CoreWeave, The GPU Champion That Isn't NVIDIA
Sundays, The Sequence Scope brings a summary of the most important research papers, technology releases and VC funding deals in the artificial intelligence space.
Next Week in The Sequence:
Edge 315: Our popular series about foundation models continues with an overview about tree-of-thought(ToT) reasoning in LLMs. The original ToT paper from Princeton University and a deep dive into the language model query language(LMQL).
Edge 316: We deep dive into Salesforce’s CodeT5+, a new open source code generation foundation model.
Go Subscribe!
📝 Editorial: CoreWeave, The GPU Champion That Isn't NVIDIA
If you've never heard about CoreWeave and you're in the AI world, believe me, you soon will. The GPU cloud platform has become synonymous with high-scale computing in the AI industry. When we talk about AI hardware, NVIDIA immediately comes to mind. How could it not? The chip giant has experienced a surge in revenue and earnings thanks to the rapid rise of AI technologies. However, NVIDIA is not the only company benefiting from the GPU craze; CoreWeave has emerged from obscurity, raising more than $400 million in equity and almost $2.3 billion in debt, making it one of the most important compute offerings of this AI generation.
CoreWeave's origins are not in the AI space; the company began in 2017 as a miner for the Ethereum blockchain. However, when AI started gaining traction, CoreWeave swiftly pivoted to building a GPU-accelerated cloud optimized for NVIDIA hardware. The CoreWeave platform includes native Kubernetes orchestration, support for the top GPU accelerators, as well as storage and networking capabilities. While CoreWeave remains highly reliant on NVIDIA GPUs today, its offerings are expanding rapidly.
The success of CoreWeave has materialized in approximately $500 million in revenue forecasted for this year and back-to-back $200M+ equity rounds in April-May. Just last week, CoreWeave announced securing a remarkable $2.3 billion in debt financing, led by Magnetar Capital and Blackstone. The proceeds of this financing will be used to purchase more NVIDIA GPUs, build 12 new data centers, and, obviously, attract talent. CoreWeave already has many of the marquee AI platforms in the market on board.
CoreWeave is becoming a common component of large-scale infrastructure for foundation models. They've expertly ridden the GPU wave.
🔎 ML Research
LLM Attacks
Researchers from Carnegie Mellon University published a paper that proposes an attack method for LLMs. Specifically, the technique uses a suffix across a large number of queries that causes the LLMs to produce more affirmative responses —> Read more.
LoraHub
Researhcers from the University of Washing and Allen Institute for AI published a paper detailing LoraHub, a framework that takes advantage of the Low-Rank Adaptation (LoRA) technique for tasks other than basic fine-tuning. LoraHub combines different LoRA modules to adapt models to unseen tasks —> Read more.
🤖 Cool AI Tech Releases
AudioCraft
Meta AI released AudioCraft, a family of generative AI models for text-to-audio generation —> Read more.
ONNX Script
Microsoft open sourced ONNX Script, a library that enables writing ONNX ML pipelines directly in Python —> Read More.
llama2.c
Adrej Karpathy released llama2.c that generates a lighweight, faster , 500 lines of C code based on the binaries of a trained version of Llama 2 —> Read more.
🛠 Real World ML
Embeddings at Uber
The Uber engineering team discusses the use of two-tower embedding models for improving recommendations —> Read more.
📡AI Radar
CoreWeave secured an astonishing $2.3 billion in debt financing to expand its GPU offering.
SoftBank launched a new Company to develop LLMs for the Japanese market.
AI processor vendor Tenstorrent raised $100 million from strategic investors.
Application performance monitoring platform Datadog unveiled Bits AI, a DevOps copilot.
Twilion announced new generative AI capabilities powered by OpenAI.
Serverless database platform Neon announced a $46 million round to add AI-native capabilities.
AI gaming startup Inworld raised $50 million in new funding.
AI platform for blockchain’s smart contract automation SettleMint raised a $17.5 million series A.
Google is adding new generative AI capabilities to its Assistant platform.
Dell announced that its expanding its Project Helix with a collaboration with NVIDIA for powering next-gen generative AI solutions.
Uber announced that is working on a new chatbot to be integrated into its app.
IBM and NASA open sourced one of the largest geospatial AI models ever built in Hugging Face.
AI video platform Wisecut secured fresh investment from legendary VC Tim Draper.
AI security platform Tromzo secured $8 million in new financing.