Technology · 14 November 2026
The Same Build-Out, Three Sets of Books
On Friday the Federal Reserve named artificial intelligence the single biggest risk to the American financial system. On Thursday Britain reported its fastest quarterly growth in two and a half years and credited AI for much of it. In Lagos, operators are raising dollar project finance for AI data centres in the same month Wall Street quietly stopped lending for them. One build-out, three different accounts of what it is.
Three institutions looked at the same activity this week and wrote it down three different ways. A central bank called it the largest risk to its financial system. A statistics office called it the main engine of national growth. A lending market called it an asset class to stop underwriting. All three are looking at AI infrastructure, and none of them is wrong.
Washington put a number on it
The Fed published its semi-annual Financial Stability Report on Friday 13 November. New York Fed staff asked 23 contacts, across broker-dealers, banks, investment funds and advisory firms, to name the salient risks to US financial stability. Eighty-seven percent named AI, against 50 percent in the spring survey six months earlier.
Federal Reserve, Financial Stability Report, November 2026, Box 5.1, Survey of Salient Risks to Financial Stability. 23 contacts surveyed in September and October 2026.
Table of values
| Risk cited | Autumn 2026 | Spring 2026 |
|---|---|---|
| AI | 87% | 50% |
| Risk assets | 52% | 35% |
| Long rates | 48% | 20% |
| Cyberattacks | 43% | 20% |
| Fiscal debt | 43% | 30% |
Worth reading the chart carefully, because the Fed adds a distinction the headline loses. AI was the most-cited risk. It was not the most feared one: respondents thought cyberattacks and fiscal debt sustainability carried the more severe consequences if they happened. AI is the thing most people are watching, not the thing most people expect to hurt most.
The report groups the AI concern into three parts: stretched equity valuations where earnings disappoint, cybersecurity, and complex forms of AI-related financing. The third is where the report stops being a survey of opinion and starts being arithmetic.
Debt issuance tied to advanced computing passed $680 billion by October 2026, a fivefold increase on the $130 billion issued in all of 2023. Outstanding public debt related to it is estimated above $1.1 trillion. The number that frames both: AI-related borrowing has gone from under 6 percent of nonfinancial investment-grade corporate debt issuance in 2023 to about 31 percent in 2026. Close to a third of America's high-grade corporate borrowing is now going into one build-out.
The Fed also notes hedge fund leverage at its highest since comprehensive data began in 2013, concentrated in a small number of large funds.
Two caveats the report makes itself. Its data cut off on 28 October, so nothing in November is in it. And it finds the system fairly stable overall; it does not forecast a bad ending. But a central bank putting a trillion-dollar figure on one sector's financing, in a stability report, is a different act from an analyst writing about a bubble.
Then the last inflation reading before the Fed votes. October CPI landed on 10 November: headline up 0.4 percent on the month against a 0.3 percent consensus, core holding at 0.3 percent against 0.2 percent expected, its third month running at 0.3. There is no annual rate this month, because the BLS never collected October 2025 prices during the shutdown. On the same day Governor Waller said a half-point increase should be considered in December if core came in around 0.3 percent or higher. It came in at 0.3. The 10-year Treasury yield went back to 5 percent. The FOMC meets on 8 and 9 December, and the next CPI lands on 10 December, the day after the vote.
What a US reader should take from this. The December meeting is now the live one and the committee walks into it without another inflation print. If you hold concentrated technology exposure, the Fed has told you in writing that the market's biggest perceived risk is the thing you own, and that its financing is the weak joint rather than its earnings.
London booked it as growth
The ONS published its first estimate of third-quarter GDP on Thursday 12 November. The economy grew 0.7 percent on the quarter, up from 0.5 percent in the second, and 1.9 percent on the year. It is the fastest quarterly figure in two and a half years.
Services did nearly all of it, expanding 0.9 percent. Within services, information and communication rose 2.1 percent, and inside that, computer programming, consultancy and related activities rose 4.4 percent. The ONS says plainly that this industry has been the largest driver of GDP growth across 2026, and links it to firms working in AI and cloud computing. Production managed 0.1 percent and construction fell 0.7 percent.
The Bank of England had already upgraded its 2026 growth forecast to 1.6 percent alongside last week's rate rise, well above the 1.2 percent the OBR used for the October Budget, and attributed part of the strength to AI spending.
So the activity the Fed counts as a systemic risk sits, in Britain's national accounts, in the sector the ONS identifies as its main engine of growth.
