Associate Professor Banking and Finance,Università Cattolica del Sacro Cuore, SWF,ETF,Islamic Finance, CNBC, @Natgasweather, Energy Working, Foresight Economics
Verona,Italy
Joined December 2014
- Tweets45.8K
- Following1.9K
- Followers7.9K
- Likes30.6K
Pinned Tweet
Safe havens or war hedges? #oil #gold #natgas during Iranian war. Below the link of the publication
pdf.sciencedirectassets.com/…
Andrea Paltry, Ph.D. retweeted
Narrative violation: @GoldmanSachs raises data center spending forecast for 2026/27 and 2028.
@GoldmanSachs sees $125bn data center spending in 2028. Texas adds timing risk.
@GoldmanSachs raises its US data center spending forecasts for 2026/27 to $80bn/$100bn and adds $125bn for 2028. Its forecast has double-digit growth through 2028, with spending remaining above 2026 levels through 2030.
The capacity forecast rises from 65GW in 2026 to 143GW in 2030. @GoldmanSachs puts the high-confidence pipeline at $900bn, with 86GW under construction and 171GW in high-probability projects. The spending outlook depends on when that pipeline gets built.
Texas is the immediate timing risk. A proposed construction moratorium would affect 20%/27% of 2027/28 under-construction additions. @GoldmanSachs sees delay and cost risks, with $IREN most exposed and $CLS / $ANET facing knock-on risks from slower AI infrastructure deployment. Its view is that post-midterm restrictions would push activity into 2029/30 while leaving the total addressable market intact. Medium/low-probability projects face significant schedule uncertainty.
Andrea Paltry, Ph.D. retweeted
GS: Down 44.5% from its June peak, KOSPI is nearing GFC and Dot-com bubble drawdown levels (~52-53%). What makes this drop unique: it's driven by valuation compression despite rising earnings.
Compared with the Dot-com bubble in particular, earnings quality is stronger
Andrea Paltry, Ph.D. retweeted
The Contrarian Data Center Trade: Goldman Raises Its Construction Forecast Into The Blue Wave, Sees Spending Through 2030 zerohedge.com/technology/con…
Andrea Paltry, Ph.D. retweeted
According to Korean media, global DRAM revenue hit nearly $200 BILLION in the third quarter.
That is up 369% from a year ago, according to Counterpoint Research.
Bit shipments only grew about 4% while the average selling price hit $1.9 per gigabit. This is a PRICE cycle, not a volume cycle.
Samsung held the number one spot at 38% and SK Hynix jumped to 29% after DRAM revenue rose more than 40% quarter on quarter while Micron is at 21%. China's CXMT slipped back to 9%.
DDR revenue was up 500% and LPDDR was up about 400%. Conventional memory is growing FASTER than HBM right now, because AI racks are stuffing more DRAM next to every CPU.
Counterpoint expects Samsung to take back the HBM crown next year on HBM4.
Stocks are crashing but the fundamentals just printed a record.
$MU $NVDA $AVGO
Qualcomm is down more than 11% today, after opening 14% lower in what was on track to be its WORST day since 2020.
The whole chip sector is selling off with it, with the PHLX Semiconductor Index down more than 5% this morning.
The company actually beat on earnings, but its handset business, which is its biggest segment, saw revenue drop 26% to $5.14 billion.
The bigger problem is costs, since Qualcomm said prices are rising across wafers, packaging, testing and especially memory.
AI data centers are soaking up so much memory supply that phone chip makers are getting squeezed on both price and availability.
Qualcomm says it is raising prices to offset this, but the full benefit is not expected until the second half of fiscal 2027.
The AI boom is creating huge winners, but it is also creating real VICTIMS across the rest of the chip industry.
$QCOM $MU $AAPL $ARM
Andrea Paltry, Ph.D. retweeted
Meritz Securities (South Korea):
Genuine demand reflected in orders received by memory chipmakers for 2027 and 2028 far exceeds their supply capacity. Even if the Democrats’ blue wave leads to some delays and adjustments, the supply shortage is unlikely to change materially.
