The AI sector is an ever-expanding space. Every week, new tools, devices or large language models (LLMs) are introduced to society, revealing a new edge to everyday life where artificial intelligence can be useful. However, the creation of these software megastructures requires a sturdy hardware foundation, which is not known to be cheap.
Hence Nvidia, famous for its chips and graphics cards, is today one of the most valuable companies in the world. The company, which offers high-end hardware such as its most recent H100 chips, has become the main supplier (and therefore bottleneck) of this technology, which is sold to tech giants such as Apple, OpenAI and Amazon so that they can process the data used to train their most powerful and famous LLMs.
All this has meant that over the last few years Nvidia's value has not stopped growing at a dizzying, even insane, rate. The company's shares have risen by 149% so far this year, making it the second best-performing stock in the S&P 500 (SPY) at present. At the same time, in the second quarter of this fiscal year, the company posted record revenues of $30 billion, higher than the expected $28.86 billion. This represents an increase of 122% compared to the same period in 2023, as well as an increase of 15% compared to the first quarter.
Faced with figures like these, many of us wonder what the company's future will be, where it will go with such financial solvency. After seeing the movements of this technological galleon, from Insights we can see where Nvidia is turning the rudder, and is that not only wants to dominate the hardware, but is heading towards a sea previously unknown to them, the software.
The NVLM language model 1.0
This September, Nvidia launched its open source AI model that promises to rival advanced systems from the likes of OpenAI and Google, NVLM 1.0. It is capable of generating detailed descriptions of images, understanding the humour of memes and performing complex coding tasks. The most advanced variant, the NVLM-D-72B, handles 72 billion parameters and competes directly with large-scale models such as GPT-4 and Llama 3.
The development of NVLM 1.0 stands out for its ability to simultaneously process text and images, performing complex tasks such as optical character recognition (OCR), reasoning, and step-by-step instruction generation for mathematical problems. The company has also stressed the importance of the quality and diversity of the data used to train this model, beyond the simple scale of the data.
In our view, the most notable feature of this announcement is that NVLM 1.0 is available under open source standards, allowing it to be used, modified and distributed freely. This follows a dynamic that companies like Meta are leading the way (at least in terms of their LLM technology advancements) and benefits smaller, independent developers.
The video game software
Nvidia is not just following in the footsteps of the big tech companies by developing its own LLMs, but is also looking to advance the software of its other big market, video games. Nvidia's chips are not its most famous source of income, but it competes with the other hardware it dominates, that of thegraphics cards. These computer components are vital for playing the most powerful video games, and some of the latest models even have AI enhancements to improve graphics performance.
But the business of software is not just there, but in the implementation of AI tools in the development of video games.
For example, Nvidia created Machinima, a platform that gives users tools to create animations using advanced graphics in real time, mainly powered by Nvidia's graphics cards. It is aimed at game-based content generation, using the power of artificial intelligence and real-time simulation. This facilitates the production of cinematics without the need for sophisticated technical equipment or the intervention of professional studios.
Also relevant is the recent announcement of its new tools for game development on Unreal Engine 5 (graphics engine), presented during the Unreal Fest Seattle 2024. These add-ons are designed to facilitate the creation of realistic MetaHuman characters and optimise the integration of AI in game development. Key new features include:
Audio2Face-3D: This plug-in lip-syncs and animates the facial expressions of MetaHuman characters using AI, providing realism in animations within Unreal Engine 5.
Nemotron-Mini 4B: It allows the generation of interactive responses in dialogues with characters, improving the fluidity and naturalness of interactions within games.
RAG (Augmented Generation Retrieval): Enhances character interaction by providing contextual information, enriching in-game responses and actions.
In addition, NVIDIA has integrated its technology with Autodesk Maya through an audio-based facial animation plug-in, which facilitates a more streamlined workflow between Maya and Unreal Engine 5. The company also announced a rendering microservice that enables high-quality streaming of MetaHuman characters across any device via WebRTC.
Finally, another service we would highlight is GeForce Now, a sort of "netflix of video games" that allows you to play titles purchased from platforms such as Steam or the Epic Games Store from the cloud. This eliminates the need for powerful hardware, as games are run on NVIDIA's servers and streamed to your device. The RTX 3080 subscription enables 1440p and 120 FPS gaming with Ray Tracing enabled, delivering a high-quality gaming experience without the need for a powerful PC.
