AI Daily: Meta’s Interest Costs Jump 330% on Higher Debt

The cost of financing AI was the day’s biggest thread on 22 October 2026, by our grouping of the news. Meta’s interest expense was $977 million in the third quarter, up from $227 million a year earlier, an increase of 330%, according to its Form 10-Q, which attributes the rise to “higher long-term debt balances.” By our arithmetic from their filings, eight companies, from Alphabet and Meta to Nvidia and SpaceX, list about $3.56 trillion of purchase commitments, leases not yet commenced, guarantees and similar obligations.

The threads below are Meta’s quarter and its bill; debt, commitments and neoclouds; earnings beyond Meta; labs that sell deployment help; agents, security and safety; and money, people and policy. Each item links to the full piece, where the primary source is cited.

AI Daily Briefing · video

Today’s stories in about a minute (22 Oct 2026).

Meta’s Quarter and Its Bill

Meta’s third quarter showed revenue of $64.60 billion, up 26%, and capital expenditures of $34.91 billion, more than the $32.79 billion of cash its operations generated, which left free cash flow at negative $2.12 billion, according to its release. Its operating margin was 20%, against 40% a year earlier, after a $9.89 billion charge for its agreement with U.S. state attorneys general. Read: Meta’s quarter, explained and the margin after the charge.

Meta had $454.52 billion of non-cancelable contractual commitments at 30 September 2026, against $81.19 billion a year earlier, according to its 10-Q; by our arithmetic, that is 5.6 times as much in twelve months. Bloomberg reported that a gauge of Meta’s credit risk hit a fresh high on its AI spending plans. Read: Meta’s commitments and its interest costs.

Shares of Meta fell about 3% in premarket trading, Reuters reported, while at least 13 brokerages raised their price targets, according to LSEG data cited by Reuters. Meta said on its earnings call that its AI agent app Muse has been adopted by more than 10 million users, TechCrunch reported. Read: the share move and Muse’s users.

Debt, Commitments and Neoclouds

A chart credited to ZeroHedge, built on a Morgan Stanley Research framework, says hyperscalers, Nvidia, Broadcom and SpaceX “now disclose $3.59tn of off-balance-sheet commitments”; our check of their filings comes to about $3.56 trillion. Anthropic and Meta each carry about $450 billion of future commitments, though the two figures are built differently. Read: the commitments check and Anthropic and Meta.

A Goldman Sachs note, as relayed by BigGo Finance, estimates CoreWeave “would still need to raise roughly $19.5 billion to $23 billion in new debt to fulfill its $129 billion backlog as of August,” while Bernstein analyst Madison Rezaei, according to TipRanks, estimates the company needs about $85 billion to fulfil existing customer contracts, leaving an estimated $76 billion to finance using outside capital. CNBC reported that CoreWeave’s unsecured debt is “sporting a yield of about 15%.” Read: CoreWeave’s funding gap and its debt yield.

Nebius said it has entered into a $3.5 billion senior secured facility provided by Citi, with a 364-day maturity. Investors betting against five US neocloud and data center companies have made $6 billion in paper profits so far in October, Reuters reported, citing ORTEX. Read: the Citi facility and the short sellers.

Chinese stocks entered a bear market, with US$2.4 trillion lost since a peak in May, Bloomberg reported, and Nasdaq 100 futures were down 1.72% before the bell, Reuters reported. Read: the AI doubts follow-up.

Earnings Beyond Meta

Tesla’s capital expenditures rose 110% to $4.73 billion while total revenue was roughly flat at $28.05 billion, and income from operations fell 56% to $718 million, the company said in its Q3 2026 Update. Elon Musk said he expects Optimus to reach a run rate of 20,000 units a week in a year or “maybe a little longer,” Axios reported. Read: Tesla’s quarter, explained and the Optimus timeline.

SAP crossed €10 billion of total revenue in a single quarter for the first time, and its current cloud backlog rose 28% to €24.1 billion. SAP put a target on its own AI use: an “efficiency run rate by end of 2028” of more than €2 billion. Read: SAP’s record quarter and SAP’s AI target.

Lam Research reported revenue of $8.22 billion, up 54% from a year earlier by our arithmetic, IBM said generative AI is “increasingly embedded across Consulting engagements representing 35% of backlog”, and Nokia said its net sales to AI & Cloud customers grew 123% on a constant currency basis, to EUR 631 million. Read: Lam, IBM and Nokia.

Spotify expects operating income of EUR 820 million in the fourth quarter, after EUR 672 million in the third, and its net income fell to €518 million from €899 million a year earlier. Its 499 million ad-supported users are 63% of its 788 million total, and its ad-supported revenue is 9.7% of revenue, by our arithmetic. Read: Spotify’s guide, the net income and the free tier.

Labs That Sell Deployment Help

OpenAI published an account of its forward deployed engineering (FDE) team: in one August 2026 example, the first deployment took two months, and extending the same solution to a second team took one day, OpenAI wrote. The post calls the distance between capability and use “the deployment gap”. Read: labs as services firms and the deployment gap, explained.

The same post lists “~48% time saved through AI-based directed test generation” for an unnamed semiconductor customer, says a voice model built with a large telecommunications customer rose from around 60% to 80% and then 90%+ on the customer’s key evals, and says that during its pilot, Tax AI built with Thrive Holdings processed 7,000 tax returns. Read: the chipmaker case, the voice case and Tax AI.

Bloomberg Law reported that law-firm leaders heard Anthropic would not let them negotiate key contract terms around privacy and data security unless they committed to a minimum amount of spending; in a survey cited by Bloomberg Law, 44% of firms said they were using or exploring Claude, but among those firms, just 7% had fully rolled it out. Read: the law-firm terms.

