Primeur Magazine: You recently published four new reports. Can you just give a few first impressions of your findings?
Addison Snell:Sure, Intersect360 Research first of all consists of industry analysts who cover high performance computing and hyperscale markets. This includes major trends like machine learning or artificial intelligence, data science, Cloud but to the extent that they transcend all of those areas. What we are really talking about here is the high performance computing portion of the market. Intersect360 Research is happy to report that in 2019, it showed healthy growth. It was up 8,2 percent which was a point better than we had originally forecast. It is about 39 billion dollars (about 33 billion euro) worldwide.
Now, the hardware on-premise portions of that, the biggest ones in terms of servers and storage, showed much more modest growth between 2 and 3 percent. Where the big growth really came in the market was of course in Cloud computing types of deployments which have been growing at double-digit rates. These are people who are using Cloud computing or public Cloud computing to run their high performance computing applications. We also saw a big spike in what we could call Cloud-like sorts of deployments that aren't strictly public Cloud but we are talking about things like managed services, contracts, facilities, outsourcing, other types of utility or outsourcing that aren't strictly public Cloud. Those were also well up in the 2019 numbers and contributed to that solid growth rate.
Primeur Magazine: Does that mean that in the end everything will be Cloud or is there still a market for the on-premise HPC in the future?
Addison Snell:Yes, certainly not everything is moving to Cloud. High performance computing is still solidly, predominantly an on-premise market. However, we have seen a very rapid growth in Cloud, starting with bursting types of applications but I think there will be a long-term role for Cloud particularly as a playground for new applications, new workloads and new technologies, and matching those together. I think organisations will find a way to use Cloud as a way to figure out where they want to go directionally. Just economically, most of high performance computing tends to run at a fairly high utilization rate on-premise. These are sophisticated users who know what they are doing. Economically, for the majority of the market, it still works out better for that to stay on-premise. Cloud is a portion of the market. It is growing very rapidly. It will take up a larger portion of the market but this will still be predominantly on-premise throughout the forecast.
Primeur Magazine: So, that were the findings for 2019. In 2020, we saw one very positive thing in high performance computing: the Fugaku supercomputer which is a marvelous machine. But apart from that, we also saw the whole high performance computing market more or less collapsed like a lot of other things.
Addison Snell:Well, 2020 is still a forecast of course. We will have to wait till the end of the year to really total up the end of the year numbers. I do want to spend a moment and talk about Fugaku. What a marvelous machine. Fujitsu has been on the top of the TOP500 list a couple of times now, and this Fugaku machine at RIKEN is really an outstanding achievement. It uses a very interesting A64 FX Arm-based processor which is the first time we have really seen that at this level of scale. It uses the Tofu interconnect from Fujitsu. I think there are going to be a lot of eyes on this machine in terms of how efficient it is for high performance computing applications. It was a real achievement for them to deliver that ahead of schedule, so that those cycles could be used for COVID-19 research which there is a great need for in the scientific computing space right now.
This bridges us to the other major point about 2020 which is that here we are in the middle of a global pandemic and that is what's really got the forecast gone in an interesting direction right now. It has had a tremendous effect on the worldwide economy and that will have an effect on high performance computing spending this year. We are now forecasting a decline of 3,7 percent year over year. Now that is not just a 3,7 percent loss but we really have to look at the fact that it is down from the growth that we would have originally forecast. It was in the neighbourhood of about 7 percent, so this is a much greater loss than that to say you are going to wipe out all of that growth. We are actually going to see the first loss in this market since the last recession period in 2009.
Primeur Magazine: When I read your report you claim that there is a decline now but the market will catch up and eventually in 2024 or something like that, we will be back on track, so we will be at the level that we had before the crisis. Is that correct?
Addison Snell:Right, overall there has been very little effect on the long-term compound annual growth rate or the health of the market. That is because we are monitoring that essentially the underlying demand for scientific and technical computing isn't going anywhere. The majority of spending this year that does not happen, most of that is deferred rather than delayed. The first thing that happens is, while we have it down 2020, then there will be a rebound spike in 2021 and what looks like this tremendously high growth rate, but a lot of that are just purchases or acquisitions that were delayed from 2020 into 2021. Then we could even see a little bit of an echo effect in 2022, being down a bit, and 2023 up a bit. This just has to do with purchase cycles, budget cycles, before it evens out again by the end of the five-year forecast and we get pretty much back to where we would have been.
Now, that has not to say there is no real loss because when I was describing those troughs or valleys dips in the market followed by the spikes in the market, what happens with that is the troughs are deeper than the peaks are high. So, there is real loss that goes on. There are some purchases that don't come back but in terms of the long-term growth rate, there is very little effect. By the time we get to the end of the five-year forecast we are about back where we would have been. This is something we have seen before, when we look back at our numbers from 2009 and the five years that followed. That is pretty much the pattern that we saw there that we were able to measure. We expect to see the same thing again.
