For years, the AI boom was described in terms of what models could do. On 2 November, it was described in terms of what they cost. A $115 billion deal for lasers and optics. About $895 billion in promised spending on computer power. And the world's most valuable phone maker warning that memory — the humble component that stores your photos and apps — is now expensive enough to squeeze its profits. Taken together, those three numbers tell a single story: AI is no longer just a research race. It is an infrastructure bill, and it has arrived.
🤖 30-Second Signal — read this, you get the story
⚡Marvell is buying Lumentum for $115 billion in enterprise value.
🎯Anthropic disclosed about $895.4 billion in compute commitments — about $213.8 billion of it is contingent.
🔭Apple made $114.5 billion in a quarter, but warned margins will fall on memory costs.
Tap to reveal what happens next →
Regulatory approvals for the Marvell deal, and whether memory costs keep biting Apple through the holiday quarter.
Tap, scroll and watch the numbers — the full story, with sources, is below. All figures are from the article, no new claims added.
Apple first: your next iPhone could cost more to make
Start with Apple, because it is the part of this story most readers will feel directly. Apple reported fiscal fourth-quarter results on 2 November, in the first report under chief executive John Ternus, who succeeded Tim Cook on 1 September. Revenue was $114.5 billion, up about 12% year-on-year, with earnings per share of $2.10, up 14%, beating analyst estimates. iPhone revenue was $57.58 billion. Gross margin was 47.9%, including about one point from tariff refunds.
Then came the warning. Apple guided December-quarter gross margin to 45–46%, mainly because of higher memory costs plus costs associated with Siri AI. Memory had already shaved about one point off margin from June to September, excluding refunds, and the chief executive called memory pricing unprecedented. Purchase obligations rose by $33 billion in the quarter as Apple locked in parts. December revenue was guided up 7–10%, and services up 8–9%, below expectations in that segment.
Marvell buys Lumentum for $115 billion in enterprise value
The largest deal of the week was Marvell's agreement to acquire Lumentum, announced on 2 November. It is an all-stock acquisition with an implied enterprise value of $115 billion, making it one of the largest tech deals on record. Lumentum shareholders would receive 4.5 Marvell shares per Lumentum share, a premium of about 23.9% to Lumentum's unaffected close on 28 October. Marvell shareholders would own about 67% of the combined company and Lumentum shareholders about 33%. Chief executive Matt Murphy would stay on, and the deal is expected to close in the first half of calendar 2028, subject to shareholder and regulatory approvals.
The logic is about AI data centres. Marvell makes chips. Lumentum makes lasers and optics — the components that move data between servers at high speed using light. Combine them, and you get a company that can sell both the computing and the connectivity inside an AI data centre. The combined company put its 2030 addressable market at $550 billion and claimed more than $1 billion in annual run-rate cost synergies. Lumentum also reported fiscal first-quarter 2027 revenue of $1.26 billion, up 136.5% year-on-year, on the same morning — a reminder that its business was already growing fast before the deal.
One figure in this section needs care. Outlets have reported conflicting equity values for the deal, around $80 billion in one report and $102.54 billion in another. This article therefore quotes only the $115 billion enterprise value, which is the figure consistently used for the deal's scale, and does not assert an equity value that sources disagree on. Enterprise value and equity value are different measures, and mixing them is how a $115 billion deal becomes a $80 billion deal in a headline.
Anthropic: $895 billion in promises, with a large caveat
Anthropic, the company behind the Claude AI models, disclosed in an amended filing on 2 November aggregate purchase commitments and leases of about $895.4 billion, substantially all for compute and infrastructure. About $447.5 billion is due before 31 December 2030, and about $69.8 billion is due within 12 months after 30 September 2026, with the remainder mostly in 2031 to 2045. Its third-quarter 2026 revenue was $16.96 billion, and nine-month revenue was $33.2 billion, with a third-quarter operating loss of $567 million.
That $895 billion figure is staggering, but it comes with a caveat that must be in any honest account of it. About $213.8 billion is tied to arrangements with termination rights or contingent conditions — financing, facilities or approvals that may not all materialise. A further $150–175 billion of arrangements without fixed amounts are not included in the $895.4 billion at all. In other words, the headline is the maximum scale of commitments as disclosed, not a single, unconditional cheque. It shows how much compute Anthropic believes it needs, and how far in advance it is trying to lock it in, but it is not the same as cash already spent.
Palantir grows 96%, and shares still fall — the twist
Palantir reported third-quarter revenue of $2.313 billion, up 96% year-on-year, its first quarter above $2 billion, beating estimates of about $2.18 billion, with adjusted earnings per share of $0.44 against $0.41 expected. US commercial revenue was $984 million, up 148%, passing US government revenue of $953 million, up 96%, for the first time. Full-year 2026 guidance was raised to $8.530–8.534 billion, about 91% growth.
And yet shares fell after the results. The twist is in contract value, not revenue. US commercial total contract value grew only 11% year-on-year, and total contract value was $2.918 billion, up 6%. Revenue reflects work already done and billed. Contract value reflects new business signed for the future. When revenue soars but new contract growth is modest, investors worry that the current pace cannot be sustained. That is why a company can beat expectations and still fall — the market was pricing in not just this quarter, but the next several, and the contract numbers suggested a slower future than the revenue numbers suggested a present.
OpenAI in Abu Dhabi: four agreements, no price tag
OpenAI signed four agreements in Abu Dhabi on 2 November, witnessed by Crown Prince Sheikh Khaled bin Mohamed bin Zayed Al Nahyan after a meeting with chief executive Sam Altman. The agreements covered: ADNOC, with G42 and Inception42, described as the first energy-sector company to deploy OpenAI Frontier, on in-country UAE inference; the Abu Dhabi Government and OpenAI on a Machine-Readable Government initiative, with a first pilot live with the Abu Dhabi Agriculture and Food Safety Authority for food imports; Mubadala, Inception42 and OpenAI for investment analysis and risk; and a five-year agreement between Inception42 and OpenAI for frontier-AI adoption across government, enterprise and education.
What the bill means
Whether that money is well spent will be judged by products, revenues and margins in the years ahead, not by the size of the commitments themselves. A $115 billion deal can fail on integration. A $895 billion commitment can be renegotiated. A margin warning can be temporary. But as a snapshot of 2 November 2026, the bill has arrived — and it is larger, and more consequential for ordinary consumers, than most people realised.
How big is the Marvell–Lumentum deal?
The deal has an implied enterprise value of $115 billion, all-stock, with 4.5 Marvell shares per Lumentum share. This article quotes only the enterprise value because outlets conflict on the equity value. Closing is expected in the first half of 2028, subject to approvals.
Did Anthropic really commit $895 billion?
Anthropic disclosed about $895.4 billion in aggregate purchase commitments and leases, substantially for compute, in an amended filing on 2 November. About $213.8 billion of that is tied to contingent arrangements, so it is not all unconditional spending.
Why did Apple warn about memory costs?
Apple guided December-quarter gross margin to 45–46%, mainly on higher memory costs plus Siri AI costs, after memory already shaved about a point off margin from June to September. It called memory pricing unprecedented, reflecting high demand partly driven by AI data centres.