Microsoft CEO Satya Nadella wrote in his annual letter that “The tech stack we are building is model-forward and agent-first.” Read: Microsoft’s stack.

Agents, Security and Safety

Major US banks are detecting unauthorized third-party AI agents operating inside customer accounts, sources told Axios, and of the five banks Axios names, four are blocking or restricting outside agents, by our count. Anthropic says 1 million Agent Skills have been published and nearly 280 million installed, as it donates the standard to the Agentic AI Foundation. Read: the banks and agents and Agent Skills.

METR and Redwood Research estimated that not monitoring compaction events “likely led to on the order of 10% of eligible internal Codex traffic not being monitored between May and July 2026”, and OpenAI told them it has patched the identified vulnerabilities that impact its production safeguards, CNBC reported; METR says it has not evaluated those claims. Read: METR’s audit.

Nine AI-linked security cases were reported between 19 and 22 October 2026; in four of them, by our classification, AI itself was the target. Among roughly 1.1 million internet-facing servers that announce a PHP version, about 758,000 run an end-of-life branch, according to a VirusTotal count. Read: the security week and the PHP follow-up.

Money, People and Policy

DeepSeek has raised about 100 billion yuan ($14.98 billion) in its latest fundraising round, Bloomberg News reported, citing people familiar with the matter. Kevin Weil’s startup, Warp Sciences, raised $215 million at a valuation of $865 million after the new money, Bloomberg’s full report says. Read: DeepSeek’s round and Warp Sciences.

Up to 1.3 billion SpaceX shares become eligible for sale under the lock-up terms in its prospectus; at the $135 IPO price the tranche is worth about $175.5 billion, by our arithmetic. Tesla and Nvidia will put about $2.8 billion into Samsung Electro-Mechanics, South Korea’s Maeil Business Newspaper reported, citing industry sources. Read: the SpaceX lock-up and the substrate deal.

Reuters’ factbox of AI-linked layoffs adds up to about 102,800 announced job cuts since October 2025, by our arithmetic, and Barclays expects to cut about 5,000 full-time roles. A taskforce whose members include Nvidia, Microsoft and Cisco told UK ministers that Britain has “roughly 12 months” to win public consent for data centres, the Guardian reported. Read: the job cuts, Barclays and the UK taskforce.

US District Judge Eumi Lee heard arguments on whether Anthropic’s use of song lyrics to train Claude is protected as fair use, and did not issue a ruling, Reuters reported. Read: the copyright docket.

The Structural Read

Meta had $84.00 billion of notes outstanding at the end of September, against $59.00 billion at the end of 2025, according to its 10-Q, and capital expenditures of $34.91 billion were more than the $32.79 billion of cash its operations generated.

Nebius said it has entered into a $3.5 billion senior secured facility provided by Citi, and the facility “has a 364-day maturity and initially bears interest at Term SOFR plus 2.75% per annum, increasing up to a maximum of 3.50% per annum, and subject to other adjustments.”

As we read it, the labs spent the same day describing what they sell besides models: OpenAI wrote that one deployment that took two months for a first team took one day for a second, and Anthropic says nearly 280 million Agent Skills have been installed.

Three Implications

LENDERS AND BOND BUYERS Bernstein analyst Madison Rezaei, according to TipRanks, estimates CoreWeave needs about $85 billion to fulfil existing customer contracts, leaving an estimated $76 billion to finance using outside capital.

BANKS WITH CONSUMER ACCOUNTS Axios names Morgan Stanley, Bank of America, JPMorgan and Wells Fargo among banks racing to contain agent risks, and reports that Morgan Stanley has begun blocking third-party agents.

TEAMS BUYING AI DEPLOYMENT HELP OpenAI says around 15 FDEs worked with a semiconductor customer on more than 10 use cases from mid-2025, and that all of them reached production within three to six months.

The Business Engineer Lens

This story maps onto the Business Engineer framework The AI Capex Map.

The framework sets out to map the financing of AI: “How the money flows, where it cracks, what it buys, who can absorb it, and what happens when revenue can’t catch up.”

As we read it, Thursday’s coverage touched the first two clauses: interest costs in Meta’s filing, analysts’ estimates of CoreWeave’s funding needs and a Citi facility for Nebius. We do not say one caused the other. The framework was published in May 2026 and does not discuss these reports.

Later on 22 October

Mohsen Rezaei, secretary of Iran’s Supreme National Security Council, said in an interview with the outlet Drop Site that “in the event of a war, cables, data centers, and artificial intelligence facilities could also be in our crosshairs.” Stargate UAE, the 1-gigawatt first phase of the UAE-US AI Campus in Abu Dhabi, is a $30 billion project, Reuters reported on 11 September. Read: the Gulf AI risk.

OpenAI made its Decisions API generally available, at a list price of $0.10 per million input tokens, about 2.4 times the $0.042 that TypeSafe lists for its rival Jev model, by our arithmetic. Read: the Decisions API.

What Is Not Established

This roundup adds no new reporting and selects from the day’s coverage. Every figure above is taken from the linked FourWeekMBA piece, each built from its own primary source. The six threads and the figures in our charts are our grouping, not a finding.

We do not claim that these stories are connected or that one caused another. The Goldman and Bernstein figures measure different things, the DeepSeek and Samsung figures rest on people familiar with the matter or industry sources, and the OpenAI case figures are the company’s own.

The Bottom Line

Meta’s 10-Q shows interest expense of $977 million, up 330%, which the filing attributes to higher long-term debt balances, and our check of eight companies’ filings comes to about $3.56 trillion of commitments.

The day’s coverage also includes OpenAI’s account of its deployment team, banks restricting outside AI agents, and METR’s audit of OpenAI’s agent monitoring, as reported.