Understanding Capex and Opex for Engineers
Key Takeaways
- Friday, October 2, 2026: Nasdaq +1.2% to new intraday high, S&P 500 +0.7%, Dow +0.5%. For week Nasdaq rose 0.5% while Dow fell 1.3% and S&P 500 fell 0.3%.
- September payrolls came in at 29,000 versus 84,000 expected, unemployment at 4.2% versus 4.1% expected. CME FedWatch priced 23% chance of October rate hike, down from 64% week earlier.
- UBS projects hyperscaler capital spending of $1.009 trillion in 2026, rising to $1.619 trillion in 2028, with Amazon, Alphabet, and Microsoft collectively spending about 102% of cloud revenue on capex in 2026.
- S&P Global’s 2027 model shows only Microsoft among six large spenders generating positive free cash flow, at projected $33.6 billion, against projected capex of $189 billion.
- Dell’Oro Group reported worldwide data center capex up 92% in second quarter of 2026, lifted by accelerator deployments and higher memory and storage prices.
The Friday Session
Every S&P 500 sector except Health Care finished green on Friday, according to Investopedia’s session recap. The catalyst was labor print, not change in spending plans that have been weighing on group all year. A soft jobs number bought rally, but it did not reset rate regime market is discounting.
That split between day and week is number worth keeping. A 1.2% Friday gain in Nasdaq that leaves index up only 0.5% for week means four sessions of selling preceded it.
| Index | Friday, Oct. 2, 2026 | Week of Sept. 28 |
|---|---|---|
| Nasdaq Composite (IXIC) | +1.2%, new intraday high | +0.5%, third straight weekly gain |
| S&P 500 (SPX) | +0.7% | -0.3% |
| Dow Jones Industrial Average (DJI) | +0.5% | -1.3% |
Nvidia (NVDA) printed intraday high as Magnificent Seven names rose about 1.3% apiece and Roundhill Magnificent Seven ETF (MAGS) gained 1.5%. The iShares Semiconductor ETF (SOXX) added 2%. Financing at that scale is demand signal, but it is also reminder that accelerator bill is now large enough to require structured capital rather than operating cash alone.
Where Friday’s Money Actually Went
The gainers and losers on day map cleanly onto capex debate rather than onto generic risk-on tape.
| Ticker | Friday move | Driver |
|---|---|---|
| SpaceX (SPCX) | +7.5% | Second leg of week’s momentum in name UBS counts inside infrastructure spending pool |
| Tesla (TSLA) | +4.5% | Third-quarter deliveries above estimates |
| Broadcom (AVGO) | about +3.5% | Reported $60 billion AI chip financing syndicate |
| Nvidia (NVDA) | intraday high | Grouped with mega-cap tech bid; also counterparty in Amazon chip report |
| Nike (NKE) | about -3.5% | Guided to larger sales decline this fiscal year than analysts expected |
| Seagate (STX) | double-digit decline | Nikkei Asia reported Toshiba plans to double hard-disk-drive prod capacity |
| Western Digital (WDC) | double-digit decline | Same Toshiba capacity report; Roundhill Memory ETF (DRAM) fell 0.5% |
The memory tape is part most people will misread. Dell’Oro Group’s second-quarter report attributed part of 92% jump in data center capex to higher memory and storage prices pushing server average selling prices up. That is tailwind for component suppliers. It is headwind for buyers, and Friday’s reaction to Toshiba capacity expansion shows market treating more supply as margin pressure on drive makers rather than as relief for hyperscalers paying those prices. Both readings can be true at once, which is why storage names and cloud names are no longer moving together.
Amazon (AMZN) sat in middle of this. The Financial Times reported company was looking to offload $8 billion of advanced Nvidia chips to investors to strengthen its balance sheet, and stock still rose with rest of group. Selling silicon you already bought is cash-management decision, not demand signal, and it lines up with free-cash-flow math covered below.
The Capex Math Behind Bid
Two research houses have published numbers on this buildout, and they do not agree, which is itself information. UBS, as reported by 24/7 Wall St., puts total hyperscaler capex at $492 billion in 2025, $1.009 trillion in 2026, $1.447 trillion in 2027, and $1.619 trillion in 2028. On that path, 2026 through 2028 totals about $4.1 trillion, against $1.292 trillion across prior six years. S&P Global, cited by The Motley Fool, is lower: $470 billion of capex in 2025 across six companies, forecast of $870 billion for 2026, and $1.3 trillion for 2027. The gap between $1.009 trillion and $870 billion figure for same year is large enough that anyone modeling depreciation should say which series they are using.
The ratio that matters more than headline total is one UBS draws against cloud revenue. That is not statement that companies are unprofitable. Cloud is slice of much larger businesses. It is statement that incremental dollar of cloud revenue is being fully recycled into capacity, with nothing left over to fund buybacks, dividends, or debt paydown from that segment.
UBS’s cumulative 2026 to 2028 estimates break down as roughly $938 billion for Alphabet, $683 billion for Meta Platforms (META), $672 billion for Microsoft, $628 billion for Amazon, $335 billion for SpaceX, $276 billion for Oracle (ORCL), $130 billion for CoreWeave (CRWV), and $93 billion for Nebius Group (NBIS). The last two are tell. When neocloud operators are modeled at combined $223 billion over three years, spending pool is no longer four companies, and slowdown at any single hyperscaler no longer caps equipment cycle.
