In 2009, the European Commission found that Intel had abused its market position with respect to AMD and slapped the company with a €1.06B ($1.45B at 2009 exchange rates) fine. In 2014, the General Court of the European Union upheld the judgment. In 2017, however, Intel won relief from the European Court of Justice, which ordered a new trial. That’s where we are today.
Intel is once again making the case that its exclusivity rebates had no impact on AMD, despite the fact that they placed steep restrictions on how many AMD systems could be sold. Intel’s counter-argument is that these deals were ones the OEMs could walk away from, that they only covered a fraction of the market, that AMD was supply limited, and that nothing in EU law states that exclusivity rebates are automatically anti-competitive.
What Happened in 2017?
When evaluating whether a company has illegally restricted competition, the EU applies what’s known as AEC (As Efficient Competitor) test. Loosely speaking — because I am not an expert in EU law — the AEC test is intended to measure whether an “as-efficient competitor” that would otherwise be competing effectively with the dominant company is being shoved out of the market by anti-competitive practices.
Back in 2009, the EC found that Intel’s behavior was so self-evidently abusive, and its practices so filled with “naked restrictions” on AMD, it did not need to perform a full AEC test but provided one anyway.
When Intel appealed the case to the General Court in 2014, it raised specific problems with the EC’s AEC analysis and argued that the analysis had been performed incorrectly. In essence, Intel took the position that other factors, including AMD’s factually constrained CPU
production, were the reason for the company’s low market share.
The General Court, however, “attached no importance” to the EC AEC report, precisely because the document was not a formal finding of fact. Because the General Court chose not to consider the AEC analysis as evidence, it also chose not to consider Intel’s arguments about the specific problems with the AEC as evidence.
This was not a breach of protocol or legal precedent. Prior to the Intel ruling in 2017, there was no requirement that a General Court perform an analysis of company-provided data regarding whether it had breached antitrust law. The EC had provided the document voluntarily.
Prior to 2017, the use of exclusivity rebates to retain customer loyalty was apparently treated as prima facie evidence of anti-competitive behavior. The EC also argued that Intel’s criticisms of its AEC report amounted to an attempt to contest the findings of fact already established in the original case and rejected them accordingly.
The text below is from the 2017 decision. In all cases, the Commission is the European Commission that handled the initial investigation, the General Court is the body Intel appealed to.
In this case, while the Commission emphasised, in the decision at issue, that the rebates at issue were by their very nature capable of restricting competition such that an analysis of all the circumstances of the case and, in particular, an AEC test were not necessary in order to find an abuse of a dominant position… it nevertheless carried out an in-depth examination of those circumstances, setting out, a very detailed analysis of the AEC test, which led it to conclude… that an as efficient competitor would have had to offer prices which would not have been viable and that, accordingly, the rebate scheme at issue was capable of having foreclosure effects on such a competitor.
The ellipses above reflect where I removed paragraph reference numbers to make the text easier to read.
In 2017, all of this changed. The ECJ ruling found that the General Court had erred in failing to consider Intel’s arguments regarding the AEC report that the court had chosen not to treat as evidence and sent the case back to the lower court for a re-hearing on Intel’s arguments.
The decision was seen as significant. Norton Rose Fullbright writes:
Although the Commission did in fact make an effects-based assessment, the General Court had not assessed Intel’s arguments that the Commission’s analysis was defective. This is the first time that the CJEU has required an effects-based analysis in an exclusivity rebate case.
Present Day: Intel May Not Want to Win This Case
Back here in 2020, Intel is arguing essentially the same point it raised in 2017, that the AEC test was badly performed and that it wasn’t actually harming AMD. As evidence, Intel points to the fact that Dell adopted AMD CPUs when rebates were at an all-time high. They aren’t wrong. Dell was fined a great deal of money for failing to report its finances properly with respect to Intel’s rebate payments.
If Intel wins this case, it will likely weaken EU antitrust law. The old standard of analysis emphasized the structure and nature of illegal agreements and did not require a demonstration of competitive harm, though the companies still had to be genuine competitors to one another. The EJC ruling did not exonerate Intel on the matter of its rebates, even though it ordered a retrial.
This may be a good time to note that this investigation is 18 years old. If it were a human, it’d be old enough to vote.
Here’s the problem. Forcing the EC to perform a full economic analysis of whatever data a huge tech company provides is going to make it significantly harder to investigate these companies at all. Substantially raising the burden of proof on the EC means its efforts to rein in Google and Qualcomm are more likely to fail.
Intel may have won its Qualcomm case in the US, but the EU investigation of Qualcomm is still ongoing. What Qualcomm did to Intel is conceptually very similar to the way the EU found Intel had harmed AMD. Intel might be shooting itself in the foot with regards to its own Qualcomm case.
As for whether the EJC made the right call? I genuinely don’t feel as though I know enough about how the EU’s justice system works to have an opinion. I typically favor giving companies and individuals the right to challenge a judgment and to have that challenge heard, but I don’t want to see that right turned into a cudgel that companies can wield to slow or overwhelm lawful investigations. Any change in the law that substantially increases the burden of proof on the EC should be matched by additional hiring and a larger budget.
Did Intel Actually Harm AMD?
AMD’s original lawsuit against Intel in the United States is, quite frankly, a damning portrayal of market abuse. While the 2005 filing focused on the US market rather than Europe, Intel’s programs to restrict AMD operated in multiple markets. Here’s one example of how much control Intel allegedly wielded:
Following two years of negotiation, Supermicro finally agreed last year to begin developing an Opteron-powered server, however, it so feared Intel retaliation that it secretly moved the AMD development to quarters behind Supermicro’s main manufacturing facility. Further, it forbade AMD from publicizing the product or beginning any marketing prior to its actual release. When, in April 2005, Supermicro finally broke away from years of exclusivity, it restricted distribution of its newly-released Opteron products to only sixty of its customers and promoted them with a glossy, upscale brochure devoid of its name and labeled “secret and confidential.”
