Tesla's Capital Spending Rose 142%. The Filing Never Says What the Money Bought.

Tesla's second-quarter capital spending more than doubled, and the update that discloses the number never says what the money bought. Capital expenditures were $5,789 million against $2,394 million a year earlier, a rise of 142%, and free cash flow swung to negative $1,092 million from positive $146 million. The supporting-infrastructure table gives physical capacity by asset, Cortex 1 above 90 megawatts and Cortex 2 above 115, both in production, alongside gigawatt-hour figures for Nevada LFP, Texas 4680, cathode materials and lithium refining, and it attaches no dollar to any of it. The same year's spending covers Cybercab lines at Giga Texas, Optimus lines replacing decommissioned Model S and X capacity at Fremont, Nevada cell production, cathode and lithium refining, a Megafactory and site work toward a semiconductor fab. No line in the filing apportions a dollar between any of them, which means every published figure for the AI share of Tesla's capital budget is somebody's estimate rather than the company's disclosure.

The full-year number in circulation came from the call rather than the filing. Chief Financial Officer Vaibhav Taneja told analysts that Tesla continues to expect capital expenditure of "more than $25 billion" this year and that it will grow for the next two or three years as the company expands the robotaxi fleet and Optimus production capacity, invests in a semiconductor fab, installs solar manufacturing capacity and builds AI compute infrastructure, in that order and with no split between them. Tesla is also "being opportunistic in securing certain debt facilities" that give it capacity to borrow up to $30 billion to accelerate the work, he said. Chief Executive Officer Elon Musk stated the pacing rule himself, saying Tesla should be spending on capital expenditure "as fast as we can without it being too wasteful" and that "it's okay to be a little less capital efficient if we get things done sooner, because that's actually going to be the higher NPV outcome for the company." That's a defensible posture, and none of it is in the filed update.

Where the Capital Is Coming From

Borrowing against a capital plan is the change that has already worked through the largest balance sheets. FactSet's analysis of Alphabet, Amazon, Meta, Microsoft and Oracle found incremental annual debt rising from 9% of capital expenditure in fiscal 2024 to 32% on a trailing twelve-month basis by the middle of this year, taking aggregate total debt to roughly $700 billion, with free cash flow expected to land near zero or below for all five except Alphabet and Microsoft. S&P cut Oracle to BBB- from BBB on 9 July, citing surging capital expenditure, negative free cash flow and customer concentration, and Alphabet priced an $84.75 billion equity raise in June. Inside those budgets the compute share has moved from about 43% in 2022 to roughly 60% this year, so the mix is tilting toward the shortest-lived assets in the build at the same moment the funding tilts toward debt. Goldman Sachs Global Institute put silicon's assumed useful life at four to six years in May, against roughly twenty for buildings, and called that assumption the single most influential variable in any forecast of the buildout's scale.

Some of the commitment sits outside the balance sheet entirely. Moody's Ratings put 2026 capital spending across Microsoft, Amazon, Alphabet, Meta, Oracle and CoreWeave at $785 billion, rising to about $1 trillion next year, with direct debt across the six near $460 billion and lease commitments of $1.2 trillion, more than $820 billion of which relates to leases that have not yet commenced. A smaller number across a larger group is a different measure rather than a contradiction. Moody's counts direct debt and holds the leases separately, where FactSet's roughly $700 billion is total debt across five companies. Which base a reader is looking at decides how large the obligation appears, which is the same problem the rest of this piece is about. Moody's wrote that the "transition from asset-light to asset-heavy models requires unprecedented levels of investment and capital raising," and that investors "will increasingly focus on these companies' ability to realize an adequate return on investment." A lease that has not commenced is an obligation with no asset standing against it yet, which is a harder thing to put in front of a board than a machine already humming in a hall.

A company that decomposes its spending is easier to argue with. Intel reported second-quarter revenue of $16.1 billion, up 25%, with the Data Center and AI segment at $6.3 billion, up 59%, and Intel Foundry at $5.8 billion, up 31%. Chief Financial Officer Dave Zinsner said the company is "meaningfully increasing our investments in equipment, clean room space, and substrates" to support expected growth across products and foundry, and Intel announced a €5 billion investment in capacity for Xeon 6 and the next generation on Intel 3. Equipment, clean rooms and substrates is a sentence analysts can price against a depreciation schedule. Installed megawatts sitting inside an undifferentiated total gives them nothing to work with.

