How TSMC is able to spend $28B in capex while making only $18B? And how Amazon took it to a different level?
Another point in the fantastic podcast about Morris & TSMC by Ben Gilbert and David Rosenthal at Acquired that could be a bit better thought through was the question about how TSMC plans to finance its capex.
Back in April, TSMC announced that it would spend at least $100B in capex, with $28B in the year of 2021 alone. In fiscal year 2020, TSMC reported a net income of TWD$518B, which is about $18.66B. Toward the end of the podcast, Ben and David quoted a profit number of $18B (if I remember correctly) and asked the question how TSMC would be able to finance $28B of capex with $18B of income. They bantered away without an answer to that.
Granted it was a fleeting question that probably was not worth digging in too much in the context of this already brilliant podcast. The audience won’t know more about TSMC and Morris than necessary by learning about the answer to this question. Still it’s worth pointing out that this is a common misunderstanding for people who are less from accounting or finance background.
Let’s jump to the conclusion: yes, TSMC can finance $30B capex despite only a $18B of income without external financing.
One cannot spend “accounting profit”
Accounting 101: accounting profit is not cash. One can only spend cash. One cannot spend accounting profit.
To illustrate the point, imagine a company making $100M revenue in a year at 50% gross margin. The gross profit is then $50M. After an operating cost, which includes R&D (Research & Development), SG&A (Selling, General and Administrative Expenses) and D&A (Depreciation & Amortization) that’s not tied directly to operation, of $30M, one arrives at a net income before tax of $20M. We’ll ignore tax for simplicity.
In the most simple scenario, all $100M of revenue was collected in cash while corresponding products/services have already been delivered to clients in full. All $70M expenses (cost of goods sold & operating cost) have also been paid out in cash in full. In this case the $20M net income is pure cash. The company can spend that $20M for capex next year (if it chooses not to issue dividends).
But this is never the case of a real company. There’s usually a time difference between recognizing an accounting item and the actual cash flow associated with it.
For examples, it might take your clients up to 30 days to pay you in cash for products/services that you have delivered. Assuming the clients are seen by your accountant as with reasonable credits, the revenues associated with this transaction will be booked as revenues — despite the fact that there’s $0 in your bank associated with it.
In accounting this is booked as account receivables on the balance sheet. VCs more used to SaaS companies would not care too much about this item since SaaS clients pay monthly or annually, with cash almost always coming before the services are actually delivered. In non-SaaS businesses you almost always have account receivables that are non-negligible on your balance sheets. This year with the supply chain disruption many of the hardware companies are looking at a ballooning account receivables on their balance sheets with wary eyes as their clients in turn have trouble getting cash payments from their own clients that are disrupted by Covid or simply lack of containers.
Let’s say the aforementioned company with $100M of revenue but only $80M of that has been collected in cash. This actually leaves the company with $0 cash on hand despite a $20M accounting profit. It will have a $20M account receivables on its balance sheet but won’t be able to make any capex until the it collects the cash from clients and converts that account receivables into cash.
To recap, accounting profit is not cash. It’s just a number in a highly stylized table called Income Statement, which most financial analysts only take a fleeting look before they move on to digging dirts in the balance sheet and cash flow statements — especially in the footnotes.
One can spend more than accounting profit
What, Jerry? Didn’t you just say that one cannot spend accounting profit since it’s not cash? How come you’re saying now that one can spend more than accounting profit? Aren’t you contradicting yourself?
No, I’m not. The fact that one cannot spend accounting profit doesn’t mean that one does not have the cash to spend. Cash on the balance sheet, which one can spend, is totally different from the accounting profit on the income statement, which you cannot spend. The two over time are interlinked but at any given snapshot of the history, are not that correlated.
In the case of TSMC, it’s true that it wanted to spend $28B in 2021 while having reported only $18.66B of net income in FY2020. But if we look at cash flow statement (see above), TSMC reported a Net Operating Cash Flow (after paying Income Tax) of TWD$823B (marked yellow in the table above) in FY2020, which is about $30B.
Now you can definitely use that $30B cash generated from operating activities to pay for $28B capex (and other things)
But how does TSMC generate $28B of cash from operation while reporting only $18.66B in accounting profit? Are they cooking the book?
Of course they’re not cooking the book. Actually what TSMC’s numbers tell is a typical story of high growth / high capex businesses.