The ONS adds a caveat worth carrying: it cannot measure the AI effect precisely. It can see the turnover and it is attributing the cause.
What a UK reader should take from this. Be careful how much weight you put on that number. A 4.4 percent quarterly rise in computer programming and consultancy is a capital-spending cycle showing up as output, and capital-spending cycles are the thing the Fed is worried about financing. If that spending slows, the growth figure that justified the Bank's upgrade gets revised down at the same time borrowing costs are going up.
Lagos is being asked to fund it in dollars
The part nobody is connecting ran in BusinessDay on 9 November. Major commercial banks and private credit lenders have begun quietly pulling back from financing new AI data centres ahead of what the report describes as a $13.4 billion AI debt wall. Syndicate desks that were competing to underwrite hyperscale campuses weeks earlier now have credit committees stalling deals, including for facilities backstopped by blue-chip technology names.
Meanwhile Lagos is building. Open Access Data Centres is executing a $240 million expansion of its Lagos facility, designed for up to 40 megawatts of AI capacity. MTN Nigeria has committed over $240 million to its Ikeja site. Equinix is rolling out LG3. All of it is dollar-denominated project finance, and all of it also requires off-grid power and cooling, which is capital on top of capital.
The demand is real. Speaking at the ITW Data Cloud Africa summit, NITDA director-general Kashifu Inuwa Abdullahi said federal agencies spent nearly $2.9 billion on technology between 2023 and mid-2026, roughly $1 billion of public IT demand a year.
Read the BusinessDay detail again, because the part that matters is easy to miss. The deals being stalled include facilities backstopped by blue-chip technology names. If lenders are hesitating on those, an operator without that kind of backing is in a weaker position still. Nigerian builders are arriving at the dollar project-finance market at the exact moment it is tightening for this asset class.
At home, money stayed expensive. At Wednesday's Treasury bill auction the DMO offered ₦800 billion and drew ₦1.12 trillion in subscriptions, a bid-to-cover of 1.40 times, allotting ₦713.30 billion. Almost all of that appetite was for one tenor: the 364-day alone drew ₦1.079 trillion against a ₦600 billion offer and cleared at 15.89 percent, up from 15.80 the week before. That leaves roughly ₦41 billion of bids across the 91-day and 182-day against ₦200 billion offered. The market is not short of money. It will only lend it for a year.
What a Nigerian reader should take from this. If you are anywhere near the data-centre build, in construction, power, cooling, fibre or the services around them, the financing environment for your client's project changed in the last three weeks and your client may not have told you. Ask what stage the funding is at before you staff up or extend credit. And for anyone borrowing in naira, the one-year rate ticked up rather than down, so the cheapening that ran from August into late September is still not coming back.
Nairobi would not take the swap
Kenya ran a Treasury bond switch auction on 9 November. It invited holders of a 2023 five-year bond, which now has about 1.7 years to run, and of 91-day, 182-day and 364-day bills, to move into a bond issued in 2019 with roughly three years left to maturity. It advertised KSh10 billion and received KSh3.79 billion, a performance rate of 37.9 percent. The CBK accepted everything bid.
Be precise about what that is and is not. The target has about three years remaining, not thirty, so this is a modest extension rather than a test of long duration, and low uptake at switch auctions is common because participation is voluntary and confined to existing holders.
What makes it worth noting is the contrast on the other side of the curve. Three days later the weekly bill auction drew KSh69.3 billion against KSh28 billion offered, about 247 percent. The shilling sat at KSh129.96 to the dollar on 12 November, unchanged from a week earlier.
So the same market that would not move from under two years out to three, at the price offered, was more than twice oversubscribed for three-month paper in the same week.
What a Kenyan reader should take from this. Investors are being paid enough at the short end that they will not give up liquidity for a modest extension. That is a judgment about relative value, not a verdict on Kenyan credit. If you hold bills and have been weighing a move further out for the extra yield, note that the institutions looked at a three-year version of that trade and declined it at this price.
Next week the financing gets tested twice
The Fed's report is a judgment about whether the money keeps flowing. Two events in the following week put that to a vote.
Nvidia lost its chief executive. The board accepted Jensen Huang's resignation as president, chief executive and director on 11 November, effective immediately, and announced it on 12 November. Colette Kress, the chief financial officer, becomes interim chief executive while retaining the CFO role. The company gave no reason for the departure. It followed reports on 8 and 9 November that the board had opened an investigation into him. The allegations in those reports are unestablished and are not detailed here. The board says it expects operations, customer commitments and the product roadmap to continue uninterrupted, and is considering internal and external candidates. Nvidia reports third-quarter results for fiscal 2027 on 17 November.
