Urban Housing Inflation, China, Nvidia

Urban Housing and Inflation

This week, I wanted to address the housing component of the inflation calculation. Shelter represents roughly 35% of CPI inflation metrics and 20% of the core PCE inflation. Shelter is the slowest metric to change as landlords can often afford to be patient, and 12-month leases take time to roll over. Owner occupancy (people who own and live in their homes) is lowest in major cities. The owner occupancy rate in Boston is around 35%, meaning shelter inflation directly impacts 2/3rds of residents. Shelter inflation flows through to service workers, meaning higher prices in restaurants, construction, offices, etc. 

When there is a supply-demand mismatch, too little housing for too many people, rents increase. However, it does not need to be this way. WSJ recently profiled the many vacant lots in Chicago. Boston has an abundance of empty lots as well. We spoke to our friends at Groma, a leading multi-family REIT in Boston run by Seth Priebatsch and Chris Lehman, and learned about the significant tensions involved in adding to the local housing supply. Developers will not build on vacant lots in Boston until policies are updated because it’s not financially feasible to build under the current system of approvals, permits, local meetings and regulations. This keeps rents and rent inflation unnecessarily high while valuable supply sits unused (and untaxed! – see Land Value Tax)

The lower-hanging fruit today is to focus on preserving and updating existing housing. Many of the 15,000 triple-deckers in Boston need repairs and updating. We could also use more regulatory flexibility to develop creative housing solutions like ADUs (accessory dwelling units), co-living and rent subscription models. There’s also a push to build six-story buildings instead of the traditional triple-deckers. Six-story walkups are not currently culturally accepted, but most of the population could physically handle the stairs, especially now with Ozempic (kidding, sort of). The NBER study that Chris shared with us shows that people are often not rewarded for moving to pursue higher-paying jobs, which negatively impacts overall growth.

Politicians need to pay more attention to the on-the-ground solutions to fix inflation. Policy changes or tax incentives would lead to the rapid development of the many currently vacant urban lots. Inflation is not purely a financial problem or solution. The $1 trillion + in interest expenses (taxpayer dollars) we wrote about last week is unsustainable. A better policy strategy would be effective and is certainly needed.

China’s Economy

Former Secretary of the Treasury Henry Paulson wrote an op-ed in the Washington Post with a nuanced review of the current state of China and its role in the global economy. I recommend reading it here.

NVidia

Nvidia’s latest specialized AI product, the H100, costs roughly $30-$40k per unit. Tesla recently purchased 10,000 units to build its supercomputer, Dojo, which trains and powers the AI/machine learning models for Tesla’s Full Self-Driving (FSD) capabilities. NVidia’s top AI chips are understandably in extremely high demand. TSM in Taiwan produces all of the leading AI chips using ASML machines. 

While NVidia is in the lead, consider that AWS, Google and AMD also aim to build top-of-the-line AI chips. It takes time to design these chips, and all of the latest technology, including the H100, was planned before ChatGPT’s public debut last November. The next generation of chips designed for specific AI or machine learning use cases like LLMs or FSD will likely be even better. Consider how NVidia GPUs were the leading crypto-mining chips until ASICs (application-specific integrated circuits) arrived as an optimal solution for crypto-mining. Flexibility will likely be a valuable attribute as models and optimizations change at this early phase. Google started developing machine learning ASICs, in this case, called TPUs (tensor processing units) as early as 2016. The H100 apparently has a profit margin of over 90%, so there is a massive financial margin up for grabs for anyone who can build AI chips with performance comparable to NVidia’s models, which are currently in a class of their own.

*Additional Disclosure: Osbon Capital Management holds investments in Groma Corp and Groma REIT. Mentions of Groma or related companies are not an endorsement or advertisement.*