The hidden financial costs of chasing AI supremacy and who is unwillingly footing the bill
Meta and Microsoft report earnings today. Will they keep overstating their profitability?
Disclaimer: This content is for educational purposes only and does not constitute financial or investment advice. Always do your own research or consult a qualified financial adviser before making any financial decisions.
We live in an age of big ‘bold’ announcements and AI sentiment driven swings. Proponents of AI, namely the tech behemoths, keep announcing bolder AI capex commitments:
Alphabet wants to plough through ~$200bn of AI capex this year.
Meta and Microsoft are expected to land in the same ballpark.
That’s close to spending 1x Sales on something with an uncertain ROIC. Within one year.
The structured financier in me keeps raising my eyebrow every time I hear these numbers. All I can think of is the amount (and type) of collateral they will have to pony up to secure that kind of funding. It’s not like they own $200bn worth of treasury shares. Neither has that kind of cash on their balance sheet. Which means either raising fresh equity or pledging real assets to lenders.
This race for capex supremacy within the AI ecosystem is breaking balance sheets.
But there’s another race happening. One that doesn’t make headlines, yet is quietly breaking the reliably profitable, free-cash-flow-generative nature these companies built their reputations on.
That’s the war for AI talent. And it’s happening literally off-books, missing from the income statement.
In this article, we will:
Show how the war for AI talent has silently ballooned recurring employee expenses, and who ultimately bears it.
Reveal Meta and Microsoft’s true profitability and free cash flow once that cost is put back on the books.
With earnings out today, watch closely whether this hidden AI cost keeps expanding, or gets brought under control.
The war on AI talent
Nowhere is this clearer than at Meta. Despite having run Reality Labs for years, Meta was late to the AI game, trailing in capability even against Copilot.
So much so that in summer 2025, Mark Zuckerberg personally assembled a list of the industry’s best AI researchers and went after them with sky-high pay packages: Apple’s AI lead Ruoming Pang joined for a package reportedly exceeding $200m; Andrew Tulloch was offered as much as $1.5bn over six years (but turned it down).
Every one of these packages carries both a cash and a stock component (‘RSU’), tied to the researcher’s ability to deliver the performance Meta wants. The cash element hits the income statement immediately. The stock component, though, is where things get murkier.
The ‘off-the-Income statement’ ruse
There is absolutely nothing wrong with granting stocks to employees. They form an important part of the overall compensation strategy for employees, incentivising people to outperform and to stay with the firm.
US GAAP sets the accounting policy, and both Microsoft and Meta apply it correctly.
When a RSU is granted by the firm, its cost is fully booked into the income statement. The cost is usually the share price at granting date, multiplied by the number of RSUs granted. The underlying shares are delivered to employees over time, on a vesting schedule.
Since FY22, Meta has granted and booked a total of ~US$70bn in RSUs to employees. Similarly, Microsoft has granted ~US$42bn (since FY23).
→ Both have duly put this cost through their Income statements. So far so good right?
Actually, not really.
The hidden cost only shows up when the shares actually vest and have to be given to the employees. At that point, Meta and Microsoft face a choice:
Option A (the market standard): issue fresh share capital – this increases the number of shares but Earnings per share (for shareholders) get diluted.
Option B (the cowboy approach): buy those shares on the open market and hand them to employees instead:
Advantage: total share count stays flat, so no EPS dilution to report.
Issue: under US GAAP, this qualifies as a share buyback, and the cost of acquiring those shares is booked straight to the cash flow statement and balance sheet.
So when time comes to giving the actual shares to employees, a company can either report E.P.S dilution or it can reduce the book equity reserves/retained earnings directly without anyone really picking up on this.
Guess which approach Meta and Microsoft (tho’ to a lesser extent) have chosen?
That’s right: the cowboy approach.
US GAAP was built on the assumption that neither route would materially move profits, because which company would ever choose to grant ‘enough’ RSUs to employees that it could move the share price right?
Meta has granted ~US$70bn in stocks. And Microsoft ~US$42bn. The sheer size of these numbers IS enough to move the needle materially. RSUs, granted in large quantities every year, compound into visible profit erosion and free cash flow consumed by “obligatory” share buybacks, as employees must receive their promised shares annually regardless of market conditions.
As of FY25, total unvested RSU’s book value was ~US$60bn and corresponding market value was ~US$80bn. That $20bn gap is money Meta will have to find, one way or another, the moment these shares vest, either by diluting shareholders with fresh stock, or by spending real cash buying shares back to hand over instead. A buyback would crystallize a gain or a loss, one that’s not reported in the income statement under US GAAP.
The impact on Free Cash Flow
Share buybacks are usually voluntary by the company and are designed as instruments to return capital to shareholders. Under US GAAP, these are financed by Free Cash Flow and accounted for in the cash flow statement and balance sheet.
But when a company is buying shares at market price purely to hand them to employees, that stops being a return of capital. It becomes a recurring cash cost of the compensation programme.
Meta’s true operating and cash flow performance till Q1 26
Using the extensive information provided by Meta’s financial notes on the RSU programme, we can back-solve the profit erosion that is being kept ‘off-the-Income-Statement’:
Since December 2024, as the share price of Meta increased and the AI talent war started, ~10 points of EBIT margin per year have been given away as additional, uncaptured employee compensation.
Meta has been over-reporting profit, materially.
When it comes to Free Cash flow:
Real Free Cash flow is a fraction of reported Free Cash Flow. In FY25, real free cash flow to firm was almost wiped out by the extra compensation to employees.
Microsoft’s true operating performance through FY25
Microsoft has been building AI capability for years, so it’s had to fight far less hard for talent than Meta. The profit erosion here is much lighter: about 2 percentage points of EBIT margin are consistently lost to buying shares on the open market for employees.
Microsoft has been over-reporting profit, to a lesser extent.
What this means for investors
There are numerous implications for investors in these companies:
Any fundamental valuation model for either company needs to account for the recurring nature of this hidden compensation cost.
It would materially cut any DCF target price, potentially by over $100/share depending on assumptions.
My aim, as a corporate investor, is to challenge Meta’s and Microsoft’s management to prove that compensation at this scale will actually be accretive to shareholders in the long run.
There’s already a mountain of AI capex with no visible ROIC. Now there’s a hidden cost of talent on top of it.
And who ultimately pays the bill? The shareholder.
Today, Microsoft releases full-year FY26 results, and Meta reports Q2 26. Let’s see which method they choose to continue pursuing.