Tuesday, September 29, 2026

A new Brookings paper by Columbia's Stijn Van Nieuwerburgh walks through the deal behind Meta's Hyperion campus. Key observations:

1. Meta sold an 80% stake to Blue Owl for about $2.5bn. The joint venture then borrowed $27bn at 6.58%, about 90% of the asset's value.

2. That yield sits at least 100 bps above what Meta would likely pay on its own debt.

3. Meta leases the campus in five four-year terms from 2029 and guarantees the assets' resale value. Current accounting keeps both off its balance sheet until they start.

4. The template is spreading. Meta's El Paso venture with BlackRock sold $12.3bn of A+ bonds in July.

The setup: Meta buys flexibility and an asset-light balance sheet. Bondholders hold Meta credit one step removed. The test comes in 2029: if Meta walks away from part of the campus, the guarantee turns into a bill.

Morgan Stanley maps what now slows the AI build, citing the tally from Data Center Watch. Key observations:

1. Local opposition is now law in 2 states. Pennsylvania and Massachusetts make new data centres bring their own power or pay into a ratepayer fund.

2. Texas is auditing 474GW of requests in its grid queue. About 90% of them are data centres.

3. Morgan Stanley sees a 57GW power shortfall for US data centres over 2026-28. On-site gas, fuel cells, nuclear sites and converted bitcoin mines cut it to 33GW. That still leaves about a third of the chips without power.

4. Grid hookups already take 5-7 years in some regions.

5. Data-centre emissions reach about 1.2 gigatonnes a year by 2030, about as much as Brazil. The bank's forecast has nearly doubled.

Who gets paid: owners of existing power, such as Vistra and Talen. Makers of on-site equipment gain too, such as GE Vernova and Cummins. The risk is federal action. Without it, Morgan Stanley sees no large-scale threat to the build.

JPMorgan's biotech team flags what the headline number leaves out. Key observations:

1. The drug, UBT251, comes from China's United Laboratories. It acts on 3 hormone receptors: GLP-1, GIP and glucagon.

2. The 19.7% came from a China-only Phase 2 trial, while Lilly's 18% came from a global one. JPMorgan notes China-only trials have sometimes shown better tolerability.

3. Global Phase 1b/2a data arrive in the first quarter of 2027.

The setup: Novo Nordisk holds a drug that kept pace with Lilly's next candidate in an early test. JPMorgan expects the market to price it only after the global data. A weaker global readout breaks the case.

Morgan Stanley's securitisation team looks at who keeps those gains as rates rise. Key observations:

1. JLL puts rent growth at about 9% a year, with 95% of new capacity leased before it is built.

2. CBRE saw rates for 3-10MW deployments rise 8.3% in the first half of 2026.

3. Hyperscaler campuses sit on 10-15 year leases with rent increases of about 2% a year. Colocation landlords reset rents as leases are signed or renewed.

4. A campus leased to a hyperscaler earns about 7.75% before debt. Borrowing above that rate eats the owner's return.

The setup: Morgan Stanley prefers colocation data-centre bonds for that reason. My read: in stocks, the same logic favours colocation landlords such as Equinix and Digital Realty. It fades once new power lets supply catch up.

Morgan Stanley's commodity strategist Amy Gower sizes AI's pull on copper. Key observations:

1. Data-centre copper demand rises from about 720kt in 2026 to 1.15Mt in 2028.

2. About 75% of it goes into power equipment, where copper is hard to replace.

3. The bank sees a deficit already this year: 27.7Mt of demand against 27.5Mt of supply. More deficits follow in 2027 and 2028.

4. Newer data-centre estimates could lift the 2028 figure by about 70%.

My read: the listed exposure is the copper miners, such as Freeport-McMoRan and Southern Copper. The view breaks if new 800V power designs cut the copper each data centre needs.

Morgan Stanley's European strategists argue Europe's earnings story is misread. Key observations:

1. Autos are just 1.5% of MSCI Europe. Yet they drove the whole "Europe has no growth" story.

2. In 2025, almost 30% of the index sat in sectors with falling earnings. This year almost every sector is growing.

3. Consensus sees European earnings up about 20% in 2026, or about 15% excluding energy. Volkswagen's write-off still has to come out, about 1.3 points.

4. Financials, real assets and compute make up 64% of European earnings. That group tends to earn more when inflation runs hot. Only 26% of earnings depend on the consumer.

The thesis: Morgan Stanley says Europe has broken a 10-year slide in its valuation against the US. Past breakouts took the discount to single digits. My read: is the simple way in. It breaks if oil slumps: the bank estimates $60 Brent would cut 2027 earnings growth by 1.7 points.

Morgan Stanley's sales desk and its European strategists see a seasonal setup forming. Key observations:

1. The bank's defence coverage booked €87bn of orders in the first half of 2026.

2. Its prime brokerage data to 24 September show funds pressing shorts in European aerospace and defence.

3. German defence orders are running below last year and below target. Morgan Stanley expects a marked pick-up into year-end. The sector has rallied from about mid-November every year since 2022.

