{"id":2530,"date":"2026-08-17T08:00:00","date_gmt":"2026-08-16T23:00:00","guid":{"rendered":"https:\/\/digirise.ai\/?post_type=news&p=2530"},"modified":"2026-08-18T11:50:07","modified_gmt":"2026-08-18T02:50:07","slug":"20260817","status":"publish","type":"news","link":"https:\/\/digirise.ai\/en\/news\/20260817\/","title":{"rendered":"Digirise, with a track record of supporting AI implementation for a cumulative total of 600 companies, has launched "Toride AI," a local LLM construction outsourcing service."},"content":{"rendered":"

We can set up a frontier-class open-weight LLM such as Kimi K3 within your company's private network environment. It can handle confidential files, and there are no usage-based charges.<\/p>\n\n\n\n

Digirise Co., Ltd. (Headquarters: Minato-ku, Tokyo \/ Representative Director and CEO: Masahiro Chaen, hereinafter referred to as "the Company") has launched a service to securely build OpenWaite LLMs (Large-Scale Language Models) in a company's own environment.Toride AI<\/strong>We have started offering "[product name]".The AI's core will be installed on your company's server.<\/strong>This will enable the use of AI generation even in tasks involving confidential files. We provide end-to-end support, from model selection and hardware procurement to construction, verification that no external transmission occurs, and employee training.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Background: Japanese companies' use of generative AI is stalled at the point where they "want to use it but can't."<\/strong>
While the use of generative AI in business operations is becoming commonplace, its adoption by Japanese companies remains limited. A survey conducted by Teikoku Databank targeting 4,705 companies revealed that17.31 TP3T are companies that "utilize" generative AI.<\/strong>Only 26.81 TP3T indicated they are "considering using it."<\/p>\n\n\n\n

The main reason for stopping is,Concerns about information leakage<\/strong>According to a survey conducted by BlackBerry in 2023,Organization 72% has banned or is considering banning the use of generative AI apps in the workplace.<\/strong>That is their response.<\/p>\n\n\n\n

This restriction is even more pronounced in regulated industries. These include areas such as finance (FISC security standards, Financial Services Agency AI discussion paper), healthcare (3 ministries, 2 guidelines), local government (LGWAN), and economic security (Act on the Protection and Utilization of Important Economic Security Information), where operations are designed with the premise that data cannot be shared outside the company. According to a survey by the Ministry of Internal Affairs and Communications,The number of municipalities that have introduced generation AI is 29.91 TP3T.<\/strong>(As of December 2024) The number of companies that have implemented it remains low, with approximately 70% still not having adopted it.<\/p>\n\n\n\n

The government itself has also issued warnings about the risks of storing data on overseas servers (Personal Information Protection Commission, February 3, 2025 \/ Digital Agency, February 6, 2025). What's important here is,The warning concerns "using APIs to send data to servers of overseas operators."<\/strong>That's the point. Even with the same model, self-hosting, where you download and run the software in your own environment, has a structurally different data flow.<\/p>\n\n\n\n

Our company provides AI reskilling and AI utilization consulting services to corporations.Over 600 companies in total<\/strong>We will support the introduction of AI, and as of the end of March 2026Number 1 in the number of companies using our AI reskilling service for businesses.<\/strong>(According to Tokyo Shoko Research \/ as of the end of March 2026) The following is what we have repeatedly heard on the ground:<\/p>\n\n\n\n

\n

"We have banned ChatGPT company-wide because we are concerned about confidential information."<\/p>\n\n\n\n

- "Industry rules prevent the use of external AI services."<\/p>\n\n\n\n

"The cost of AI will increase the more you use it."<\/p>\n\n\n\n

"We can't lift the ban because we can't manage the 'scope of use' for each employee."<\/p>\n<\/div><\/div>\n\n\n\n

"Toride AI" was designed as an option to solve all four of these problems together.<\/p>\n\n\n\n

\n
\"\"
The "5-stage AI maturation model" is based on our company's definition.<\/figcaption><\/figure>\n<\/figure>\n\n\n\n

Turning point: The era has arrived where top-tier AI can be downloaded and run in-house.<\/strong><\/p>\n\n\n\n

