Then Dell asked for at least $62m to $70m more to deliver it, according to XTX's court filing in Ireland reported by the FT.
That's 11% to 13% on top of the agreed price (our maths). Dell said it couldn't absorb higher memory costs.
The FT says it isn't a one off. Dell, HPE, Lenovo and Supermicro are asking some buyers for more at delivery on AI server deals signed months earlier. One person told the paper of rises of 25%, 30% and 45%. JPMorgan puts memory cost rises at over 400% since early 2024.
Who carries the risk? The buyer who signed first. Trading firms like Jane Street and Optiver have mostly paid up, the FT says, because they need the kit and there's nowhere else to get it fast.
So the AI bill isn't just the chips. It's whatever memory costs on delivery day.
XTX's side is a claim in a lawsuit, not a ruling.
2
Sources:
FT report via Times of India, 22 Nov 2026 (XTX v Dell filing in Ireland: $549m fixed price, $5m for price certainty, at least $62m to $70m more; suppliers named; 25 to 45% quotes; JPMorgan 400%+ memory since early 2024; trading firms largely paid): timesofindia.indiatimes.com/…
Seeking Alpha summary of FT, 22 Nov 2026: seekingalpha.com/news/467540…
11% to 13% is our sum: $62m and $70m / $549m.
4
New York's power grid will run short from 2033, its grid operator says. And new data centres are the swing factor.
NYISO's new reliability report finds the state's first ever statewide supply gap: 225 MW in 2033, growing to 1,850 MW by 2036.
What's behind it:
2,700 MW of big new loads in the forecast
1,700 MW of that is data centres
2,600 MW of old plant assumed unavailable by 2036
The key line in the report: if that big load growth doesn't turn up, NYISO says there's no statewide gap at all.
And if new data centres cut their use at peak times, the 2036 gap drops from 1,850 MW to 850 MW. More than half gone.
So the size of this problem isn't fixed yet. The real question is whether data centres sign up to flex on the hottest days, or the grid builds for their worst day. NYISO picks the fixes in its next plan.
1
8
Sources:
NYISO press release, 18 Nov 2026 (225 MW in 2033, 1,850 MW by 2036, 2,700 MW large loads incl 1,700 MW data centres, 2,600 MW unavailable by 2036): nyiso.com/-/press-release-ny…
NYISO 2026 RNA report draft, 1 Oct 2026 (no statewide gap if large load growth doesn't materialise; flexible new data centres cut 2036 need to 850 MW): nyiso.com/documents/20142/59…
Anthropic's record IPO pulled in $119.5bn. About $26bn of that came from the firms whose chips and cloud it runs on.
Its closing release names seven buyers of a private share sale at the same $83 price: AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC.
Where the money came from:
$91.6bn from IPO shares Anthropic sold
$26.0bn from those seven companies
$1.9bn to early holders who sold (Anthropic got none of that)
So roughly 22 cents of every dollar Anthropic took in came from its own supply chain (our maths: 26.0 of 117.6).
Why it matters: some of that cash can flow back to them as orders for chips, memory and cloud. That's a known pattern in the AI build, not proof of anything wrong. But it's worth remembering when these firms report AI sales.
The share price is a separate story. ANT closed Friday at $88.91, down 8.5% on the day but still above $83. A falling price doesn't take back cash Anthropic has already banked. The risk sits with anyone who paid more since Thursday.
1
1
9
Sources:
Anthropic, IPO closing release, 20 Nov 2026 ($93.5bn IPO gross, $91.6bn to Anthropic, $26.0bn Class N private placement to AMD, Broadcom, Google, Micron, Microsoft, Nvidia, TSMC at $83): anthropic.com/news/ipo-close
Reuters via Investing.com, 20 Nov 2026 (close $88.91, down 8.5%): investing.com/news/economy-n…
22 cents is our sum: $26.0bn / ($91.6bn + $26.0bn). Gross figures before fees.
2
UK debt interest hit £11.2bn in October. That's the highest October since records began in 1997, and £2.8bn more than a year ago.
The twist: £4.2bn of it was the inflation top up on index linked gilts. RPI rose 0.6% from July to August, and that part of the bill rose with it.
That £4.2bn isn't cash out the door this month. It builds up and gets paid when each gilt matures. But it's still owed.
