Headlines: New Pricing strategies in SaaS businesses, AI Opensource movement, AI enterprise deployments with Cohere, AI Inference startup survival game and finally Tesla's big plans from Q2 earning call

AI in Businesses

• The per-seat model that dominated the software licensing world is likely up for a disruption as AI chatbots taking the seats occupied by humans in several tasks - as independent actors or assistants. This article analyses the well known SaaS products and their pricing strategies to remain profitable. (Tomasz Tunguz - Theory Ventures)

• Meta has released a new model, trained with 15 Trillion tokens and using the compute power of a 15,000 * H100 GPU cluster. The resulting model is 405 billion parameter behemoth and beats GPT-4 mini-o in several benchmarks. Yet it is outsourced to the AI community. Mark Zuckerberg makes arguments for Meta for the wider AI community for taking this approach (Mark Zuckerberg - Meta Blog)

• Cohere is mission is deploying LLMs for large-scale enterprises; and not building LLMs themselves. Cohere brings capabilities in fine-tuning, retrieval augmented generation (RAG), parameter-based tuning & reinforcement learning from human feedback (RLHF) techniques to mention a few. As a testimony to the success of its strategy; Cohere recently raised $500 million in funding raising its valuation to $5.5 Billion

• AI inference startups (8 of them with ~1 Billion in funding raised) are making models of choice available for the customers saving them the pain of building and maintaining complex models. These startups abstract the challenges with building and maintaining the infrastructure (e.g. orchestration of GPU fleets, OS/CUDA/configuration management, monitoring/observability) and then optimising for things like maximizing utilization. However the TAM (Total Addressable Market) for these start ups are in the range of $1 Billion currently and the pricing is already aggressive. Article discusses potential exit paths and likely outcomes for these startups

Macro developments

• Ray Dalio measures the chances of a civil war soon after the US presidential election. Ray posits that given the far right tendencies from Republican party under Trump's leadership, there is a higher chance that the Republican party will negate the outcome of the election if the verdict is not favourable to Trump. Therefore, the chances of such a conflict are reduced with the improving odds of the Republican victory. Last month's dramatic events has done just that with three dramatic events - 1) the Trump / Biden presidential debate, 2) Trump's assassination attempt 3) Biden's withdrawal from the presidential race. (Ray Dalio - LinkedIn)

Story of the week

• Tesla had a punishing Q2 quarterly result - its adjusted operating margins are at 14.4% down from 18.7% a year ago, the lowest in 3 years (CNBC) and subsequently, the stock is down by 8%. But that decline obfuscates the strategic initiatives Tesla is making in a range of spheres. Tesla Q2 earnings call itself provides an insightful Q&A with Elon Musk & his leadership team across EVs, Storage Batteries, Humanoid Robots and Robotaxi businesses (Tesla YouTube channel).

  • FSD (Full Scale Driving) which Tesla building is far more generalised than anything else anyone else is building

  • FSD will arrive in Q4 and then Elon says, "I will be shocked if it is not by 2025"

  • Robotaxi will work on the same principles as Airbnb: implying Tesla can operate a fleet on a shared usage basis with Tesla car owners rather than investing in a fleet on its own

  • Primus 2 Robots are going into production in 2026. Tesla will have the scale to beat the competition by significant margins

  • DOJO supercomputer is a strategic necessity for Tesla given the premium pricing for NVIDIA GPUs and will double down on building it

  • Tesla acknowledges that if there is a withdrawal of subsidies for EV vehicles that will have an adverse impact, but that will be felt least likely by Tesla compared to the competition

  • Tesla plans to invest in xAI (AI company from Elon) and will go for Tesla board approval for investment of $5 Billion to enable xAI to work on a generalised AI than Tesla's narrow AI

  • Tesla's battery storage business is growing significantly from the data centre's needs to have an alternate energy supply source anticipating power shortages and subsequent fluctuations in the grid capacity.

  • US utilities have 3 times the capacity than the actual energy generated by them. With forecasted shortage in energy supply, utilities are investing in generating energy during off hours and storing in storage batteries

  • Tesla will harness distributed computing utilising the under utilised computing capacity available via Tesla EVs and Primus Robots and thereby reducing the compute costs currently constraining growth