Eco-responsible AIs. Today.
With half the world now online, the growth of the digital ecosystem is transforming the way we live -- and our environment as well.
Today, ICT's carbon footprint accounts for about 4% of total greenhouse gas emissions (GHG). That's already a lot.
Within ten years and the universalization of AI-based applications, trend studies estimate this footprint in excess of 15% -- equivalent to half the emissions of the global transport sector -- if nothing is done to significantly reduce it, with climate neutrality as the ultimate objective.
east That is why, as a technologies factory, LEXISTEMS is dedicated to designing eco‑responsible AIs based on six company‑wide commitments:
Time to information is a function of data being fully findable, accessible, interoperable and re-usable. This requires applications capable of understanding both users and the data they address. IDC estimates that only 0.5% of data is analyzed (3% being merely tagged), while TRUE Global Intelligence locates "dark" – i.e unquantified and untapped – data north of 50% in average worldwide. Thime to information is where LEXISTEMS' meaning-based technologies shine, especially when compared to keywords and other obsolete alternatives. Think 1 API call vs 5 to 10. Meaning is just as green as it is powerful.
As HPC veterans, we know pretty well how efficient programming requires less hardware, compute energy and data exchanges at the edge, core and cloud levels. More than just "green coding", our technologies systematically implement HPC best practices -- the only ones capable of making exaflops supercomputers a sustainable reality. Compound that with optimized training cycles of mostly transfer learning-based models... and you've got the cleanest possible act.
We take pride in developing AIs that are efficient and smart enough to be embedded into low-consumption devices or mobile apps (15-60W switchable vs 800W in average + permanent cooling for an API server). When models sizes or inference requirements don't cut it, we consistenly design the leanest possible APIs and payloads to reduce service and network energy needs to the minimum.
Storage is no longer just the persistence of files. Today's public and business applications have become multi-sources and multi-repositories, with increasingly distributed and hierarchized information. LEXISTEMS' AIs can read and digest data and documents from within their datastores, regardless of updates frequency. No duplication, no back-and-forth: using the right combination of locality, virtuality and vectorization, our technologies require 0 additional storage energy.
Claiming eco-responsibility is one thing, proving it another. Internally, our implementation decisions result from methodologically assessing applicative and energy efficiency. And to best help our customers achieve their climate neutrality objectives, every Sensible Solution comes with its own carbon footprint report, which includes overall training and typical inference metrics.
Whether in compute, inference, storage or delivery, theres' a lot of data services providers to choose from. We selectively partner with -- and recommend to our customers -- those with the most ambitious environmental commitments.