I can partially confirm this story. When I heard the first AMD Supermicro motherboard was being released, I searched Supermicro’s website and couldn’t find it. If you punched the model number into Supermicro’s own search engine, it would return no results. The only way to find a link to the motherboard was to use Google to search the site. The box and motherboard were unbranded.
I called the company to ask about the board and was told Supermicro focused on Intel products. It wasn’t until I mentioned the specific model number that the company representative suddenly knew what I was talking about. Even then, she was cagey, and only barely willing to confirm the motherboard existed.
There’s also the alleged HP chip debacle. To make a long story short, AMD once offered to give HP with one million free processors in order to break Intel’s hold on AMD. Here’s the 2005 filing:
On the eve of the launch, HP disclosed its plan to Intel, which told HP it considered AMD’s entry into HP’s commercial line a “Richter 10” event. It immediately pressured HP into (1) withdrawing the AMD offering from its premier “Evo” brand and 2). withholding the AMD-powered computer from HP’s network of independent value-added resellers, the HP’s principal point of access to small business users for whom the computer was designed in the first place. Intel went so far as to pressure HP’s senior management to consider firing the HP executive who spearheaded the AMD commercial proposal. As a result of Intel’s coercion, the HP-AMD desktop offering was dead on arrival.
This paragraph, ultimately, was at the core of the Intel / AMD antitrust case. Intel’s response to these issues was often to argue that AMD capacity-constrained and therefore incapable of providing the entire market with enough CPUs (which was and is factually true).
But AMD had a potent counter to that argument. Intel’s exclusive rebate system didn’t just restrict AMD’s market share, it restricted the markets AMD was able to sell parts into and the prices it could charge. A million free processors are a lot of chips — but HP couldn’t afford to lose Intel’s rebates. Even if Intel adjusted its rebate policies to specifically account for AMD’s total manufacturing capacity, it was still allegedly impacting AMD’s revenue by mostly confining its CPUs to the desktop market. In AMD’s telling, Opteron took major market share in servers in spite of Intel’s active opposition.
AMD and Intel eventually settled their case, with Intel paying AMD $1.25B and renegotiating the x86 license terms. The allegations in AMD’s filings against Intel were never adjudicated, but the EJC affirmed that Intel’s exclusivity rebates were an intrinsic abuse of power in 2017:
In that regard, the Court has already held that an undertaking which is in a dominant position on a market and ties purchasers — even if it does so at their request — by an obligation or promise on their part to obtain all or most of their requirements exclusively from that undertaking abuses its dominant position within the meaning of Article 102 TFEU, whether the obligation is stipulated without further qualification or whether it is undertaken in consideration of the grant of a rebate.
Allow me to translate: Offering exclusivity rebates is an abuse of power within the meaning of EU law. Period. Nevertheless, the EC is now required to fully examine Intel’s line of reasoning and consider whether it has merit.
Do I think some of Intel’s actions harmed AMD? Yes. Do I know if they would have been found guilty of antitrust violations in a court of law, especially in the United States, which has different and steeper requirements for these kinds of cases? No. Do I think AMD was also responsible for some of its own problems during this time period? Yes. Where does that leave us? Debating an investigation that’s old enough to vote.
The funny thing is, I’m not sure AMD actually got the short end of the stick. I’m sure the $1.25B was nice, but the renegotiated x86 licensing terms are what allowed AMD to spin off its foundries, something it was contractually forbidden to do before.
Historically, AMD had trouble with new nodes. After beating Intel to 180nm, AMD lagged on 130nm and 90nm. By 45nm, it was a year behind Intel. It’s not clear that AMD could have made the jump to FinFETs within a reasonable period of time if it had retained its own fabs. One of the reasons AMD’s fab transitions lagged Intel is because AMD had to balance capacity constraints and bringing up a new node, much as Intel is doing now.
I’ve taken the time to unpack both the court case findings and AMD’s original allegations is because I think it’s worth understanding the subtleties of the case. The EJC didn’t give Intel a clean bill of health in 2017 and it didn’t examine all of Intel’s legal claims. Intel will now have a chance to present its own data asserting that AMD was not harmed by its rebate practices. I’m in favor of that in principle but wary of the impact in practice — the 737 MAX debacle can be traced, in part, to the FAA’s decision to allow manufacturers to self-certify their own planes. The FAA took that step because it didn’t have the resources or manpower to continue its previous level of oversight.
Intel’s past treatment of AMD tends to be a pretty hot topic, and certain issues, like compiler optimizations, still come up in the present day. I am curious to see how the ruling goes and what evidence Intel will present, but I’m also pretty ready to move on from the battles of 2005. It’s a new era. If you look at the list of executives at Intel in 2005 and compare it with 2020, it’s pretty different. The same people aren’t running the company. While AMD has performed exceptionally well since 2017, there’s no denying that Intel has also substantially improved its prices and overall performance.
Chances are good that this case isn’t over, no matter how the EC rules. It’ll make its way back up the food chain to the ECJ one way or another. Might be time to start planning its 21st birthday party.
Now Read:
- AMD Analyst Day 2020: Zen 3, Infinity Fabric 3, and 3D Packaging
- AMD Announces CDNA, RDNA2 Architectures, Significant Leap in Performance-per-Watt
- From 4.3GHz All-Core Overclocking to SMT Scaling: A Comprehensive Review of the AMD Threadripper 3990X
No comments:
Post a Comment