Bullpen.News From Edwin Dorsey
Get Daily Recap Feedback

X

SouthernValue

Everything we have picked from this writer, newest first.

20 stories Visit SouthernValue

@SouthernValue95

SouthernValue on X

$MSFT Cloud Chief Scott Guthrie: “Over time we expect Maia to be the dominant compute for our inference fleet,” “Multi-GW deployments begin in CY27,” “The future of AI won’t be defined by a single model, or a single chip, we will continue to apply the right model and technology to the right problem to bend down the cost curve,” “We won’t ever get custom silicon to 100%, but do expect it to get to the majority of our internal consumption,”
@SouthernValue95

SouthernValue on X

$AMZN paid <$400M for Annapurna … crazy how easy it is to spend money these days. Interesting acquisition, though. Decart has a world model and a video model, but the core piece AMZN is interested in is its “DOS” Decart Optimization Stack which helps AI models to train or inference over any chip technology, including already working with Trainium. This is a piece of technology AMZN wouldn’t want acquired by $NVDA. Makes a lot of sense within AMZN as a way to help customers use NVDA AMD CRBS TRN at optimal performance.
@SouthernValue95

SouthernValue on X

Best thing I’ve read in 3Q. One can be quite punitive on $AMZN assumptions and it still looks reasonably cheap on SOTP. I personally don’t expect results to look very “punitive” anytime soon. Bedrock is growing faster than the frontier labs and generates ~50% EBIT margins for AWS. Demand environment is supporting rising prices for each incremental GW deployed, and significant price increases on the install base (pure incremental profit requiring no additional capital).
@SouthernValue95

SouthernValue on X

$MSFT outlining their view of digital work and application stack of the future: - Copilot as the UI/OS for getting work done - Supported by the context layer of Microsoft IQ, bringing in data from all of your other tools - Connected to an “infinite software factory” to build personalized apps that help you get work done more efficiently, running on a managed infra layer that keeps data within the rails and avoiding app sprawl - and he is clear that existing Systems of Record need to evolve to support this structure or be replaced.
@SouthernValue95

SouthernValue on X

$SKYD should probably be a ~$4 stock. Unlike $PARA and $WBD it cannot breakup/sell - the only way out is through. Thats no easy road: starting almost 7x levered and needs to achieve MSD growth and to find $6B synergies from two companies that have already been thoroughly gutted to hit their deleveraging plan. Interest expense 150% of FCF. Rates exploding higher so they better hope they can de-lever. Potential seismic changes to entertainment in next 3-5yrs from generative content and short-form competition are existential threats. With any decline in revenue $SKYD will not be able to afford its sports rights renewals and the entire legacy business falls apart from there. Putting ~30% odds $NFLX gets another shot at WB studio in eventual bankruptcy. If you disagree and you are optimistic, please share why? I'm openminded but do not currently see it. Mgmt seems shell-shocked in Faber interview this AM getting a few tough questions. https://t.co/RKBucH3xlT
@SouthernValue95

SouthernValue on X

Danaher $DHR is combining technology from several of its bioprocessing and life science companies with Lab Robotics startup Automata to launch an AI-powered robotic autonomous research lab. From the Press Release: Danaher's first AI-powered autonomous lab is designed to deliver: - Speed: Up to 8x faster discovery of affinity reagents, helping researchers get from idea to breakthrough sooner - Scale: Unlocking up to a tenfold increase in annual reagent generation, from tens to hundreds of reagents per year as the lab scales - Quality: Each design built in the lab would also be verified, tested and fed back to improve the next, without trading speed for scientific rigor - Sustainability: Over time, this approach is expected to complement traditional immunization-based methods used in the discovery of antibodies and other research tools Based at Abcam, a Danaher company, the lab uses AI to help create new affinity reagents, including antibodies, then builds and tests each one, feeding every result back into the model to improve the next design. The lab brings together industry-leading technologies and expertise from several Danaher operating companies, including Beckman Coulter Life Sciences, Cytiva, Genedata, Integrated DNA Technologies, and Molecular Devices, in one connected workflow. The device orchestration and robotic automation is the result of collaborative work with Automata*. This continuous design-make-test-learn loop can run in a fraction of the time it takes today, without trading speed for scientific rigor. This approach applies the same test-and-verify discipline of continuous improvement long used across Danaher operations to an AI-enabled laboratory environment. The lab is set to be an early proof point for how Danaher's smart instruments, AI-ready data, and expert agentic systems can accelerate scientific discovery and medical progress. https://t.co/mw00CBjDC4
@SouthernValue95

SouthernValue on X

Morgan Stanley's Wigg: " $GOOG AI infra chief Amin Vahdat says Power lead times are directly gating capacity though GOOG would still rather remain grid-connected than fully self-generate as supplying 1 GW at 99.99%+ reliability on a standalone basis can effectively require ~2 GW of generation because of redundancy requirements. GOOGL is targeting a doubling of token-serving capability every six months. GOOGL’s 7-8 year-old TPUs are still running at 100% utilization despite the company using roughly a 6-year depreciation life,"
@SouthernValue95

SouthernValue on X

This is frontier lab only, $AMZN $MSFT would be the opposite and inference dominated. Possible that Anthropic is essentially going to be a nonprofit that will always be running at ~breakeven where training R&D = inference gross profit. In periods when they are accidentally profitable (now) that means they are behind on spend. And then partners like the clouds and inference marketplaces dedicate more compute to serve inference, which becomes capital light revenue to labs.