Until now, "AI that can run on in-house servers" has generally lagged behind cutting-edge cloud models in terms of performance. This assumption is beginning to break down as we enter 2026.<\/p>\n\n\n\n

July 2026,The "Kimi K3" (2.8T parameter), an open-weight LLM with frontier-class performance, has been unveiled.<\/strong>This has been done (Hugging Face official model card). The official blog for the inference engine vLLM has published deployment procedures and measured throughput values assuming real-world operation, making it clear how companies can run it on their own GPU servers.<\/p>\n\n\n\n

Actual hiring has already begun overseas. According to Forbes (July 27, 2026),Approximately 60% of the tokens consumed by US companies via OpenRouter are from the Chinese company OpenWeight Model.<\/strong>And,DoorDash, Coinbase, and others have adopted Kimi.<\/strong>It has been reported that... Also, in its FY2026 earnings report, NVIDIA said that sales related to "sovereign AI," where countries have their own AI infrastructure...It expanded to about three times the previous year.<\/strong>They have made it public that they did so.<\/p>\n\n\n\n

The performance gap is also narrowing. In the "Intelligence Index," a comprehensive metric from the AI model evaluation organization Artificial Analysis, the open-weight Kimi K3 scored 57 points, a difference of only 3 points from the top-performing closed model's 60 points (as of July 2026). Epoch AI's analysis also indicates that the average time it takes for the top-performing open-weight models to lag behind the most advanced closed models is approximately 4 months (as of May 2026).<\/p>\n\n\n\n

The assumption that "the best AI is only available on external clouds" is already becoming a thing of the past.<\/p>\n\n\n\n

<\/p>\n\n\n\n

\n
\"\"<\/figure>\n<\/figure>\n\n\n\n

Performance difference between open-weighted LLMs and closed-loop models. Source: Artificial Analysis (publicly available data as of July 2026) \/ Epoch AI. Created by our company based on publicly available data from various companies.<\/p>\n\n\n\n

"Toride AI" is a service that creates a dedicated AI for your company, in-house.<\/h2>\n\n\n\n

"Toride AI" provides an AI chat environment with a user-friendly interface similar to ChatGPT.Inside your company's server<\/strong>This is the service we will be building.<\/p>\n\n\n\n

FEATURE 01 | You can upload confidential files as they are.<\/h3>\n\n\n\n

It can handle contracts, blueprints, customer lists, and more. The imported data never leaves your company's server. The open-weight model is essentially a file of numerical parameters and does not have communication capabilities itself, allowing for operation in a closed network configuration.<\/p>\n\n\n\n

FEATURE 02 | You can manage what you "can do" and "cannot do" by department and position.<\/h3>\n\n\n\n

You can configure available features, whether file attachments are allowed, and whether external sharing is permitted on an organizational basis. Instead of a simple "company-wide ban or company-wide ban" choice, you can gradually lift restrictions based on the level of risk.<\/p>\n\n\n\n

FEATURE 03 | No usage-based billing is required.<\/h3>\n\n\n\n

Because the system does not involve user-based or token-based billing, operating costs are predictable even when deployed company-wide. The only ongoing costs after implementation are electricity and maintenance fees.<\/p>\n\n\n\n

FEATURE 04 | We measure and deliver "zero external transmissions" in practice.<\/h3>\n\n\n\n

During the setup process, we verified through communication logs that no external communication occurred.Verification Report<\/strong>We will provide it to you in this form. Instead of saying "it probably didn't appear," we will ensure that it is recorded.<\/p>\n\n\n\n

FEATURE 05 | Models are selected by our company based on their intended use, budget, and licensing requirements.<\/h3>\n\n\n\n

In addition to Kimi K3, we support gpt-oss (OpenAI), Gemma (Google), Qwen, DeepSeek, and Japanese models (ELYZA, Sarashina, etc.). We will propose a solution after carefully reviewing publicly available benchmarks for Japanese language performance, our own verification, and licensing conditions, including whether commercial use is permitted. If you are hesitant about using overseas models, we can also configure a system using US-made or Japanese models.Regardless of which model you choose, the mechanism that ensures "data does not leave the company" remains the same.<\/strong><\/p>\n\n\n\n