4. France raises its defence budget by 13% this year and 11% next, per Morgan Stanley Investment Management.

The setup: Morgan Stanley's preferred names are BAE Systems, Rheinmetall, Leonardo, Saab and Indra. The seasonal case breaks if German orders do not pick up by year-end.

Morgan Stanley analyst Terence Tsui just turned negative on Ericsson. Key observations:

1. He cut Ericsson to Underweight and its target to $9 from $11.

2. Nokia's rally has orders behind it: a record €2.8bn order book in the second quarter, almost three times the first.

3. Morgan Stanley rates Nokia Overweight with a €14 target.

The setup: Morgan Stanley favours the company with AI orders over the one riding its coattails. The pair breaks if Ericsson starts booking data-centre networking orders.

A Morgan Stanley sales note tracks how far the index and the average stock have split. Key observations:

1. The record counts stocks with a negative beta over three months, meaning they move against the index. In Europe the share is 11% of the Stoxx 600.

2. The S&P 500 beat the small-cap Russell 2000 by 11% in three months.

3. Meta is up 36% this month, its best month in 13 years.

4. Jay Pelosky, a Morgan Stanley alum quoted in the note, finds the average stock's link to the 10-year yield at its most negative since 1960.

5. Oil and the 10-year yield now move together more tightly than at any time in 35 years.

6. Hedge funds lean the index's way. US long-short net leverage rose to 54% as funds bought semis, memory and agentic AI.

The index rides AI earnings while the average stock trades like a bond. Pelosky's trigger for the laggards is a US-Iran deal that reopens Hormuz and pulls oil and yields lower.

Morgan Stanley's economists and oil strategist track what happens as the buffer runs out. Key observations:

1. The Strategic Petroleum Reserve holds 284.6mn barrels. Draws of about 30mn barrels every four weeks reach the roughly 70mn operating floor by March 2027.

2. Each monthly draw covers only about 6% of US crude use.

3. Diesel trades about $100 a barrel above Brent. The gap has widened since early July.

4. Strategist Martijn Rats says a full halt of US diesel exports could cut refinery runs by about 2mn barrels a day. Gasoline supply would drop by about 650,000 barrels a day.

My read: with the reserve near its floor, fuel-export curbs become the next lever. Morgan Stanley ranks Valero and Delek most exposed, Suncor and Imperial Oil best placed. It sees a partial curb as possible and a full ban as unlikely.

Morgan Stanley's European strategists break down who did the lifting. Key observations:

1. The index is up about 10% in local currency. Banks, semis, capital goods and energy did most of the work.

2. Direct AI capex beneficiaries are about 13% of the index but drove 45% of its gain.

3. Since the momentum peak on 22 June, semis have been the biggest drag. Banks kept adding.

Europe's rally is narrower than the index level suggests. ASML carries the AI leg and HSBC the bank leg. Only the bank leg kept contributing after June.

Morgan Stanley's securitisation team shows the rate sell-off reaching the debt behind the AI build. Key observations:

1. Top-rated data-centre bonds now yield above 6%, against under 5% at the start of the year. A-rated ones are near 7%, up from about 5.5%.

2. High-yield bonds on sites leased to CoreWeave carry coupons up to 9 7/8%.

3. Developers that build and power sites for tenants, such as Applied Digital and Core Scientific, see some deals turn unprofitable at 9-10% debt costs. They have no other source of funding.

4. The bank cut its 2027 forecast for data-centre securitisations to $40-50bn from about $55bn.

Who gets hurt: developers that can only borrow, such as Applied Digital and Core Scientific. Hyperscaler backing is the escape route: TeraWulf's bond carries a Google backstop. Otherwise the squeeze eases only if yields fall.

A new Brookings paper by Columbia's Stijn Van Nieuwerburgh tests how the payoff changes with chip life. Key observations:

1. The paper assumes six years for IT gear, which is 68% of a campus's cost, and 20 years for everything else.

2. At three years, the required rental price rises from about $5.5 to $8.8 per GPU-hour at full use.

3. Morgan Stanley's hardware analysts already see enterprise servers replaced faster: every 3.3 years in 2025, every 2.8 years by 2027.

4. Listed data-centre landlords no longer work as a hedge. Their beta to the stock market has risen from about 0.5 to about 1.

My read: chip life is the hidden variable in every AI return model. Owners of long-lived assets, such as power plants, carry less of that risk than companies renting out GPUs.

JPMorgan's biotech team asks whether Chinese trial data hold up in Western patients. Key observations:

1. The drug is ivonescimab, which Summit licensed from China's Akeso. The trial excluded anyone over 75.

2. The typical US lung-cancer patient is about 70.

3. JPMorgan expects obesity data to travel well and cancer data to be judged trial by trial. Clarity should come over the next 6-12 months as global trials read out.

