Home

Subscriber portal home and product launcher.

Monday, October 26, 2026

Explore Atlas by Creative Strategies

Use Atlas and its tools to put Creative Strategies research to work.

Notes

View all

The Diligence StackOct 23, 2026Ben Bajarin
Prometheus and the Economics of Useful Compute
The strategic advantage of best in class compute
Ben BajarinOpen
Atlas Research NoteOct 23, 2026Ben Bajarin
Meta Q3 2026: Muse Works, and Now It Has to Get Cheaper to Serve
Creative Strategies earnings read | October 23, 2026 | Meta Platforms third-quarter 2026 results and conference call, October 21, 2026 Meta showed that its personal agent has found an audience, said the product is too expensive to serve at global scale today, and committed to cutting that cost by more than tenfold in a year, while the capex guide implies a December quarter near $54 billion. This report pairs with our Prometheus report, "Prometheus and the Economics of Useful Compute," which covers how Meta's training cluster produces the models Muse runs on. Operating view: Muse is the first consumer AI product we have seen where serving cost, more than model quality, sets how fast it can grow.
Ben BajarinPreview
Atlas Research NoteOct 23, 2026Ben Bajarin
Texas Instruments Q3 2026: Analog Turns Into a Capacity Story, and Our Models Are Behind It
Creative Strategies earnings read | October 22, 2026 | Texas Instruments third-quarter 2026 results and conference call, October 21, 2026 Texas Instruments told investors that little analog capacity has been added outside China in four years, that pricing has started to move, and that its cleanrooms are ready to fill, while its data center business nearly doubled from a year earlier. Operating view: The analog recovery is becoming a supply story, and TI is the supplier with the cleanroom space to meet it. Management said little analog capacity has been added outside China since the 2022 peak, that some foundries are moving capacity away from analog, and that a fab takes about four years from moving dirt to output.
Ben BajarinPreview
Atlas Research NoteOct 22, 2026Ben Bajarin
Lam Research Q3 2026: A Constrained $200 Billion, and NAND Gets Its Own Trade Ratio
Creative Strategies earnings read | October 22, 2026 | Lam Research September 2026 quarter results and conference call, October 21, 2026 Lam told investors that cleanroom space and component supply will set the pace of 2027 equipment spending while demand runs above it, and it gave NAND its first new reason in years to need more wafers for every bit it ships. Operating view: Lam's preliminary view of 2027 wafer fab equipment, more than $200 billion on a mid-$150 billion 2026, sits above our WFE Base case ($187 billion) and just under our High case ($208 billion), and management described it as a constrained number with real demand above it. That is the second equipment maker in two weeks to land on our High case for 2027, after ASML's roughly 85-system EUV plan.
Ben BajarinPreview
Atlas Research NoteOct 21, 2026Ben Bajarin
SpaceX Q3 2026: The Newest Hyperscaler Prices Its Watts at Spot
Creative Strategies earnings read | October 21, 2026 | SpaceX Q3 2026 results and earnings call, October 20, 2026 SpaceX is adding compute capacity faster than any operator we track and renting it for two to three times what our models assume a gigawatt earns, and its 2027 plan depends on chips, power and pricing all holding at the same time. Operating view: Superintelligence (SI) revenue of $7.6 billion was more than half of SpaceX's $14.6 billion quarter, and the Q1 2027 guide of $33 billion to $37 billion has SI becoming most of the company within two quarters. We think SpaceX is now best modeled as a capacity business, with revenue set by the gigawatts it energizes and the price each one earns.
Ben BajarinPreview
Atlas Research NoteOct 21, 2026Ben Bajarin
Micron's AI Memory Roadmap: Bandwidth Moves Onto the Compute Die, and Wafers Stay Tight
Creative Strategies Atlas note | October 21, 2026 | Micron's AI Memory and Storage Roadmap technology webinar with Scott DeBoer, President and Chief Technology and Products Officer, plus our follow-up conversation with Scott after the event Operating view: Each new AI memory product on Micron's roadmap takes more wafers per bit than the one before it. HBM5, vertical HBM (vHBM) and vertical edge inference memory (vEIM) all target volume in 2028, and Scott told us trade ratios rise at each step. Our Core Memory model already has HBM supply-limited from 2029, and it does not yet carry vHBM or vEIM, so our current theory is that the risk to our supply balance skews tighter.
Ben BajarinPreview

Need help?

Send the team a support ticket about account access, the research agent, billing, or the iOS app.

Contact support

Explore Atlas by Creative Strategies

Ask the Agent

Notes

Prometheus and the Economics of Useful Compute

Meta Q3 2026: Muse Works, and Now It Has to Get Cheaper to Serve

Texas Instruments Q3 2026: Analog Turns Into a Capacity Story, and Our Models Are Behind It

Lam Research Q3 2026: A Constrained $200 Billion, and NAND Gets Its Own Trade Ratio

SpaceX Q3 2026: The Newest Hyperscaler Prices Its Watts at Spot

Micron's AI Memory Roadmap: Bandwidth Moves Onto the Compute Die, and Wafers Stay Tight

Need help?