I will say one more quick thing. There is a quick caveat which is that it is hard for us as HPC analysts to forecast a pandemic. We are not epidemiologists. Our forecast premise is based on the idea that the economic effects of the pandemic are primarily constrained to this calendar year and that by 2021, early 2022 at the least, we get back to more normal business spending, even if the shape of some of it changes. If that doesn't happen and we are still seeing the pandemic having a major economic effect in 2021, then we are going to have to rip up the forecast and go do a new one again.
Primeur Magazine: But then you assume that there will be no major change in the demand part of the market and that we will be back at the old expected level?
Addison Snell:That is right.
Primeur Magazine: But in previous crises, there was more a supplier side crisis, or an overall crisis. This time it is perhaps different. We could for instance say that the airline industry is staying very low and will not recover. That could have an impact on engineering industry and hence also on the need for HPC systems, a change in the demand side. There are also other industries like that, so are you taking that into account?
Addison Snell:That is right. Right now, we don't. I suppose that if you say we are going to wipe out the airline industry and people are not going to fly or they are not going to fly to the extent they did before, then that could have a deleterious effect where you are taking out HPC spending from the aerospace industry. At this point, we think that mostly people return to flying, people return to buying cars, people return to needing gas for their cars, which is the energy market, people return to spending money in retail in some level. Now, that could change. They could be spending more online and less on brick and mortar. But the HPC that goes into the analytics of retail is still there. Some of it might shift more to Cloud and less to on-premise. That could happen but really, when we look vertical market by vertical market or workload by workload, right now it looks like there is not any sort of long-term impact that would really wipe those out.
Now, I mentioned Cloud and I will say one other thing quickly about Cloud. This is that Cloud, we think, is going to have the reverse effect of the other sorts of spending that we are monitoring because when you constrain on-premise spending, Cloud is something that can absorb a lot of that gap. When everything else goes down, Cloud will go up to an extent. Not to where it fills in the entire gap but Cloud will follow the reverse trend from everything else because of the nature of Cloud to be flexible in times of uncertainty. That will be an interesting effect that the Cloud spending should actually be up in 2020 while everything else is down.
Primeur Magazine: Yes, we understand it is very difficult to make predictions in these times.
Addison Snell:That is what we do.
Primeur Magazine: That is why we are talking because you make predictions. One other thing which we would like to discuss is the following. We mentioned the Fugaku already but there are other exascale programmes in the US and in Europe. They all try to to have an impact on the development of high performance computing but the question is: do they have a real influence in the sense that the complete market is affected in a positive way? Or do they just have a minor ripple effect on the industry? Do they really help, these exascale support programmes in Japan, the US, China, and Europe?
Addison Snell:They do have influence that goes beyond the spending on those systems themselves. Now, the race to exascale of course does put real money into the market in terms of those supercomputers that get bought and deployed and we count those. But then, there is a halo effect that goes through the rest of the market. One thing that we have noticed that started in 2019, is that government spending really came back to the forefront, leading the growth rate by sector ahead even of commercial. That ended a string of six straight years in which the commercial markets had the highest growth rate. Before that, it was academia.
We have really come through years of relative government austerity in multiple geographies. US government spending was constrained; European or Eurozone spending was constrained for many years. We had similar constraints in Japan, in China. What we are really seeing that in part is related to the race to exascale, in part is related to artificial intelligence, is that really a lot of that government spending is coming back. We are now forecasting that over the next five year period it is going to be the government sector that has the highest growth rate over that five year period. It will also be more stable relative to the commercial spending which is going to have a lot more of the variability due to COVID-19. Government will have maybe some variability but really should be much more stable through the pandemic. Out of all the commercial segments the one that will be the most stable is biosciences related to pharmaceuticals and other sorts of research which of course is getting a lot of attention during the pandemic. Generally, what we will see is that government will be more stable and have a higher growth rate. The commercial sector will still have a good growth rate but will have a lot of variability as a result of the pandemic.
Primeur Magazine: Did we discuss all the major findings and predictions of your survey and of your reports?
Addison Snell:Well, there are many other things that I could go on highlighting and the great thing with high performance computing is that I could talk about it all day. It is a fascinating area. I think the big thing I would underline is that the need for high performance computing doesn't go away. The great thing about science is that we don't get to the end of it. You solve one problem and there is another problem on the horizon that leads to engineering and other forms of innovation. This is a long-term stable market because we do not run out of problems to solve. Beyond that, we are adding in data science and machine learning, everything with artificial intelligence, so really, it are those underlying demand side growth factors that lead us to conclude that the market comes back. There is going to be a need for this type of computing. That is what keeps us busy as analysts. It is an exciting time to be in high performance computing.
Primeur Magazine: Thanks very much for this interview. I surely hope we can meet at some real live conference.
Addison Snell:I hope I will see you back on the conference circuit maybe in 2021. Thanks for having me on your screen.