\n
\n
\n
\"\"
This is a scene from a site visit to a construction company client. As an FDE (Forward Deployed Engineer), I enter the site to identify issues that are not visible on screen.<\/figcaption><\/figure>\n<\/figure>\n<\/div>\n<\/div>\n\n\n\n

Five features of "Toride AI"<\/p>\n\n\n\n

Plan Structure \u2500\u2500 Three plans tailored to scale and purpose<\/h2>\n\n\n\n

We offer three plans to suit your usage needs.<\/p>\n\n\n\n

<\/th>Light<\/th>Standard<\/th>premium<\/th><\/tr>
Image of scale<\/td>For use within the department (1 GPU)<\/td>For company-wide use (2-8 GPUs)<\/td>Exclusive access to top performance (8 GPUs or more)<\/td><\/tr>
Estimated construction period<\/td>Approximately 1 month<\/td>1-2 months<\/td>2-3 months<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

*Hardware will be purchased by your company (at actual cost), and we will provide configuration design and procurement support.<\/p>\n\n\n\n

*For companies that want to try it out on a small scale first,Closed-network proof-of-concept (trial verification) using cloud GPUs<\/strong>We also have that available.<\/p>\n\n\n\n

*This project may be eligible for subsidies such as the Digitalization\/AI Introduction Subsidy 2026 and the Small and Medium-Sized Enterprise Labor-Saving Investment Subsidy (General Type). Please check eligibility on a case-by-case basis.<\/p>\n\n\n\n

The cost will vary depending on the number of users, intended use, and security requirements, and will be determined through an individual quote. The price range for each plan is published on the pricing page of our service website.<\/p>\n\n\n\n

Pricing plans:https:\/\/toride.digirise.ai\/#pricing<\/a><\/p>\n\n\n\n

On the service website,A cost simulator that shows the break-even point for cloud AI when you enter the number of users.<\/strong>We have also made it public.<\/p>\n\n\n\n

Cost simulator:https:\/\/toride.digirise.ai\/#simulator<\/a><\/p>\n\n\n\n

Implementation Process \u2500\u2500 Your "in-house AI" can be operational in as little as two months.<\/h2>\n\n\n\n

\u3000STEP 1: Free consultation and requirements definition (Free, 60 minutes)<\/strong>We will discuss the purpose, number of people, and budget, and then provide you with an estimate.<\/p>\n\n\n\n

\u3000STEP 2: Trial Verification\/PoC (2-4 weeks)<\/strong>We can verify the effectiveness using your company's data before making an investment decision.<\/p>\n\n\n\n

\u3000STEP 3: Equipment preparation (2-6 weeks)<\/strong>We design the server configuration and installation location. Procurement is optimized through competitive bidding.<\/p>\n\n\n\n

\u3000STEP 4: Construction and Safety Check (2-4 weeks)<\/strong>We will build the chat screen and access control system, verify that there are zero external transmissions, and deliver the product to you.<\/p>\n\n\n\n

\u3000STEP 5: Training and Start of Use (Continued Support)<\/strong>We provide training for employees, supporting them until they can "master" the technology.<\/p>\n\n\n\n

If securing a server location or power supply is difficult, we can propose installation strategies including colocation to a high-power data center or a private network plan for a domestic GPU cloud.<\/p>\n\n\n\n

<\/p>\n\n\n\n

\n
\"\"<\/figure>\n<\/figure>\n\n\n\n

Implementation process (Start using the service in as little as 2 months, from the free consultation in STEP 1)<\/p>\n\n\n\n

The reason for offering an "honest comparison" with cloud AI<\/h2>\n\n\n\n

Our company also handles the corporate implementation of cloud AI as part of our "Claude Code Corporate Implementation Support" service. Therefore, in this service,I do not unilaterally recommend local LLMs.<\/strong><\/p>\n\n\n\n

\u30fbThe top performer in overall performance still lies with cloud AI.<\/strong>The advantage of always having access to the latest model at no extra cost is something that locals don't offer.<\/p>\n\n\n\n

\u30fbFor low usage volumes, cloud APIs can be more cost-effective.<\/strong>We will openly present this break-even point.<\/p>\n\n\n\n