Put it next to the rest of October's central government spending:
Interest £11.2bn
Net investment £11.1bn
Cash to councils £12.9bn
Tax and other income was up £6.7bn on the year. The extra interest matched £2.8bn of that, about 42p in every extra £1 (our rough maths).
So inflation doesn't only hit the shopping. Each RPI rise also lands on the interest bill. Next ONS update is 22 December.
1
4
Sources:
ONS, Public sector finances UK: October 2026, released 20 Nov 2026 (interest £11.2bn, £4.2bn RPI uplift, receipts £93.6bn up £6.7bn, net investment £11.1bn, transfers to councils £12.9bn): ons.gov.uk/economy/governmen…
42p is our sum: £2.8bn extra interest divided by £6.7bn extra receipts. Figures not adjusted for inflation.
1
Oracle says its OpenAI build is now fully funded. Here's who is paying for it.
This month Oracle raised about $20bn:
86.2m new shares sold at $116, about $9.9bn after fees
$10bn of new bonds paying 7% to 8.65% a year, some running to 2046
Our rough maths: about $777m a year in interest on those bonds alone.
It also says vendor financing is in place, so suppliers are carrying part of the bill until they get paid.
The catch: Oracle still doesn't expect positive free cash flow until the second half of FY2028, roughly Dec 2027 to May 2028. It targets over $30bn in FY2029.
Fully funded means the money is lined up. It doesn't mean the bet has paid off. That rests on OpenAI using and paying for the capacity it ordered.
1
5
Sources:
Oracle statement, 20 Nov 2026 (fully funded, vendor financing, FCF targets): oracle.com/news/announcement…
Oracle, pricing of $20bn stock and notes, 2 Nov 2026 (86.2m shares at $116, five notes 7% to 8.65%): investor.oracle.com/investor…
Reuters via Investing.com, 20 Nov 2026: investing.com/news/stock-mar…
Interest is our sum of coupon x amount for each note. Oracle's fiscal year ends in May.
Wall Street just raised $44bn to buy Nvidia chips for SpaceX. SpaceX won't own them.
Apollo and Goldman announced the pricing today. A trust run by Apollo sells the bonds, buys the new Vera Rubin chips and rents them to SpaceX on a 5 year lease starting in 2027.
Who carries what:
Bond buyers get up to 8.26% a year on the biggest slice, $25.2bn due 2033
The chips themselves are the collateral if SpaceX stops paying
SpaceX pays $7.1bn of rent upfront
IFR reports the debt stays off SpaceX's balance sheet
The catch: ratings firm KBRA warns GPUs lose value fast, so nobody knows what the chips would fetch in a default. One fund manager told IFR it is really SpaceX risk dressed up as asset backed debt, and passed.
So if SpaceX keeps paying, bond buyers collect 8% plus. If it doesn't, they get racks of chips that are worth less every year.
1
9
Sources:
Apollo, pricing of $44bn SpaceX GPU financing, 20 Nov 2026: ir.apollo.com/news-events/pr…
SpaceX, $40bn GPU finance lease, 26 Oct 2026: ir.spacex.com/updates/releas…
IFR, 20 Nov 2026 (tranches, 8.259% yield, $7.1bn upfront rent, KBRA and fund manager quotes): ifre.com/bonds/2516476/yield…
2
UK retail sales jumped 0.7% in October. A big chunk of that was tins and UHT milk.
ONS says supermarkets sold more long life food after the official Be Ready advice on 22 Oct to keep 3 days of supplies at home.
What the numbers show:
Food store volumes up 0.8%, after a 0.3% drop in September
Clothing up as the weather turned colder
Fuel volumes down, with pump prices up for a third month
Every economist in the Reuters poll expected less
The catch: shoppers spent 1.2% more in cash but took home only 0.7% more stuff. Our rough maths says about half a point of the rise was simply higher prices.
So this looks like a stock up, not a spending boom. You only buy a 3 day stash once, and accountants MHA call it probably a one off. The real test is Christmas, with inflation at 3.6% and forecasters ITEM Club expecting energy prices to rise more than 20% in January.