This year, the memory order fulfillment rate—available supply capacity relative to order volume—has remained in the low-to-mid 80% range, driving sharp increases in selling prices. The shortage is expected to worsen in 2027, with the fulfillment rate falling to around 80%.
Meanwhile, during a recent visit to South Korea, OpenAI CEO Sam Altman asked SK hynix and Samsung Electronics to supply specialty memory, including HBM, for 2027–2030 in volumes beyond what they can accommodate.
Andrea Paltry, Ph.D. retweeted
J.P. Morgan says third-quarter memory results came in with a clear positive bias. Makers beat on earnings and margins despite currency headwinds.
DRAM average selling prices rose 23% quarter-on-quarter, ahead of the high-teens to 20% consensus range. Bit guidance for the fourth quarter points to double-digit sequential growth. High-bandwidth memory 4 shipments have begun. The profitability crossover between HBM4 and conventional DRAM is expected in the third or fourth quarter of 2026. The firm sees high-bandwidth memory average selling prices rising about 96% in 2027. Company-specific issues, including Samsung’s delayed shareholder returns and Kioxia’s soft fourth-quarter guidance, look temporary rather than structural.
Demand remains mixed. Hyperscaler AI capital expenditure stays robust, with upward revisions for next year. PC and mobile face greater pressure. Bottom-up supply-demand fundamentals look solid through 2028. Multiple makers indicate 2027 capacity is already largely sold out. The setup supports continued constructive views on memory stocks near term, even as the market weighs longer-term AI funding and power constraints. Risk-reward remains attractive across the coverage universe.
$DRAM
If high-bandwidth memory 4 shipments have already begun and multiple makers say 2027 capacity is largely sold out, does the next constraint stay on how fast those HBM4 volumes can ramp, or on whether hyperscaler AI spending keeps rising once the current pricing surge cools?
JPM Memory Market Update:
The order visibility is extending as we progress through the year, but we believe the market is anticipating AI funding risk and power bottleneck driven memory downcycle as early as in 1H27E.
This puts memory in a tricky spot as investors debate between “too-cheap to ignore” vs. “it is cheap for a reason.” We believe EPS cuts are unlikely to happen in the next 6M as the memory S-D gap is too wide to narrow over a short period of time and memory earnings will have a buffer protected by LTAs and prepayments.
On a medium-term basis, we continue to view risk-reward as attractive and remain OW across all our coverage companies and rank our next three months order of preference based on shareholder return policy upside: SKH (OW) > SEC (OW) > Kioxia (OW).
Andrea Paltry, Ph.D. retweeted
US JOLTS Job Openings Sep: 7.523M (est 7.164M; prev 7.079M; prev R 7.19M)
- Job Openings Rate: 4,5% (prev 4.3%)
- Quits Level: 3.365M (est 3.096M; prev 3.066M; prev R 3044M)
- Quits Rate: 2.1% (prev 1.9%)
- Layoffs Level: 1.641M (est 1.679M; prev 1.641M; prev R 1768M)
- Layoffs Rate: 1.0% (prev 1.0%; prev R 1.1%)
Andrea Paltry, Ph.D. retweeted
Chairman of Macronix, Wu Min-hiu, says memory may be short in supply until 2030.
Wu Minqiu said in an interview that there may be a shortage of memory chips until 2030. Considering the current scarcity of production capacity, he will make the best use of the available capacity. Currently, equipment delivery times are not only long, but prices are also rising. He specifically visited equipment suppliers in hopes of accelerating delivery.
Wu Minqiu stated that Tesla and SpaceX CEO Elon Musk has money, but having money is only the first step. Musk is not a leader in the IC field, and Americans value lifestyle, so he does not believe the Terafab project will succeed.
$SPCX $MU $DRAM $EWY $SKHY
Andrea Paltry, Ph.D. retweeted
🚨 NOMURA: THE MARKET MAY BE MISREADING THE ENTIRE AI DATA CENTER BOOM.
Big Tech is spending hundreds of billions on AI infrastructure. Free cash flow is under pressure. Investors are worried the AI boom isn't profitable.
But Nomura argues something very different:
AI infrastructure is already highly profitable. The cash flows simply haven't caught up with the investment cycle.
