BEP ResearchFree to read
Free Isn’t Cheap Enough
Read best config against best config, SemiAnalysis's own dashboard gives NVIDIA 9.7x outperformance on the TPU it just crowned.
Independent AI infrastructure research
From silicon
to cash flow.
Research for investment and strategy teams.
BEP connects engineering, market data, and company economics to help you assess the businesses being built around AI.
Work with BEP Research
Bring the research into your team’s decisions.
Discuss team research access, private briefings, or a focused project around your investment or strategic questions.
Latest research
Ideas worth spending time with.
BEP ResearchFree to read
The Only Company That Can Afford Free
The reported Hugging Face deal makes NVIDIA the Compute Patron of open source AI.
BEP ResearchFree to read
The Write Problem: Which AI Bytes Actually Belong on NAND
AI is pulling NAND into memory tiers it has never served. Third in our free memory series with Rick Xie.
An observation from The Stack
AI changes fast.
Measure the change.
LLMflation follows a weighted basket of model output prices against GPT-4’s March 2023 launch price. It’s one way our research moves from a narrative to a number.
Explore the market dashboard The full dashboard is included with a paid subscription.LLMflation
As of 2026-09-11
21.4GPT-4 launch = 100
How to read this comparison
A price-basket comparison, not a measure of equivalent model quality or task success. Weights: OpenAI 30%, Anthropic 25%, Google 20%, DeepSeek 15%, open-weight models 10%. The launch reference is $60 per million output tokens.
Put the research to work
Read the thesis. Check the numbers.
01
The Stack
GPU and token prices, infrastructure costs, and the margins behind AI.
Subscriber accessOpen The Stack ↗
02
Compute Scarcity Index
Track how tight compute markets are, which way they’re moving, and what drives the change.
LevelMomentumComponent drivers
Subscriber accessExplore BCSI ↗
03
Open research tools
Interactive supply chains, AI task economics, and the research call sheet.
Open accessBrowse free tools ↗
People behind the work
Engineering depth.
Investment perspective.
Our work brings together the BEP team and specialist research collaborators. The memory series is one example: engineering and standards analysis paired with the economics of the business.
Founder · BEP Research
Ben Pouladian
Electrical engineer, operator, and investor. Ben leads BEP Research and the investment framing of our collaborative memory series.
About Ben ↗Research collaborator · memory systems
Rui “Rick” Xie
A memory-systems researcher and co-author of our memory series. Rick leads its engineering and standards analysis, from DRAM to SSDs.
Read the collaboration ↗Investment activities · BEP Holdings
BEP Syndicate
Private investment across the
AI infrastructure stack.
Through BEP Holdings, we organize special-purpose vehicles to invest in private companies across AI infrastructure. Our investment activities draw on the sectors and technologies we study through BEP Research.
- Focus
- Private AI infrastructure companies
- Structure
- Deal-specific SPVs
- Participation
- Verified accredited investors
Register interest in future opportunities. Joining the list creates no investment commitment. Participation is subject to verification and the applicable offering documents.
Research subscriptions and SPV participation are separate engagements.
BEP’s investment activities may create financial interests in companies or sectors covered by its research. Consider those interests when evaluating the analysis.
BEP Research membership
A clearer view of
the AI economy.
In-depth research, market dashboards, and working models in one subscription.
View subscription options ↗Already a subscriber? Sign in ↗
Use your subscription email to receive a one-time sign-in code.