Whatever the cause, the practical fact is that the company at the centre of the AI supply chain reports earnings on Tuesday under an interim leader, with no permanent successor named.
And one of the largest listings ever attempted prices the following day. On 10 November Anthropic set a range of $70 to $78 a share for its Nasdaq listing under the ticker ANT: 825 million Class A shares, with an underwriters' option on a further 123.75 million. At the top of the range that raises about $64.4 billion, and the company's own filing puts the implied valuation near $1.4 trillion fully diluted. Pricing is set for 18 November, first trade for 19 November.
For scale, SpaceX raised $75 billion in June and $85.7 billion once its underwriters exercised their option, against Saudi Aramco's $29.4 billion in 2019. So Anthropic at the top of its range, with the full over-allotment taken, reaches about $74 billion: second on record, close to what SpaceX raised at pricing, still clearly below what SpaceX raised in the end. Reuters had earlier reported the company was seeking as much as $100 billion, so the range as set is well below what was once discussed.
The filings show third-quarter revenue of about $17.0 billion against $1.4 billion a year earlier, and $33.2 billion across the first nine months, so more than half the nine-month figure landed in the third quarter alone. They also show a third-quarter operating loss of $567 million, following an operating profit of $974 million in the second. Revenue accelerating and operating profit going backwards in the same quarter is the AI build-out in one company's accounts.
So the sequence is this. The Fed says the risk is in the financing. On Tuesday the sector's most important supplier reports earnings without a permanent chief executive. On Wednesday the market is asked to absorb one of the two largest equity raises ever attempted, from a company in the same sector.
Neither event tells any individual project whether it gets funded. What they do is set the price of appetite for AI risk, and everything still looking for money is quoted off that price, whether it sits in Virginia, Slough or Ikeja. By Thursday evening that appetite will have been marked twice in public.
What actually happened this week
The AI build-out stopped being a technology story and became a credit story.
Britain is counting it as growth. Washington is counting it as the largest single risk to its financial system. Both are reading the same activity, and both are right about what they see, because a capital-spending boom is genuine output while it is being financed and a genuine problem when it is not.
The Fed's own numbers reconcile them, and they split in a way worth getting right. Beyond the $1.1 trillion already borrowed sit more than $1 trillion of leases that have not yet commenced, and more than $2 trillion of take-or-pay contracts and residual value guarantees. Only the first of those is optional. A lease that has not commenced can be left uncommenced. A take-or-pay contract is a promise to pay whether or not you take delivery, which is precisely what it is for.
That is the real shape of the exposure. Roughly a trillion dollars of this build-out can still be called off, and roughly two trillion cannot. Wall Street has started calling off what it can, and the first projects to feel it are the ones furthest from a balance sheet that can self-fund.
Lagos is one of those places. It is building AI capacity with dollar project finance, for real domestic demand, into a lending market that has just begun to close for exactly this asset class. That is not a reason not to build. It is a reason to know where your funding actually stands.
And the week ends with the question unusually well framed. The central bank says the financing is the weak joint. On Tuesday the sector's largest supplier reports under an interim chief executive. On Wednesday a listing second in size only to SpaceX asks the market for sixty-four billion dollars. Rarely does a thesis get examined this quickly.
Method
Every figure comes from a primary release or a named report, linked below. Where sources disagree we give the range and say why. Where a number will not stand up we leave it out. Every call is dated, and the next edition grades it.
Written on 14 November 2026, covering 8 to 14 November. Every anchor falls inside that window.
Fed figures are read from the report itself rather than from coverage. It separates not-yet-commenced leases, above $1 trillion, from take-or-pay contracts and residual value guarantees, above $2 trillion; accounts giving a combined $3 trillion of commitments not yet begun merge the two. There is no October US annual inflation rate because October 2025 prices were never collected during the shutdown, and the ONS states that it cannot measure the AI contribution precisely.
The roughly ₦41 billion bid across Nigeria's 91-day and 182-day is a residual: ₦1.12 trillion of total subscriptions less ₦1.079 trillion on the 364-day. Kenya's switch targeted FXD4/2019/010, which has about three years left to run rather than the ten its original tenor implies.
Reports of a board investigation at Nvidia concern allegations that are unestablished and are not detailed here. Anthropic's pricing and listing dates can move. SpaceX raised $75 billion at pricing and $85.7 billion after its underwriters exercised their option, against Saudi Aramco's $29.4 billion in 2019.