Here's the breakdown 🧵👇
1/ The profitability numbers tell a different story.
According to Nomura's November 2 report:
• AWS revenue: $46B, +39% YoY
• AWS operating margin: 38.7%
• Google Cloud revenue: $30.7B, +103% YoY
• Google Cloud operating margin: 36.2%
• Azure growth: +48%
These are not the numbers of an industry struggling to generate operating profits.
2/ The real issue is timing, not necessarily profitability.
Big Tech pays for data centers, GPUs, networking and memory upfront.
Revenue and cash generation arrive over subsequent years.
That creates a classic free-cash-flow J-curve:
CAPEX accelerates → FCF deteriorates → Infrastructure generates revenue → Investment growth moderates → FCF recovers.
3/ Nomura expects operating income to surge before free cash flow.
Its illustrative forecasts suggest operating-income growth could peak around 2029, while FCF growth accelerates later.
That distinction matters enormously for investors valuing AI infrastructure businesses.
4/ Why can't hyperscalers simply stop spending?
Because AI increasingly threatens their existing businesses.
Search, advertising, shopping, software and cloud services are being reshaped by AI agents.
Companies that underinvest today risk surrendering customers, data and market share tomorrow.
5/ Demand may not be the primary constraint. Capacity is.
Nomura cites enormous contracted cloud backlogs:
• Google Cloud: $560.8B
• AWS: ~$496B
• Microsoft commercial RPO: $677B
These figures have different reporting definitions, but collectively illustrate substantial contracted demand.
6/ Big Tech has financing alternatives to cutting CAPEX.
Nomura identifies four paths:
① Accelerate investment
② Freeze CAPEX
③ Use joint ventures and operating leases
④ Monetize contracted future cash flows
Its preferred expectation is continued investment supported by increasingly sophisticated financing structures.
7/ Oracle offers a real-world example.
Nomura highlights Oracle's combination of long-term leases, project financing, customer prepayments and customer-provided hardware.
The broader message: hyperscalers can restructure who finances infrastructure without necessarily reducing how much infrastructure gets built.
8/ This distinction is critical for semiconductors and memory.
Negative hyperscaler FCF does not automatically imply collapsing demand for GPUs, DRAM, HBM or other data-center components.
If spending continues through financing structures, semiconductor demand can remain resilient even while hyperscalers optimize their balance sheets.
9/ What could invalidate the thesis?
A genuine compute oversupply.
Watch:
• Cloud revenue growth
• GPU rental pricing
• Contracted backlog growth
• Data-center utilization
• Memory supply commitments
• Hyperscaler investment guidance
These are more useful indicators than negative FCF alone.
10/ The investment takeaway:
The key question is not simply:
"Why is Big Tech burning so much cash?"
It is:
"Will the infrastructure being built today generate enough durable cash flow to justify its cost of capital?"
Nomura is optimistic.
But the answer ultimately depends on pricing, utilization, investment discipline and the durability of AI demand.
The AI infrastructure cycle may be a cash-flow timing problem rather than a profitability crisis.
$NVDA $MU $GOOGL $MSFT $AMZN $META $ORCL
Andrea Paltry, Ph.D. retweeted
UBS projects global DRAM capacity to rise about 147% to 89 exabytes by 2030. That is up from 36 exabytes in 2025.
Samsung is projected at 31 exabytes. Micron is at 26 exabytes, up 100% from the 2027 estimate. SK Hynix is at 22 exabytes. CXMT is at 10 exabytes. The growth rates from 2027 to 2030 are 63% for Samsung, 100% for Micron, 38% for SK Hynix, and 67% for CXMT. Even this expansion covers less than one-third of UBS’s full agentic AI scenario, which would require about 282 exabytes of DRAM by 2030. AI CPUs already need about three times the memory of traditional processors. Industry estimates sit far lower, around 77 exabytes.
Supply growth looks strong on paper. It still falls well short of unconstrained AI demand. Memory remains the bottleneck.
$MU $SKHY #cxmt #samsung $DRAM
A capacity forecast is not a finished wafer.