On the other hand, local LLMs have:Data does not physically leave the company \/ Verified model versions can be permanently fixed \/ The more you use it, the lower the unit price becomes.<\/strong>This offers a value that is difficult to replace with cloud services.<\/p>\n\n\n\n

For many companies, the practical solution ishybrid<\/strong>That's right. We use local LLM for highly confidential processing and cloud AI for general-purpose processing that requires the latest performance. From our position of being able to handle both, we design the optimal allocation in a neutral manner.<\/p>\n\n\n\n

Representative's comment<\/h2>\n\n\n\n

I've seen countless companies that have banned AI. They do so because they have a high level of security awareness. I think that decision itself is correct. The problem is when they keep it banned for years.<\/p>\n\n\n\n

The release of Kimi K3 as openware in July 2026 shattered the assumption that "the highest-performing AI is only available on external clouds." This means we've entered an era where cutting-edge AI can be deployed on a company's own servers. However, running it requires specialized design and infrastructure, which remains a barrier for many companies.<\/p>\n\n\n\n

Toride AI is a service that overcomes those barriers for you. And we're not just a company that sells local LLMs. Because we also provide support for cloud AI implementation, we're in a position to say, "In your case, the cloud is sufficient." We don't just sell what we want to sell; let us work with you to determine the optimal solution for your company.<\/p>\n\n\n\n

-- Masahiro Chaen, Representative Director and CEO, Digirise Co., Ltd.<\/p>\n\n\n\n

Supervisor Profile<\/h2>\n\n\n\n

\u8336\u5713 \u5c06\u88d5\uff08\u3061\u3083\u3048\u3093\u30fb\u307e\u3055\u3072\u308d\uff09\uff5c\u682a\u5f0f\u4f1a\u793e\u30c7\u30b8\u30e9\u30a4\u30ba \u4ee3\u8868\u53d6\u7de0\u5f79<\/strong><\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n
\n

- Official Gemini Advisor at Google Japan \/ Advisor to GMO AI&Web3. Serves as an AI advisor to several listed companies.<\/p>\n\n\n\n

\u30fbX (formerly Twitter) @masahirochaen has over 190,000 followers. He shares cutting-edge generative AI through his YouTube channel "Chaen's AI Lab" and other channels.<\/p>\n\n\n\n

He speaks at over 150 conferences, corporate training sessions, and local government support events annually. He also appears regularly on NewsPicks' program.<\/p>\n\n\n\n

\u2022 Author of the book "AI Brain: A Thinking Revolution That Creates Unlimited Free Time" (KADOKAWA)<\/p>\n<\/div><\/div>\n\n\n\n

Free white paper "The Textbook for Local LLM Implementation 2026" released.<\/h2>\n\n\n\n

On the service website, the white paper "Local LLM Implementation Textbook 2026<\/strong>We are offering this document (44 pages in total) for free. It summarizes the structural differences between local LLMs and cloud AI, global and Japanese implementation examples, and criteria for deciding whether it is necessary for your company.<\/p>\n\n\n\n

Download materials:https:\/\/toride.digirise.ai\/#contact<\/a><\/p>\n\n\n\n

Future outlook<\/h2>\n\n\n\n

Digirise has been supporting AI-driven management for Japanese companies by building organizational foundations through AI training, supporting business transformation through consulting, and going as far as advanced AI implementation through development. "Toride AI" is located there.Data sovereignty and regulatory compliance<\/strong>This service adds that as a key element.<\/p>\n\n\n\n

We aim to achieve our mission of "making Japan the world's No. 1 in AI utilization rate and productivity" by creating an environment where companies that have stopped using AI because they "want to use it but can't" can safely take that first step.<\/p>\n\n\n\n

Service Overview<\/h2>\n\n\n\n
\n
\n

\u2022 Service name: Toride AI (Local LLM construction outsourcing)
\u30fbStart of provision: August 17, 2026
Services provided: Model selection \/ Hardware configuration design and procurement support \/ Private network environment construction \/ RAG and internal system integration \/ Verification and report delivery of zero external transmissions \/ Operation and maintenance \/ Training for employees on how to use the service
\u30fbExamples of compatible models: Kimi K3, Kimi K2.6, gpt-oss, Gemma, Qwen, GLM, DeepSeek, and Japanese models (ELYZA\/Sarashina, etc.)
\u2022 Plans: Three plans available: Light, Standard, and Premium, plus a private PoC (Proof of Concept). Pricing varies depending on requirements and will be determined through individual quotes (price ranges are listed on the service website).
\u2022 Development period: Minimum of approximately 1 month (minimum of 2 months with standard progress including PoC)<\/p>\n\n\n\n