Reading the Expense Lines

American Express raised full-year revenue growth guidance to 10% on the strength of the first half, and Chairman and Chief Executive Officer Stephen J. Squeri said the company will "reinvest this outperformance in growth initiatives given the significant opportunities we see ahead." The same release puts consolidated expenses at $14.5 billion for the quarter, up 12% against revenue growth of 10%, and attributes the increase primarily to higher variable customer engagement costs from increased Card Member spending, the U.S. Platinum Card refresh and usage of Card Member benefits. The filed expense table shows which lines carried it. Marketing rose 6% to $1,650 million. Data processing and equipment rose 13% to $817 million. Salaries and employee benefits rose 9%, business development 10%, professional services 5%. Card Member services rose 50%, to $1,949 million from $1,301 million.

There is no line called technology on that statement. The two that come closest grew 6% and 13% in the quarter, while the cost of the customer proposition grew 50%, and the aggregate outgrew the revenue it was funded from. Growth investment is a phrase that covers every one of those lines equally well.

Management discussed technology at length anyway. Squeri said on Amex's earnings webcast that the company has "a large backlog of technology projects, and so we're getting to more things quicker," that "there is no shortage of technology investments or enhancements or refreshes that need to occur," and that "across a wide range of technology platforms, we're able to pull some of those investments into the second half of the year." Payments Dive reported that Amex disclosed no dollar amount for the additional technology spending, and that Chief Financial Officer Christophe Le Caillec put the second-half marketing increase at 10% against 2025. That guidance is the origin of the 10% now traveling as a spending story, and the quarter itself reported 6%. Amex publishes the table underneath both numbers, so the argument about whether the reinvestment is the right one can actually be had.

The category finance can least easily decompose is the one arriving now. Accenture research reported by Fortune puts roughly 8.7 trillion tokens a week through one of the firm's internal platforms, estimates that only 10 to 20% of enterprise tasks are complex enough to justify a frontier model, and finds that routing work to cheaper models cuts cost to around a sixth of running everything at the frontier. Usage of one internal AI tool rose 113-fold in ten weeks, with 19% of users accounting for about 80% of the spend. Lan Guan, Accenture's Chief AI and Data Officer, said clients "are ready to scale AI, but then they hit this unexpected cost wall," and that on their own token consumption "they're literally telling me that they are walking in the dark." Accenture sells the token-management work its own research describes, and the internal usage figures are unaudited, both worth holding in view.

The Seat That Signs for It

None of that explains why a finance chair changes hands, and the filing that recorded the most visible change of the month does not try to. GE HealthCare disclosed that James K. (Jay) Saccaro notified the company on 21 July that he would resign as Chief Financial Officer "for a role outside of the medical technology industry," with a last day of 14 August. The filing names no destination and gives no other reason. Two days later the company named George Newcomb, its Controller and Chief Accounting Officer, interim Chief Financial Officer from the same date, keeping both existing titles with no change to his compensation arrangements while a search for a permanent successor runs. Nothing in the record ties the departure to capital spending, to AI investment or to expense pressure.

The appointment fits a hiring pattern that predates the quarter. Interim appointments reached 12% of new CFO hires globally in the first quarter, double the 6% of 2025, in Russell Reynolds Associates' Q1 2026 Global CFO Turnover Index, which counted 89 appointments against 95 a year earlier and found 42% of new CFOs had already run finance at a public company, up from a seven-year average of 35%. Boards want people who have done the job and there aren't enough of them, so the gap gets covered by whoever already signs the filings. Crist|Kolder Associates counted 120 CFO turnovers across 664 large US companies during 2025, up 17.7% on the prior year, with 65% of replacements hired internally, which is the same instinct without the deadline. Newcomb holds the Controller's title and the CFO's at once.

Tesla's update, Amex's expense table and GE HealthCare's 8-K are short documents and none of them is difficult to read, which is what makes the composition question tractable for anyone who opens them. Take the total, find which lines moved and by how much, and mark the places where a company declined to break down the figure it's asking to be judged on. A budget defended on its disclosed composition can survive a hostile question, while a budget defended on a headline total is being defended on an estimate of what that total contains, and the estimate usually belongs to whoever wants the money.