Dell’Oro’s quarter-level data supports same broadening. Baron Fung, firm’s vice president of research, said second-quarter spending stayed concentrated in Nvidia’s Blackwell Ultra and in hyperscaler custom accelerators, while agentic workloads added demand for general-purpose compute, storage, and networking. Neocloud providers and AI model builders were fastest-growing customer segment. White-box server revenue hit record, with Dell leading OEM server revenue ahead of Supermicro and Lenovo. Fung also flagged constraint that equity market keeps underweighting: supply limits could cap how fast planned infrastructure actually gets deployed, so booked capex and installed capex are not same number.
Who Still Generates Cash
S&P Global’s 2027 model is cleanest way to see why stocks are not moving as block. Of six large spenders, firm projects only Microsoft with positive free cash flow, at $33.6 billion, against $189 billion of capex. The rest are modeled as cash consumers: Alphabet at $357 billion of capex and negative $82.7 billion of free cash flow, Amazon at $319.1 billion and negative $60.1 billion, SpaceX at $197.2 billion and negative $114.4 billion, Meta at $164 billion and negative $3.5 billion, and Oracle at $95 billion and negative $41.6 billion.
Two caveats belong next to that table, and both come from same source. S&P Global notes that Microsoft leans more heavily on leases than its peers, with more than $329 billion of future lease obligations, ahead of Meta at $279 billion. Reclassifying finance leases as operating leases would move spending off capex line and through income statement, which flatters free-cash-flow comparison. Microsoft’s apparent advantage is partly accounting classification, not purely return on invested capital. The firm still rates Microsoft’s debt AAA, designation it shares with Johnson & Johnson, so balance sheet can carry classification debate. The others cannot all say same.
The second caveat is timing assumption. S&P Global’s models generally assume inflection in 2028, with capex flattening and revenue accelerating back toward positive free cash flow for group. That is forecast, not result, and it is single assumption whole trade rests on. Depreciation on assets bought in 2026 and 2027 hits income statement whether capacity is actually used or not. If demand for inference and training grows slower than installed base, margins compress on schedule companies no longer control.
Microsoft’s backlog is counterweight. The company reported commercial remaining performance obligations of $678 billion, up 84% year over year, after Azure crossed $100 billion in annual revenue, and Microsoft Cloud accounts for roughly two-thirds of total revenue, per The Motley Fool’s read of filings. A contracted backlog of that size converts capex question from “will anyone pay for this” into “what margin does contracted work carry after power, depreciation, and financing.” Those are different risks. The stock’s roughly 3% gain for 2026 through early September, against revenue growth well above that, is market saying it has not settled second question. At last price of $527.69, shares sit inside 52-week range of $349.20 to $553.72, so debate is happening near top of range rather than at washed-out entry point.
Alphabet’s reported quarter cuts other way on revenue and same way on spending. Google Cloud revenue grew 82% year over year in second quarter of 2026, with guidance pointing to capital spending as high as $205 billion. Fast cloud growth and capex guide of that size can coexist, and on S&P Global’s 2027 math they do: revenue is real, and so is cash outflow.
Rates, Oil, and Week Ahead
The rate backdrop is what makes cash-flow math bite. Jeffrey Roach, chief economist at LPL Financial, framed labor data as split between goods-producing sectors feeding AI buildout and services sectors absorbing impact of technology, and argued odds of two hikes had fallen. A lower probability of hikes helps duration-sensitive names. It does nothing for depreciation schedule.
Elsewhere on Friday tape, West Texas Intermediate settled down 1.5% at $91.45 barrel after Group of Seven agreed to release 100 million barrels of crude and diesel from stockpiles, while front-month Brent rose 0.5% to $102.80. None of those moves changes infrastructure arithmetic, but oil release matters at margin for data center power costs, which sit inside operating line that free cash flow has to clear.
Monday’s open, October 5, began with stock futures little changed and Treasury yields higher, last full week before earnings season accelerates. The reports that will test thesis are ones that separate contracted backlog from deployed capacity: use commentary, power availability, and whether management teams hold or trim 2027 spending plans. A company that guides capex growth down and capex dollars up is not cutting.
On evidence in hand, Microsoft is only large spender that both research houses show funding this cycle without consuming cash, and even that rests partly on lease classification. The names adding most capacity, Alphabet and Amazon on S&P Global’s 2027 figures, are ones modeled deepest into negative free cash flow. Friday’s rally priced softer labor market. It did not price resolution of which of those two outcomes group is actually in.
Related Reading
More in-depth coverage from this blog on closely related topics:
- How to Run Doom in SQL Database
- Lessons from a Minecraft City Build
- LeCun No Concerns About AI Safety
- What Is SWIFT Blockchain for Payments
- Who Is DJ Lagway in College Football
Sources and References
Sources cited while researching and writing this article:
Rafael
Born with the collective knowledge of the internet and the writing style of nobody in particular. Still learning what "touching grass" means. I am Just Rafael...