Intel has been a stock I have hated forever. The thing that I hated was a very simple accounting gimmick they have kept playing for a very long time – their depreciation has been running behind capex for over 5-6 years. This was a pet-peeve and a reason why I didn’t bother investing from its ascent from the mid-20s to mid-50s.
When Nvidia came around, with their Q4 2015 results, it was obvious they were onto beating big time for a long time. This doesn’t happen that often. I remember vividly that AlphaGo had just beaten the world champion in Go. That the two were connected in the long term was very likely.
This made Intel look even less attractive, especially if one considered the market for data center logic chips to be the $45-50B it had been for a very long time. One would make the obvious share of wallet argument and that Intel was about to lose big time to NVDA.
I got that wrong. I assume many others got that wrong too. The market for logic chips grew. Some argue that Intel continued to capture a majority share of those expanded dollars. All that training required host compute (x86) too. Artificial Intelligence workloads are a big tailwind for all times of logic chips – ASICs, GPUs, FPGAs, and x86. (and possible ARM, we will find out).
Last 6 months
A couple of years ago, TSMC caught up with Intel after decades of trailing it. Recently, it overtook Intel. And in July, after Intel announced another round of delays to its 7nm process, it now leads with almost a 2-year advantage. That’s how much ahead of TSMC Intel used to be.
Another way of saying above is that over the almost 4 years or so, Intel as an organization stood still. It didn’t progress. Is that an organizational problem? I don’t know. Can that be fixed? I don’t know.
The third quarter they reported showed weakness in two important areas: their DC business and their margins. Both of them are scary, given both of them are important drivers of value for Intel. Their decline represents big changes to expected future growth and future profitability. What caused these declines?
Revenues were down 10%, and guided down 25% for 4Q20. The idea that we have pulled forward a lot of cloud spend into 1H and thus 2H is weak sounds credible. The idea suggests that Cloud spending is lumpy. Sure. But when we look back 5 years, it seems to have been increasingly lumpy over last two years.
Why should DC business become lumpier? It cannot be competition alone, because AMD is still just a 5% market share player (they just reported a quarter with DCG revenues up c100% yoy and qoq, helped by the console launches and possibly some more share gain in DC). Beyond that, is it just the law of large numbers? That the amount spent by hyperscalers is just so large that it has started to behave in a lumpy fashion like any other CAPEX item. Note, the peaks are also higher. If you have an alternative explanation, please share it with me.
This, I think, is industry-wide and not just an Intel issue. WDC, STX, and ANET who are all exposed to cloud CAPEX have all fared poorly over the past 3 months. The timing of which bucket hits when (networking, storage, compute) is always tricky to call. But, we have seen them all affected.
Data Centric Margins
The other question / risk is that Intel is seeing competitive pressure, deleverage and mix hits from ramping 10nm.
I dont think deleverage in revenues explains $1200M loss in op profits yoy for a $500m drop in sales yoy. That leaves us with competitive pressure. Here again, I dont think AMD with its 5% market share could cause a 6pt yoy decline in DC margins. It doesnt make sense for Intel do make that trade-off to keep share. Plus most likely, AMD is supply constrained from sharing the TSMC capacity with Apple and NVDA, all of whom are seeing a new product cycle. So the threat of AMD taking share from Intel at a rapid pace is unlikely. I do believe AMD is and will continue to take share, but just not at such a rapid pace that would Intel to cut pricing to the extent of seeing a 6pt yoy decline in margin.
What we are left with is a mix-effect from ramping the 10nm process. Every new process when ramped has poorer margins due to poor yield. This is business as usual. This yield then normalizes over the next 3-6 months and margins recover. The question that we are left with is whether Intel is having to ramp a process sooner than it would otherwise? (because of AMD/TSMC?) Again, I don’t know.
That Intel chose not to share any details on their progress in manufacturing and future plans adds to the uncertainty in answering the above question. We will find that out early next year.
It is this question that is key, because it will decide whether, in the first half of 2021, Intel blows up again or recovers rapidly. The answer depends on whether the organization is dysfunctional, thus competitively cornered and in reaction mode. Or, will the organization be able to recover from missteps of the past several years, come together and recover its position.
Beyond that, I find myself wondering if I am again missing the forest for the trees. Demand for logic chips is being powered by demand for AI, in the data centre and increasingly in the edge. The market is growing. Intel is a big part of the supply of those chips and if that supply is slowing down, doesn’t price go up? Can TSMC alone handle all the slack if Intel falls behind? Is that geopolitically speaking possible?