Timothy Arcuri is a Managing Director at UBS, leading their equity research for semiconductors and IT equipment
He was one of the few analysts who called the massive bull run in Micron early on and still believes the memory supply chain is undervalued today.
Here is his bull thesis on Micron and why Wall Street still mistakenly trades the stock at a 3x forward P/E
"Memory has always been cyclical, because it has always been about supply more than demand. People trade it like nothing has changed, but in fact, a lot has changed.
You now have HBM, which is on annual cadence because every new generation of ASICs or XPUs requires a new generation of HBM. So now there is this constant connectivity you need to have with the memory producer."
Introducing Praxis-1, a new 27B MoE world action model from @runwayml, and Praxis-1 Fast (73ms median) - an optimized checkpoint from our research team.
We’re proud to ship Day 0 support with DROID and YAM endpoints live today on the Reactor API.
. @AnthropicAI generated $1.72 of revenue per $1 of compute in Q3... up from $0.40 in Q1'25 but down from $2.08 in Q2
Expensive short-term capacity hurt the ratio. @OpenAI has done better with early contracts in '25
Procurement & capacity planning matter alongside model quality
Coding got enterprise budgets first because it's the one process with a built-in verifier i.e tests pass or not. ROI is easy to measure... so the spend keeps growing.
Most business processes don't have a cheap way to check the work yet.. so spend grows, then stalls when they can't show ROI.
Applied AI will most likely scale one process at a time as each gets its own verifier.
Today, we're expanding our eighth-generation TPU portfolio with two new systems built specifically for the agentic era — bringing our total to four distinct eighth-generation TPU systems launched in a single year.
TPU 8i Max was built for massive context windows and the highest compute density per rack, and TPU 8tx is optimized for secure and confidential inference. While flexible across workloads, the four systems in our 8th gen TPU portfolio give developers and organizations a purpose-built accelerator for every stage of the agentic lifecycle.
@AminVahdat, SVP and Chief Technologist, AI and Infrastructure, explains why we're expanding our TPU portfolio and what everyone needs to know about TPU 8tx and 8i Max ⬇️
Interesting to see Americans more pessimistic about AI than many other countries.
Is it because how frontier labs communicate ie focus on job losses and doomsday scenarios?
Other countries heard those too… they land differently here.
@nrmehta Agreed on moving beyond tools to business outcomes. But competition may pass some AI savings to customers, so flat margins don’t necessarily mean no value created.
Getting a robot policy into production takes more than model weights. Low-latency inference, streaming and deployment all need to work together.
@reactorworld is putting that infrastructure behind an API so robotics teams can focus on making their robots more capable.
Well done team @_bschmidtchen@taiuti 🦾
Introducing Remote Policy API on Reactor.
You can now:
• Run policies on Reactor’s infrastructure with ultra-low latency
• Use Cosmos Nano, FLUX Action, and DreamZero out of the box
• Power your robot control loop with <30 ms global rtt.
• Deploy, scale and stream with your own weights with Reactor Serve.
Get started on DROID checkpoints today with more coming each week.
Not sure data center jobs impact gets enough attention.
According to Goldman DC-exposed construction employment 315k above broader construction trend since 2022 i.e. electricians, HVAC crews, utility workers, etc.
AI productivity gains don’t automatically accrue to shareholders
Goldman finds almost no relationship between S&P 500 companies’ AI exposure & margin growth 👇
Great guidance from @jpmorgan team for enterprises adopting AI agents @patopet https://t.co/8mvImm4fMw
Hardware security is important part of the AI resilience.
@eclypsium secures that foundation by verifying hardware/firmware integrity from procurement through operation @c7zero
BofA on GPU spot rental pricing:
Continued strong GPU rental pricing data suggests demand for compute is still high, with B200 rental price of ~$6.47/hr near all-time high level.
Estimated market pricing for 1M tokens continue to decline, now at ~$0.92 vs. TTM median of $1.45, suggesting tokens are becoming more accessible and affordable over time (which could lead to greater usage and token generation
$NVDA $SPCX $NBIS $IREN $CRWV
Hardware and firmware security across IT/OT deserve far more attention.
The consequences reach well beyond stolen data.
I appreciate the effort @c7zero@eclypsium team have spent working on this problem.
Love Janan’s takes on FT
This one applies well to most VCs on X:
“You very likely have nothing additive to say about politics. Most opinions are correct but banal, or distinctive but wrong - while quite a few are both banal and wrong.
To air them is no great service to the public realm. To hold back, to abstain, often is”