One of the key signatures in the semiconductor foundry business in the modern days is the depreciation of the most expensive equipments over 4 years, especially those EUV lithography equipments made by ASML that costs easily $100M each.
That’s not to say that these machines will break down after 4 years and won’t be able to generate revenues. It’s simply an assumption that every generation of lithography equipments will no longer be operationally profitable after 4 years, as newer generations of equipments come int to produce larger, denser, more powerful and therefore more valuable chips, which in turn squeeze the profitability of the previous generations.
It is basically the accounting version of the famous Moore’s Law.
It’s a way to keep the foundry management teams and investors highly alerted of the paranoid competitive nature of the fab/foundry businesses — If a fab can still make profit after 4 years and fully depreciated, good to know! But don’t count on it and better make all the money you want in 4 years!
Now coming back to TSMC. In the CF (cash flow) statement above, one finds that the biggest non-cash expense to be added back to the net income to arrive at the OCF (operating cash flow) to be Depreciation expense, worth a total of TWD$324B.
This depreciation expense is basically a mix of all the new and remaining depreciation expenses of the equipments that TSMC paid cash for in the previous 4 years. The cash was already out of the door in the previous 4 years, but the depreciation still impacts the net profit directly and in a significant way.
In fact, the depreciations of foundry equipments is so key to the business that companies recognize them in the cost of revenue:
As seen above, out of the TWD$324B depreciation in FY2020, TWD$299B was recognized in cost of revenue, which is almost half of the total cost of revenue TWD$628B!
But mathematically how does it work? Can a company continue to generate more annual operating cash flow than accounting profit?
The answer is yes. For those that have read my previous article “TSMC at $0 Pre-Money?”, you probably remember that TSMC hasn’t done any capital increase since 2000. In the past 20 years TSMC has been growing revenues and profits and escalating capex year over year — all with the cash generated from the business.
Mathematically, a quick back-of-the-envelope calculation would tell us that in FY2020 the total depreciation is about 24% of total revenues (TWD$1,339B). Now while the absolute amount of (cash) capex and therefore (accounting) depreciation will grow every, as long as TSMC continues to be super profitable (FY2020 gross margin at 53%) it will be able to finance its capex very comfortably without resorting to external financing.
Amazon did it by intention
Now TSMC as a manufacturing company probably sounded too old and too irrelevant for the cool modern Silicon Valley people. But in fact the story of accounting profit and capex is not only that of hardware businesses. Amazon has actually had the same story for years, and even more — they did it on purpose.
Some of you might remember that around the time of 2013~2015 when Amazon share price & market cap were skyrocketing, starting to propel Jeff Bezos toward a jaw-dropping wealth, a lot of the journalists “criticized” how Amazon was most of the years making effectively no profit and that they thought the exponentially rising share price was a bubble. The more sophisticated ones pointed to how Amazon used tax strategy to reduce its taxable income to almost $0 so that it was able to pay very little or no corporate tax at all.
As seen on the graph above, during the decade of 2011–2020, one can see that as revenues (blue bars) climbed almost exponentially, incomes before tax (orange bars) stayed for a long time, indeed, close to non-existent and didn’t start to emerge out of the horizontal axis obviously until maybe 2016. But the share prices already started growing exponentially somewhere around 2013, if not 2012.
So how could the share price of a seemingly “non-profitable” company rise so fast during those years? Were the investors crazy?
Again, of course not.
What Amazon did was deliberately pursuing the depreciation to minimize (for the moment) taxable income.
While TSMC is generating healthy net income so that its OCF perennially outdoes the accounting profit due to the nature of depreciation of expensive equipments, Amazon saw an opportunity to suppress its taxable income by aggressively growing its capex and depreciate heavily to net out the accounting profit.
The weapon at hand was AWS.
AWS is a cloud business where most of the operating costs are the expenditures on data centers, including both building them and filling them with servers, as well as operating them with staffs.
When AWS initially showed signs of success and the revenue started growing significantly, someone at Amazon — wouldn’t be surprised if it was Jeff himself as he’s one of the most financially savvy Silicon Valley founders— saw an opportunity: given the speed of growth in this business and given that servers are typically depreciated over 3 years (even shorter than a fab!), if Amazon could continue increasing its capex building new data centers and growing revenue accordingly, there’s a chance that Amazon could use fast-growing depreciation of servers to offset the profit from AWS operation!
