In coming months, Intel has promised to break out financials for Intel Foundry Services (IFS) its fab operations and nascent foundry business. As soon as we heard about this we started to think about what that model would look like. We imagined a deep sea of red ink, this business is capital intensive and currently only has a tiny amount of external revenue. Here we want to take a stab at thinking through that model a bit more rigorously. Ultimately, there are too many unknowable variables, so we offer up this analysis as a way to gauge the trade-offs the company makes when it ultimately reveals its numbers.
The first step is to try to tease out IFS revenue. In theory, this should equate to Intel’s cost of good sold today. IFS’s chief customer will be Intel Products, the design of the business. A few things upset this naive equation. First, not all of COGS is necessarily wafer costs. It can include channel payments, field support, warranties and a a host of other charges. Moreover, the company has a fair amount of discretion in allocating costs between the two side of the business. We have no ready way of calculating this, so we just made an estimate of 20% of COGS remains on the product side.
A second factor is that Intel generates a fair amount of revenue from parts fabbed at TSMC. This includes products in its PSG group (aka Altera), Mobileye, and a growing amount of PC CPUs. Fortunately, we can factor those out. After some trolling around on the Internet we found a sell-side report from JP Morgan which broke out their estimate of TSMC revenue from Intel. These work out to roughly 5% of 2023 Intel revenue, growing to 10% next year as new Intel CPUs ramp at TSMC.
Adding all that up, we get to IFS revenue in 2024 as $23 billion on Intel revenue (from consensus) of $57 billion.
Next comes gross margins. This is the major swing factor in these results, and there are a wide array of methods for getting this figure. Calculating semiconductor gross margins is tricky. Some of it is straightforward – wafers produced multiplied by cost per wafer. Then there is the significant matter of depreciation by which all fab owners’ live and die. But how do we allocate that depreciation as well as a host of other overhead?
In terms of wafers, we think Intel has capacity to produce about 800,000 wafers per month, but utilization of these is probably still fairly low, around 80% (a guess). If each wafer costs $1,200 (another guess), we end up with direct wafer costs of about $9.2 billion a year in 2024. Then we have to factor in depreciation. Last year depreciation was about 24% of total company COGS. If we use that as a proxy for capex this year, we get $8.2 billion in depreciation.
Tack on another 10% for miscellaneous expenses and we get IFS gross profits of $7.1 billion in 2024, or 24% gross margins.
We then made some estimates as IFS’s operating expenses. We looked at current Intel opex and made estimates as to how much of those to attribute to the fabs – R&D a high percentage, SG&A much lower.
Netting all this out leaves us with IFS losing $5 billion in 2024 and $7 billion in 2025. The numbers increase because In this period IFS will be losing Intel Product share to TSMC just as depreciation spikes, reflecting the costs of the company’s five nodes in four years program. Assuming the company reaches stability in manufacturing soon, 2025 should be the bottom and IFS should turn profitable by 2027. We chose not to model that far out because the variables become too big – how competitive will the 18A manufacturing process be? How competitive will Intel’s design products be by then? How many external customers will sign up to IFS? We are comfortable that 18A will be competitive with TSMC. We are decidedly mixed about Intel’s CPUs in that time frame. And we strongly question how quickly external customers will become receptive to IFS. But the range of outcomes is too broad to model at this point. We will return to all of these topics in future posts.
That being said, we see these IFS numbers as closer to the worst case scenario than what the company will report. Under our model, IFS has been loss making for years. We think that means we are over-allocating a fair-sized portion of costs to IFS that should be borne by the product side. As noted, ultimately those decisions will get made by Intel. And while they will obviously follow accounting principles, that still leaves them with considerable leeway.
At this point, we do not think many people will be surprised by this level of losses. The company has telegraphed big losses are in the works. The company’s latest moves to push costs for hot lots and the like back to the product teams on the design team likely mean those really large numbers above will not materialize. Moreover, we would caution investors to not get overly concerned about this level of losses. The company is on a multi-year transformation, this was always going to be expensive. And assuming the design side of Intel can stabilize its share position in PCs and the data center, the whole company should remain fairly profitable. The real challenge comes later in this decade by which time IFS will need to demonstrate it can attract outside customers. This is a long process and it is too early to judge progress. So be forewarned, IFS numbers are going to be hard to swallow now, but hopefully point to a better picture down the road.
Here is our model, please e-mail us if you would like a working version.