The 8 and 9 December FOMC meeting, the 10 December CPI release and the 17 December Bank of England decision are scheduled future events.
What would prove this wrong
If the Anthropic offering prices at or above the top of its range on 18 November and holds that level through its first week, the market's appetite for AI equity risk is intact and the Fed's financing concern is early. If it prices below the range or breaks issue, the concern is not early at all, and every AI project still looking for money reprices behind it. If the next round of large AI data-centre financings closes on normal terms through December, the lending pullback was a short repricing rather than a withdrawal. If UK computer services output holds up in the October and November monthly estimates, Britain's growth composition is sturdier than this reading allows. If the Fed holds on 9 December after an October core print at exactly Waller's stated threshold, the hawkish read of this week is wrong and the committee is more tolerant than its own speakers suggest. And if Kenya's next conventional long-dated auction clears with solid cover, the weak switch uptake was about the mechanics of that instrument rather than any reluctance to extend.
Next move
- CareerThis week was unusually legible if you work anywhere near computing infrastructure, and you do not need to be in finance for it to matter. In Britain, computer programming and consultancy is the industry the ONS now calls the largest driver of growth across 2026, which is where the hiring is. In Nigeria, the data-centre build is creating demand for high-voltage electrical work, cooling, structured cabling and facilities management, and those are trainable skills with a visible pipeline in Lagos right now. But take the financing warning seriously in how you choose. A project with committed funding and a signed anchor tenant is a different employer from a project that has announced an investment figure and not named its investors. Ask which one you are joining, because the lenders have started asking.
- Business owners and operatorsTwo concrete things. If any part of your business depends on a data-centre project completing, whether you are a supplier, a contractor or a tenant planning to move workloads, verify the funding status before you commit capacity or inventory. Lenders stalling deals is not the same as cancelling them, but it moves timelines by quarters. And if you are raising dollar project finance for anything capital-intensive in Nigeria or Kenya, the window on this asset class is narrowing while your naira funding costs are not improving, so bring forward whatever you can close now rather than optimising terms into the first quarter.
- InvestorsThe Fed has told you where it thinks the weak joint is, and it is not valuations, it is financing. If you hold AI or semiconductor exposure through a global broker or a dollar app, which many Nigerians and Kenyans now do, the thing to watch is not the earnings reports but whether the debt market keeps funding the build. That gets tested on 17 and 18 November, when Nvidia reports and Anthropic prices, and the honest position is that nobody knows which way they go. Note too what the Fed separated: about a trillion dollars of this build-out sits in leases that have not commenced and can still be called off, and about two trillion sits in take-or-pay contracts and guarantees that cannot. A slowdown shows up in the first number long before the second. Closer to home, Nigeria's one-year bill at 15.89 percent is slightly better than last week and still the crowded trade, with ₦1.079 trillion chasing ₦600 billion while the shorter tenors drew almost nothing. And in Kenya, the institutions declined a switch into paper with about three years left while bidding three-month bills at well over twice the amount offered, which tells you where they think the value is.
Sources
- Federal Reserve, Financial Stability Report, November 2026
- CNBC, AI risks are the biggest threat to the financial system, Fed survey finds
- US Bureau of Labor Statistics, Consumer Price Index, October 2026
- CNBC, 10-year Treasury yield races back to 5% as inflation data and Waller comments drive rate-hike expectations
- Office for National Statistics, GDP first quarterly estimate, UK, July to September 2026
- BusinessDay Nigeria, Wall Street quietly freezes data centre loans ahead of $13.4bn AI debt wall
- Nairametrics, NTB investors pour N1.08 trillion into the 364-day bill
- Nairametrics, DMO offers N800 billion in T-bills as N148.6 billion matures
- TechCrunch, Nvidia's Jensen Huang resigns, chipmaker names CFO Colette Kress as interim CEO
- CNBC, Anthropic sets IPO price range of $70 to $78, looks to start trading 19 November
- CNBC, SpaceX raises $75 billion in record-setting IPO ahead of Nasdaq debut
- Forbes, SpaceX says historic IPO raised more than $85 billion
Cite this Signal
ZeroToAct, The Same Build-Out, Three Sets of Books, 14 November 2026, https://www.zerotoact.com/signals/same-buildout-three-sets-of-books/
Disclosure
Tolu Adetuyi is co-founder and Chief Innovation Officer of Prembly, which builds identity and compliance infrastructure. Signals regularly cover payments, identity and regulation, which is his commercial interest as well as his subject.