If even a 147% rise to 89 exabytes still covers less than one-third of the full agentic AI scenario, does the next constraint stay on how fast those new fabs can ramp, or on whether demand from AI CPUs and data centers keeps outrunning the added capacity once it arrives?
Andrea Paltry, Ph.D. retweeted
According to Korean media, Korea Investment & Securities just set a price target of 3,200,000 won on SK Hynix.
That is more than DOUBLE where the stock is trading this morning at about 1,389,000 won.
The firm also set a target of 440,000 won on Samsung, which is about 100% above today's price of 219,000 won.
The earnings forecasts behind those targets are enormous, with 619 trillion won of operating profit for SK Hynix in 2027 and 400 trillion won for Samsung.
The analysts point to HBM4 leadership and rising memory prices for SK Hynix, and rising DRAM and NAND prices for Samsung.
So the brokers keep raising their numbers while the stocks keep getting sold, and that gap is the whole story in Korean memory right now.
According to Korean media, KB Securities now expects SK Hynix to earn ₩104 trillion in operating profit in the fourth quarter alone.
That would be the first quarter EVER above ₩100 trillion, up 441% from a year ago, at an operating margin of roughly 83%.
KB points to higher DRAM and NAND shipments, rising average selling prices and a bigger mix of HBM4 as the drivers.
The firm kept its Buy rating and its ₩3.5 million price target.
For context, that single quarter of profit would be larger than the annual operating profit of almost every company in the world.
$SKHY $MU $NVDA
Andrea Paltry, Ph.D. retweeted
The MAG7 free cash flow is estimated to explode into 2030 per Bloomberg:
Andrea Paltry, Ph.D. retweeted
According to Korean media, South Korea just exported $56.4 BILLION of chips in October, up 259% from a year ago.
That is the second best month in the country's history, behind only September's record of $60.3 billion.
Memory alone was about $50.1 billion of that, so this is almost entirely a DRAM, HBM and server SSD story.
Chips are now roughly half of everything South Korea ships to the world, and total exports grew 93.7%, the fastest pace since 1974.
Exports to the United States nearly tripled to $24.9 billion, which tells you exactly where the AI data center buildout is happening.
The stocks sold off last week, but the order book did not.
$SSNLF $SKHY $MU $NVDA
According to Korean media, brokers are raising their 2027 profit forecasts for Samsung and SK Hynix.
Meritz Securities now expects 654 trillion won of operating profit for Samsung next year and it expects 481 trillion won for SK Hynix.
Together that is about 1,135 trillion won, or roughly $790 billion, in a single year from two memory companies.
Hana Securities raised its Samsung 2027 forecast by 20% and expects SK Hynix to earn more than 100 trillion won in Q4 alone, a FIRST for the company.
SK Hynix says its DRAM and HBM output is effectively sold out through 2029.
Wall Street is still treating memory like a cyclical business, but Korean analysts keep raising their numbers.
$SSNLF $SKHY $MU
Andrea Paltry, Ph.D. retweeted
Morgan Stanley: memory as a % of global cloud capex to rise from 48% in 2026 to 53% in 2027.
$NVDA $AMD $MU $DRAM $SNDK $SKHY
Andrea Paltry, Ph.D. retweeted
UBS on CoreWeave:
"Apart from rising rates (which absolutely do impact future funding costs), we don’t see any fresh 'crack' in the story to warrant the sell-off, in fact the demand, GPU pricing and 'up- the-stack' moves have been encouraging."
$CRWV
Coreweave $105 PT $CRWV
“The recent share price weakness and widening of CDS/credit spreads have investors in search of an explanation, a source of incremental fundamental deterioration. Apart from rising rates (which absolutely do impact future funding costs), we don’t see any fresh 'crack' in the story to warrant the sell-off, in fact the demand, GPU pricing and 'up- the-stack' moves have been encouraging. Pinning the sell-off on CoreWeave-specific factors is akin to arguing that a share price decline must be a function of deteriorating fundamentals (sometimes it is, often it isn’t). In our view, this is a reflection of a broader confidence erosion in AI spending and stocks.”
