KT, which suffered a setback in the government's domestic AI foundation model project, is seeking a turnaround through its 'AI for Everyone' business. While the foundation model project was a stage to compete on proprietary AI model technology, AI for Everyone requires the delivery of nationwide services and stable operations, where KT is seen as having room to show its strengths.

KT has assembled an alliance spanning domestic AI model companies and infrastructure firms to platform operators that have nationwide services. It is seen as a strategy to combine experience in telecommunications, cloud and enterprise and consumer services to make up in service competition for setbacks suffered in model competition.

Elite team elimination in foundation model project stings...retrying in a different way after 1 year

KT formed a consortium with domestic AI service, model, platform and infrastructure companies and entered the government's 'AI for Everyone' project. Like the foundation model project, AI for Everyone is a large AI initiative led by the government. In particular, as a nationwide service, it is a suitable stage to show corporate recognition and technical capabilities.

KT took part in last year's foundation model project with a consortium but was eliminated at the first stage of selecting elite teams. At the time, 15 consortia competed, and five consortia including Naver Cloud, Upstage, SK Telecom, NC AI and LG AI Research were selected as elite teams. The Ministry of Science and ICT planned to select 4 of the 5 elite teams in the first evaluation, but passed only 3 due to issues such as originality. KT did not take part in the repechage for the remaining slot.

The foundation model project puts weight on securing the technological competitiveness of foundation models independently developed by domestic companies. In the initial elite-team evaluation, experience in AI model development and technological capabilities were key assessment factors. Competition then continued in a way that verified usability through benchmarks and expert and user evaluations.

KT failed to clear the bar for the elite team despite having its in-house model, 'Mideum'. Some in the industry pointed to a possible lack of competitiveness in its own model. It also contrasted with SK Telecom being included in an elite team and LG Uplus being selected as an elite team through the LG AI Research consortium.

From Upstage to Rebellion and Motif...attention on a national 'alliance'

Unlike the foundation model project, AI for Everyone focuses on building a general-purpose AI chatbot and public AI agents that anyone can use. It must apply domestic AI models with originality and performance on par with the foundation model project to more than 50 percent of total usage, while also using models from other domestic companies for at least 30 percent. That makes the ability to properly connect multiple models and run them reliably more important than building a single in-house model well.

A feature of the KT consortium is that it brought in a wide range of specialised companies across models, infrastructure and services. Upstage, Motif Technologies, NC AI and ActionPower take part in AI model and platform areas. Among them, Upstage is an elite team that has survived so far in the foundation model competition. Motif Technologies and NC AI also experienced stage evaluations in the foundation model project.

Motif Technologies was eliminated in the second stage evaluation of the foundation model project, but its model competitiveness received high marks. It scored 11.9 points in a benchmark evaluation by global AI evaluator AAII (Artificial Analysis Intelligence Index), worth 25 points, the highest among teams that advanced to the second stage evaluation. Upstage had no first-place category but received broadly good evaluations and advanced to the third evaluation.

For KT, it can generate synergy by using domestic models that entered the foundation model competition, including Motif Technologies and Upstage. The same applies to infrastructure. Rableup, Rebellion and Vessel AI Korea are taking part. In the foundation model project, KT had to prove the performance of its in-house model 'Mideum', but in AI for Everyone the new battleground is how well it can bundle and serve strong domestic AI technologies.

Telecom B2C experience seen as strength...Daum, Musinsa and Zigbang placed at the forefront of services

The services area is seen as the part where KT is most confident in this project. The KT consortium includes Daum, Saltlux, Musinsa, Zigbang, ConnectWave, Mathpresso, EBS, BC Card and DeepSearch. They are operators that already run services directly connected to people's daily lives, spanning search to shopping, housing, education and finance.

KT has also run large-scale business-to-consumer operations centred on mobile telecommunications. A telecom industry official said, "For large-scale services like AI for Everyone, experience dealing with a large number of users is important, including handling simultaneous connections, responding to failures and security," and added, "It is also an area KT can do best."

That is also what KT is emphasising in this project. KT plans to combine its telecom, cloud and AI operations capabilities with the expertise of each participating company to establish a foundation for AI services that the public can use reliably.

Competition, however, will not be easy. SKT, which also joined the AI for Everyone tender, has both a proven in-house model and large-scale B2C service experience. Kakao joined hands with LG AI Research, which is participating as an elite team in the foundation model project. LG Uplus is also participating in the Kakao consortium. It is a structure that combines model competitiveness from LG AI Research and telecom service operations experience from LG Uplus with large user touchpoints including KakaoTalk.

SKT in particular has A.X K2, which passed through the second stage evaluation of the foundation model project. In the foundation model evaluation, it received good marks in demonstrations of deployment in large-scale commercial services and application of industry-specific agents for defence, manufacturing, legal affairs and tax. The Kakao consortium is also a rival equipped with its in-house model 'Kanana', models from LG AI Research and large-scale service operations capabilities.

An industry official said, "It is hard to say KT is ahead of competitors if you look only at its in-house model, but experience operating services for a large subscriber base is a clear strength." The official added, "AI for Everyone is a project where KT can show its strengths more than it did in the foundation model project. If it delivers results, it could also be a good reference for winning future public-sector or enterprise AI projects."