4. Some drugmakers have stopped disclosing how their Phase 1 drugs work, to slow Chinese copycats.

My read: Summit's global trial, with final data due in the first half of 2027, is the test for every China-sourced cancer deal.

Morgan Stanley tracks how European regulators ration grid access. Key observations:

1. Data centres drove about 85% of Ireland's electricity demand growth over a decade. Ireland lifted its 2021 ban around Dublin only for sites that bring their own generation.

2. Spain's draft asks for 80% of power matched with renewables hour by hour, with no six-year ramp-up. Its new capacity charge already made 16 facilities give back 1,040MW of grid capacity.

3. Britain's contracted demand rose from 41GW in November 2024 to 125GW in June 2025. At least 80GW of it is data centres.

4. Denmark froze new grid connections for three months while it reviewed about 60GW of requests.

5. European data-centre power demand rises from 78TWh in 2025 to 195TWh by 2030. Gas-turbine order books already run to 2029.

Who gets paid: owners of existing plants and connected land. Morgan Stanley's utilities team names RWE, EDP, Fortum and PPC as the next beneficiaries. The bank's base case is a softer final Spanish decree, which would ease the squeeze.

Morgan Stanley's internet and media analysts single out the private company as a key name in sports. Key observations:

1. Fanatics links the jersey, the trading card and the bet in one account. Its loyalty scheme grew from about 10mn members at launch to over 30mn.

2. Total revenue reaches about $13-14bn this year.

3. Global sports revenues have grown about 8% a year for three years, against about 5% for global GDP.

4. US team values have compounded at about 12% a year since 2000.

Sport is turning into one connected shopping and betting account. Fanatics is private, so there is no listed way in yet.

Morgan Stanley's European strategists size the flows. Key observations:

1. Net passive buying comes to about $0.9bn. National Bank of Greece gets about $327mn, Eurobank $258mn, Piraeus $257mn and Alpha Bank $170mn.

2. Jumbo, GEK Terna and Allwyn drop out of the developed index and face 10-20 days of passive selling.

3. FTSE and STOXX already moved Greece on 21 September.

4. Greek banks trade at a 10% P/E discount to European banks. They have only matched them over the past year, after almost four years of outperformance.

5. EU fund inflows equal about 10% of Greece's GDP, per Morgan Stanley Investment Management.

The setup: this is a bank trade into a dated event. It breaks if active emerging-market funds sell first. In Morgan Stanley's worst case they sell up to $22bn against $15bn of developed buying.

Jitania Kandhari at Morgan Stanley Investment Management explains the catch: the costliest plant needed each hour sets the price for all power. That plant is usually gas. Key observations:

1. Germany's model rested on 3 pillars: cheap Russian energy, Chinese demand for its machines and an American security umbrella. The model is now fractured.

2. Eurozone industrial output is 3% below its 2021 level. Chemicals capacity is down 9% since 2022.

3. More than 90% of electricity in Norway and Sweden comes from low-carbon sources. That is drawing hyperscaler investment at scale.

Cheap renewables do little for the industrial power bill while gas sets the price. Countries with their own baseload power, like nuclear France and the Nordics, gain an edge as data centres arrive. Industry moves east: Poland's output rises as Germany's contracts.

Two Morgan Stanley notes, a cross-asset study and Andrew Sheets' Sunday Start, weigh rates against stocks. Key observations:

1. In the Fed's own model, with no Fed offset, GDP ends 1.5-2pp lower after two years. Business investment falls 4.9pp and unemployment rises a point.

2. Hyperscaler capex, about $1.5trn in 2027 and $1.6trn in 2028, barely reacts to rates.

3. The US equity risk premium sits at the 3rd percentile of the past 25 years.

4. Sheets finds the gap between stock and bond yields explains only about 10% of the next year's relative return since 1998. Over three years, about half.

5. US 10-year yields rose 100 bps this year, yet the S&P 500 gained 13%.

6. Investment-grade issuance reached $1.6trn through August, up 30%. Hyperscalers alone sold more than $220bn of bonds.

The shock lands on households, small firms and stocks priced off a discount rate. AI spending, which carries index earnings, barely moves. Morgan Stanley sizes a yield spike without growth at a 5-10% index pullback.

Morgan Stanley's US economists argue that slowdown, rather than the size of the spending, is what cools the economy. Key observations:

1. After September's 25 bps hike, futures price almost 100 bps more of Fed hikes.

2. Morgan Stanley's own path only parts from the market after the first quarter of 2027.

3. None of the 3 catalysts that could cool hike bets looks close: better inflation prints, a Middle East deal or slower AI spending.

4. Unemployment below 4% would raise the odds of an October hike.

Relief on rates depends on capex growth slowing in 2027. Until then the risk runs toward more hikes.