The graph above is the OCF (Operating Cash Flows) and ICF (Investment Cash Flows) of Amazon between 2011 and 2020. As one can see, while Amazon’s accounting profit did not grow a lot during the decade of explosive revenue growth, its OCF was indeed growing exponentially, a substantial part of that was the highly lucrative AWS business.
What Amazon did was then taking that OCF, turning around and investing it on servers that would depreciate over 3 years — en masse. This was captured in the negative ICF almost inverting that of OCF over the decade.
Now just as TSMC’s EUV equipments don’t become unprofitable on the eve of 4th-year anniversary, obviously AWS servers don’t all break down after 3 years and require replacements, either. Many of them could operate with proper maintenance for longer periods. And AWS could always move older servers to less demanding cloud services (such as backup storage) that continue to generate decent profit. So it’s safe to say that the servers, though depreciating over 3 years, will probably reasonably and profitably operate for much longer than that.
Now if AWS’s server numbers and revenues are fixed every year, after year 3 when depreciation is over, the operating profit will jump up like crazy, leading to taxable income. But since AWS was growing exponentially, every year for the amount of servers whose depreciation is over, a lot more new servers have been purchased and entered into depreciation cycle, allowing the magic wheel to continue.
Obviously this game cannot be played forever. The corporate finance professor will tell you that it simply delays the taxable profit to the future. One day when you could no longer grow exponentially, you will no longer be able to offset the profit with depreciation and then you have to pay tax.
On the other hand paying tax in the future is always better than paying it now, assuming Amazon keeps growing. Heavy capex also keeps your less cash-rich competitors at bay — just like TSMC out-capexed UMC, Charter and Global Foundry, pushing them out of the advanced-node competition. (The fact that Microsoft Azure was able to catch up with AWS is worth another blogpost though).
Now coming back to whether the hyper growth in Amazon’s share price and market cap in those years was a “bubble” like some journalists suggested. With the OCF graph above now you see why the Wall Street was pushing up the share price despite the “lack of profitability” of Amazon. Any qualified financial analyst would immediately see that Amazon was crazily “profitable” based on the exponential growth of OCF — which is much more valuable than Operating Income — and that Amazon was just “hiding” them by turning around and investing them in even more new servers. There’s really no reason, therefore, to whine about the “lack of probability”.
Oh, btw, this is also why next time a self-claimed investing expert starts talking about P/E ratio of a company, you should immediately walk away (or turn off the TV).
It’s actually same with startups
Note that a VC does not need to know all the intricacies explained above about TSMC and Amazon and still be able to be super successful by backing Uber or TikTok early. We invest in people, not financial statements, after all. Otherwise my industry will have already been filled with suits-and-tie former Goldman bankers, when in fact all you see are mostly nerdy ex-engineers in clumsy Patagonia vests.
But with startups there is still the same lesson to learn here. Again, many journalists continue to focus on the “not profitable” side of unicorn startups when they file to go public and financial statements are finally available for the public.
However, accounting profitability of a unicorn startup should be the last thing one should care about, coz’ the startup could be unprofitable due to:
(1) Hyper growth fueled by expensive user/client acquisition: in this case as long as the LTV per user/client is high, one should keep spending marketing and sales to grow even faster and forget about the annual accounting profitability.
(2) Stock-based compensation plans: there’s usually a one-off stock compensation expense when the company goes public. This is due to the accounting principle of expending stock-based compensation based on the market price. Depends on the cost basis and the IPO share price, this expense item could single-handed be the largest item in all lines of cost, leading to net loss at the bottom line.
So if we can’t look at the accounting profit of the startups going public, what should we look at? Well, by now you should know: cash flows.
In general if a startup is growing its OCF exponentially in the years leading up to going public, I would already not be worrying about it. And if they were able to grow FCF (Free Cash Flow, most of the time simply OCF+ICF) exponentially from negative to breaking positive, then I would bet on a very high multiple for their 1st day of trading.
Or put it this way, a startup could be posting a $100M accounting loss on a $200M annual revenue but I’m still willing to price it at 20x ($4B) of market cap, as long as I see an exponential growth trend in OCF and FCF.
Just like what they said in the 1976 movie All the President’s Men:
Follow the money (cash)!
