Latest: Hyperscaler Q3 Earnings Confirm AI Infrastructure Boom, Supply Crunch Deepens
The convergence of record AI-driven cloud revenues, massive debt issuance to fund continued buildout, and supply constraints spreading from GPUs to packaging, memory, storage, and lithography tools means the infrastructure cycle is not self-correcting quickly. The cascade is now affecting cost structures for consumer hardware companies like Apple, whose gross margins are compressing from memory price inflation driven by data-center competition.
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.
AWS, Google Cloud, Microsoft, and AMD all posted record AI-driven results in Q3 2026, and hyperscalers are now issuing tens of billions in bonds to sustain the buildout. AWS grew 39% to $46 billion with Bedrock usage up nearly 17x. Google Cloud more than doubled year-on-year to $30.7 billion with operating income tripling to $11.1 billion, though AI data center spending pushed free cash flow to a deficit exceeding $9 billion. Microsoft Cloud surpassed $63 billion with growth accelerating to 30%, Azure rising 48%, and commercial remaining performance obligations jumping 73% to $677 billion; quarterly capital expenditure reached $45.2 billion with the next quarter expected to exceed $50 billion. AMD's data-center segment hit a record $8.55 billion, up 97%, with CEO Lisa Su projecting data-center chips will account for 80% of next year's sales. Meta reported Q3 revenue of $64.6 billion, beating estimates, while EPS of $3.83 missed the consensus of $6.80 as infrastructure spending weighed on earnings; it guided 2026 capital expenditures at $135 to $145 billion.
Supply constraints now span nearly every AI infrastructure component. TSMC CoWoS capacity is fully allocated through at least mid-2027, H100 and H200 lead times are 36 to 52 weeks, hard drives are at 50 weeks, ABF substrates at 48 to 56 weeks, and DRAM at 20 weeks. Nvidia Vera Rubin racks require roughly 600 NOR flash chips each, leaving Winbond and Macronix sold out with new capacity not arriving until 2030. The constraint chain reaches chip-making equipment: ASML has formally told customers it cannot produce enough EUV tools to meet 2027 and 2028 demand, and Samsung and SK Hynix are each committing between roughly 10 trillion and 20 trillion won to secure orders.
Meta reported Q3 revenue of $64.6 billion, beating estimates of $63.4 billion, while EPS of $3.83 missed the consensus of $6.80 as AI infrastructure spending weighed on results.
Apple guided holiday-quarter revenue growth of 7% to 10%, ahead of analyst consensus of 6.1%, while holding a gross margin forecast of 45% to 46% weighed down by AI memory costs that shaved roughly one percentage point off its September-quarter margin.
Samsung and SK Hynix each committing 10 to 20 trillion won to ASML EUV orders; ASML formally told customers it cannot fill 2027 and 2028 demand
Microsoft declined to renew 2023 Oracle AI server contract; older-generation Nvidia chips no longer matched Microsoft's specifications
Nvidia approached Apollo about converting Brightspeed telephone exchange buildings into AI data centers spanning 20 states with roughly 1 gigawatt of power.
Apple CEO John Ternus called memory pricing 'unprecedented'; memory costs shaved approximately 1 percentage point off September-quarter gross margin; Apple guided holiday-quarter gross margin at 45 to 46%
AMD reported record data-center Q3 revenue of $8.55 billion, up 97%, and raised 2027 revenue growth outlook above 80%
KB Securities projected SK Hynix Q4 operating profit at roughly 104 trillion won, a 441% year-on-year increase at approximately 83% operating margin
Nvidia and Google co-designing custom HBM base dies with memory makers, moving memory controller off GPU into HBM base die in a shift Nvidia calls NVHBM
SK Hynix formed new Americas headquarters effective November 1, led by HBM development head Ahn Hyun, to work directly with US AI customers
AWS Q3 revenue grew 39% to $46 billion with Bedrock usage up nearly 17x; trailing free cash flow swung from a $14.8 billion inflow to a $42.4 billion outflow
Microsoft Cloud surpassed $63 billion with 30% growth; Azure rose 48%; Microsoft raised at least $25 billion in US investment-grade bonds
Google Cloud reported Q3 revenue up 103% year-on-year to $30.7 billion, with operating income tripling to $11.1 billion, as AI data center spending pushed free cash flow to a deficit exceeding $9 billion
Meta reported Q3 revenue of $64.6 billion beating estimates while EPS of $3.83 missed consensus of $6.80, and guided 2026 capex at $135 to $145 billion
Sundar Pichai states Google Cloud is foregoing short-term revenue to redirect compute toward frontier AI
TrendForce report shows hyperscaler HDD, DRAM, NAND, and ABF lead times far exceeding normal levels, with purchasing commitments extending to 2028
Ancora report: hyperscalers have issued nearly $200B in debt in 2026; AI-related debt through 2030 projected at $4 trillion
H100 and H200 lead times reach 36-52 weeks; TSMC CoWoS packaging capacity reported fully allocated through at least mid-2027
BackgroundCompute: the chips, clouds and data centers behind AI
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.