\u2022 Service website:https:\/\/toride.digirise.ai\/<\/a><\/p>\n<\/div><\/div>\n<\/div><\/div>\n\n\n\n

\u25a0Company Overview<\/strong>
Company name: Digirise Inc.https:\/\/digirise.ai<\/a>)
Address: 10F Tokyo Port City Takeshiba, 1-7-1 Kaigan, Minato-ku, Tokyo
Representative: Representative Director Masahiro Chaen
Business details:
\u25cf Number 1 in the number of companies using our AI risk-killing service for corporations* "Corporate Risk-Skilling\u00ae"
https:\/\/digirise.ai\/business\/reskilling\/<\/a>
\u25cf AI system development and AI utilization consulting
\u25cfClaude Code Corporate Implementation Support
\u25cf Operates "DigiRise AI School," an AI school for individuals.
\u25cf Development and operation of AI products for businesses<\/p>\n\n\n\n

[Inquiries regarding this press release]<\/strong>
Digirise Co., Ltd. Public Relations Department
Email: pr@digirise.ai<\/p>\n\n\n\n

Notes and Sources<\/h2>\n\n\n\n

- "No. 1 in the number of companies adopting our AI reskilling service for corporations." Source: Tokyo Shoko Research Co., Ltd. "Survey on Reskilling Services for Corporations" (2026 \/ as of the end of March 2026)
- Generation AI utilization rate: 17.31 TP3T, Under consideration: 26.81 TP3T: Teikoku Databank "Corporate Survey on the Utilization of Generation AI" (4,705 valid responses)
- Prohibition\/Consideration of prohibition of the use of generative AI apps in the workplace: 72%: BlackBerry survey (2023, targeting IT decision-makers)
- Introduction of AI for generating municipal data: 29.9%: Ministry of Internal Affairs and Communications data (as of December 2024)
- Government warnings: Personal Information Protection Commission (February 3, 2025) \/ Digital Agency administrative notice (February 6, 2025)
\u2022 Kimi K3: Hugging Face Official Model Card \/ vLLM Official Deployment Blog
- US companies adopt open weight models (approximately 60% - DoorDash\/Coinbase): Forbes (July 27, 2026)
- Increased sovereign AI-related sales: NVIDIA FY2026 earnings report
\u2022 Intelligence Index (Kimi K3 = 57 \/ Closed Top = 60): Artificial Analysis (Publicly available values as of July 2026). Figures were created by our company based on publicly available values from each company.
- Difference between open weight and closed weight: "approximately 4 months on average": Epoch AI "Epoch Capabilities Index" (as of May 2026)
\u2022 Subsidies: Digitalization and AI Introduction Subsidy 2026 \/ Small and Medium-Sized Enterprise Labor-Saving Investment Subsidy (General Type)
The stated timeframes are estimates only and may vary depending on the requirements definition. Results are not guaranteed.
\u2022 Model names and product names are trademarks or registered trademarks of their respective companies.<\/p>\n\n\n\n

<\/p>","protected":false},"featured_media":0,"template":"","meta":{"_acf_changed":false},"news_cat":[37,16],"class_list":["post-2530","news","type-news","status-publish","hentry","news_cat-press","news_cat-product-service-information"],"acf":[],"_links":{"self":[{"href":"https:\/\/digirise.ai\/en\/wp-json\/wp\/v2\/news\/2530","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/digirise.ai\/en\/wp-json\/wp\/v2\/news"}],"about":[{"href":"https:\/\/digirise.ai\/en\/wp-json\/wp\/v2\/types\/news"}],"wp:attachment":[{"href":"https:\/\/digirise.ai\/en\/wp-json\/wp\/v2\/media?parent=2530"}],"wp:term":[{"taxonomy":"news_cat","embeddable":true,"href":"https:\/\/digirise.ai\/en\/wp-json\/wp\/v2\/news_cat?post=2530"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}