to throw at Intel due to Barcelona features. Barcelona may perform 20% better than Core 2, with an associated higher cost. All that means is that Intel will take their lower-priced CPUs they’re binning today for sales goals and re-bin them for higher clock speeds and increase their profits further while meeting AMD’s highest performance rating. Intel’s expenses have not gone up, and yet due to the overclockability on air seen today in Intel’s Core 2 Duos, they will be able to meet the performance increases of Intel through raw clockspeed increases alone.

And beyond the current overclockability of 65 nm parts comes the 45 nm parts. And then beyond that we must consider the next phase of the tick-tock cycle. Intel’s next generation CPU will be coming out before AMD has a chance to re-engineer anything. It will strike at AMD’s highest product line and may well be the capper for AMD proceeding autonomously. At that point, a buyout may be inevitable.

Options

AMD has mounting debt, consistent quarterly losses, unresolved manufacturing problems, a lack of significant cash on hand, they are capacity constrained, walking in uncharted waters with Fusion, Bobcat and Barcelona and they’re missing product deadlines while disclosing big future plans. All of this while Intel is railing them from all sides with its tick tock strategy after hitting defined technology milestones and accomplishments, one after another.

AMD doesn’t have much choice. They need something to fall into their court, they need a favorable court ruling in their multi-jurisdiction anti-trust case against Intel, or they need to be bought out by another company. AMD cannot compete against Intel’s resources, capacity and technology without something significant happening.

Change in history ? If the courts were to find Intel did illegally impede AMD’s ability to sell products, then significant punitive damages could be imposed on the chip giant. These could benefit AMD both financially and technologically as Intel could potentially be constrained for a period of time from manufacturing more than a particular number of x86-based processors, for example. Or, they would have to give AMD a huge cash payout which would provide AMD with what they need most : more money.

The other possibility is a buyout. I’ve had several discussions with people about buyout conditions. Any company increasing its long-term debt notably becomes less attractive to potential buyers. There has also been some discussion that this might be part of AMD’s plan. By making themselves less attractive to buyout, they are moving forward with their "all in" philosophy. They are of a "make it or break it" mindset. It may also be AMD’s plan to continue selling senior convertible notes every few quarters to raise additional cash that they’re not getting from revenues. These are all possibilities and speculation, but what AMD needs most right now is cash.

There has been quite a bit of speculation recently about a private investment group or a company like IBM or Samsung buying AMD. The latter would provide enough cash and capacity to really drive AMD’s product vision forward. With the mounting debt, I believe the possibility becomes more complicated and less likely.

Conclusion

No doubt. Intel will surpass AMD in all measurable ways : Performance, manufacturing technology, capacity, execution, profitability and long-term product feasibility. In my opinion, AMD has better goals for x86 than Intel does. They’ve listened to the market and brought us more desirable products and visions, including a non-proprietary path to 64-bits, cross-technology fusion-like initiatives, open-source multi-port HyperTransport, NUMA, native quad-cores, multiple power planes and more. They’re literally the Chrysler of the x86-based semiconductor industry in terms of feature sets (and not just manufacturing abilities). They continue to push ideas and visions forward, many of which Intel seemingly later grabs and runs with.

It looks to us. Intel simply has so much capacity and cash that they will win. AMD’s future appears to be one of borrowed time. Without a court ruling, a buyout, or something truly unprecedented happening like Intel having a major flaw in their upcoming processors, then I don’t see any way AMD can make it otherwise.

Reader Contribution

These are my opinions and they come from much analysis and consideration of AMD. I welcome everyone to post their responses or opinions on this matter. If there are aspects I haven’t considered above, I’ll be more than happy to give them a visit and update this article. Just post your comments below and we’ll see how this thing plays out.