1
9
Sources:
ONS, Retail sales GB, October 2026 (20 Nov): volumes +0.7% m/m, +2.3% y/y, total spend +1.2%, Be Ready campaign, fuel: ons.gov.uk/businessindustrya…
Reuters, 20 Nov 2026: food stores +0.8%, poll, CPI 3.6%, MHA and ITEM Club quotes: investing.com/news/economy-n…
Price gap is our rough maths (value minus volume), not an ONS figure.
South Korea just sold $1 trillion of goods abroad in a single year. Only the US, Germany and China got there before.
The engine is AI memory chips.
January to October:
Chip exports: $396.8bn, up 193% on a year ago
That's 42.7% of everything Korea exported
Everything else: up 20%
Our rough maths from those figures: chips added about $261bn of extra sales, the rest about $89bn. So roughly 3 in every 4 new export dollars came from chips.
A big slice of that is price, not just volume. The trade ministry says higher DRAM and NAND prices pushed the numbers up, and it flags chip price swings as a risk.
Winners so far: Samsung, SK hynix and Korea's trade surplus, $294.9bn in 10 months, triple the old full year record from 2017.
The catch is the same thing. When memory prices turn, a whole country's headline turns with them.
Worth watching memory contract prices, not the milestone.
1
2
Sources: Yonhap, 19 Nov 2026, ministry data on $1tn milestone, chips $396.8bn (+193.1%), 42.7% share, $294.9bn surplus: en.yna.co.kr/view/AEN2026111…
Korea Times, 19 Nov 2026, non-chip exports +20%, memory prices: koreatimes.co.kr/economy/pol…
3 in 4 is our estimate from rounded figures. Not investment advice.
1
Alibaba's profit tripled to $9.6bn. Its cash went the other way.
Three months to September:
Cash from running the business: $2.3bn
Spent on AI kit and buildings: $7.5bn, up 60%
Free cash flow: minus $5.1bn
So where did the profit come from? $10.9bn of interest and investment income, mostly paper gains on shares it holds in other companies. Alibaba strips those out of its own adjusted profit, which was $3.0bn.
The AI cloud is growing fast. Sales hit $9.1bn, up 53%, and its profit more than doubled to $1.3bn. The Qwen AI apps lost $1.6bn.
The cash gap was covered by new shareholders. Alibaba sold HK$80bn (about US$10bn) of new shares in the quarter and paid $2.5bn in dividends.
With $72.8bn of cash and liquid investments, this isn't a crunch. But for now the AI build runs on investors' money, not cloud profits.
Next test: the positive free cash flow it says it's now aiming for.
1
9
Sources: Alibaba Group September quarter 2026 results, 19 Nov 2026 (US$ at company rate RMB6.7045): alibabagroup.com/en-US/docum…
Profit, cash flow, capex, investment income, cloud and AI apps figures all from that release. HK$80bn placing is in the financing cash flow section.
Not investment advice.
1
Energy price cap forecast for January: £2,166 a year. That's £443 more than today's £1,723.
Most of it is gas, and almost all of it is in the energy you actually use.
Typical home, Cornwall Insight forecast vs today's Ofgem cap:
Gas: £865 to £1,149, up £283 (33%)
Electric: £858 to £1,017, up £159 (19%)
Gas unit price: 7.97p to 10.90p a kWh
Electric unit price: 26.32p to 32.54p a kWh
Standing charges: up about £8 a year between them
So roughly 98% of the rise is in what you burn, not the daily fee. Use half the typical energy and you'd see roughly half the rise. A big house heated by gas sees more.
In October the same forecaster had January at £1,999. Wholesale gas has added £167 since.
For scale, food prices rose 0.8% in a year to October (ONS), the slowest since 2021.
This is a forecast. Ofgem confirms the real cap before it starts on 1 January.
1
9
Sources:
Cornwall Insight final January forecast, 18 Nov 2026 (gas £1,148.80, electric £1,017.09, unit rates and standing charges): cornwall-insight.com/press-r…
Today's cap, Ofgem, 1 Oct to 31 Dec 2026, Direct Debit: ofgem.gov.uk/energy-price-ca…
ONS CPI October 2026, food 0.8%: ons.gov.uk/economy/inflation…
Per fuel splits and the 98% are our maths at typical use (2,500 kWh electric, 9,500 kWh gas).
2











