Meta Needs US$95 Billion of Equity to Pay for One Year of Capex: the 3 Lines I Changed in Our Engineering Budget

Most Singapore engineering leads can safely ignore a US equity offering. This one is worth twenty minutes, because of what it says about the price of the things we all rent. A company with Meta’s cash generation has decided it needs to sell up to US$95 billion of stock to fund a single year of infrastructure, and three of its own suppliers are buying in. That is not a story about social media. It is a story about who pays for the compute, and the answer eventually includes us.

What Was Actually Announced, in Its Parts

The headline number circulating is US$95 billion, and it is two different things stacked together. The live piece, launched on 26 October 2026, is roughly US$35 billion of underwritten offerings. According to Meta’s own free writing prospectus filed with the SEC, that splits into about US$22.5 billion of Class A common stock sold under forward sale agreements and about US$12.5 billion of depositary shares, each representing a one-twentieth interest in newly issued Series A Mandatory Convertible Preferred Stock, converting on or about 15 August 2029. Underwriters hold a 30-day option on up to US$3.375 billion of additional Class A stock.

The second piece is an at-the-market programme of about US$60 billion, tied to the previously announced Class C common stock, which DigiTimes reports is not expected to begin until the first quarter of 2027, once Class C trading starts. So the US$95 billion is a ceiling across two instruments and two calendar years, not a single sale this week. Several outlets collapsed that distinction, which is why the figure looked different depending on where you read it.

The purpose is stated plainly. CNBC and The Irish Times both report Meta guiding to US$135 billion to US$145 billion of capital expenditure in 2026, funded through a mix of operating cash flow, investment-grade debt, strategic partnerships and equity, and warning investors it expects to need hundreds of billions of dollars over several years. The context from the quarter is the part worth holding onto: the 21 October results showed revenue up about 26 per cent year on year alongside negative free cash flow of roughly US$2.12 billion, with quarterly capital expenditure near US$34.9 billion. Revenue growing strongly and cash flow going negative at the same time is the whole argument for the raise.

Then the cornerstone investors, which is where this stops being an ordinary financing. Meta named Nvidia, AMD and Broadcom as having indicated interest in purchasing up to an aggregate US$5 billion of Class A stock, severally and not jointly. Nvidia has since said it will invest US$3 billion and AMD US$1 billion. Broadcom appears in Meta’s release without an individual figure.

Line 1: We Put a Price-Change Clause in Every Renewal

The first thing I changed was not a headcount. It was the shape of our vendor agreements for next year. When a sector is funding capacity with equity and debt rather than operating cash flow, price increases are not a risk that might materialise, they are the repayment mechanism. The question is only whether they arrive as a headline rate change, as a quiet reduction in what a tier includes, or as a commitment structure that is cheaper per unit and much harder to leave.

Three clauses are worth insisting on at renewal, and none of them are exotic. A notice period on price changes, long enough that you can plan a migration rather than absorb a surprise. A definition of the unit being priced that cannot be revised unilaterally, because a per-token price is meaningless if the tokeniser changes. And an export path for your own data, specified in a format and a timeframe, not as a support ticket you can file in theory. If you are modelling what a step change does to your run rate, the team cost calculator is a reasonable starting point for the people side, and project estimation for the delivery side.

Line 2: Concentration Became a Named Risk With an Owner

The cornerstone structure is the genuinely new thing here, and it deserves more attention than it got. Suppliers taking equity positions in a customer that buys their hardware tightens an already tight loop. None of that is improper and all of it is disclosed, but it changes how a buyer three steps down the chain should think about concentration. The commercial interests of the companies selling you silicon, the companies selling you capacity and the companies selling you models are more aligned with each other than they were a year ago.

So we wrote it down as a risk with a name against it, the way we would for a single-supplier dependency in any other part of the business. The useful version of that exercise is not a paragraph in a document. It is three questions with dated answers: if our primary model provider doubled its price next quarter, what would we do, who would do it, and how long would it take. If the answer to the third question is “we have never tried”, that is the finding.

Line 3: We Stopped Budgeting AI as a Project

The third change is the least glamorous and the one I would keep if I could only keep one. We moved AI spend out of project budgets and into a run-rate line with a named owner, reviewed monthly against usage, the way we treat any other utility we consume at variable volume.

Project budgeting hides the thing that matters. A pilot that costs a modest amount per month looks harmless in a project column and becomes a material commitment once it reaches production traffic, at which point the cost curve bends in a way nobody forecast because nobody owned the forecast. Moving it to a run-rate line forces three questions into the open every month: what did we spend, on which features, and what is the cost per unit of the thing the business actually counts. Teams that cannot answer the third question are the ones that will be surprised by a 2027 renewal.

This is also what makes the two roles above defensible at budget time. A headcount request to “work on AI” competes against every other project. A headcount request that arrives attached to a run-rate line, with a cost-per-request figure and a stated target, is arguing in the language the finance team already uses. The same logic applies when you build the team around a workload: cloud application development in Singapore is well supplied, but the people who can tell you what a design decision costs per request are a narrower group, and they are the ones worth the competitive offer.

What This News Is Not

It is not a bubble call. Reporting on the market reaction was contradictory within hours, with some outlets recording a small premarket rise and others a decline of a similar size during the session, which is a reasonable indication that nobody knows what it means yet. Anyone telling you confidently this week that the raise proves either strength or distress is reading their own position back to you.

It is also not a reason to slow your own hiring. The spending is going up, not down, and the volume of engineering work that sits downstream of it, integration, evaluation, data plumbing, cost control, goes up with it. What changes is which roles are easy to justify. A generic AI engineer requisition will get harder to defend through 2027. A requisition tied to a number that appears in a board pack will get easier.

And it is not an argument for avoiding the major providers. Concentration is a risk to manage, not a reason to build your own infrastructure, which for almost every Singapore business is a more expensive mistake than the one it is trying to avoid. The point of the portability work is to hold a credible alternative, because a buyer with a tested exit negotiates differently from one without.

FAQ: Meta’s Equity Raise and Singapore Engineering Budgets