GDEP ADVANCE,Inc. Q2 FY2026 earnings call

• Overall Financial Results
  • Cumulative first half revenue: 3.083 billion yen, down 22.8% year-over-year. The revenue decline was partially expected due to large projects booked in the first half of the prior year, though project delays in automotive and manufacturing impacted results more than expected.
  • Gross profit: 885 million yen, up 23.8% year-over-year; gross margin improved 10.8 percentage points to 28.7% from 17.9%.
  • Operating profit: 613 million yen, up 23.1% year-over-year. Q2 standalone operating profit increased 38.7% year-over-year.
  • Ordinary profit: 649 million yen, up 30.6% year-over-year, supported by 31 million yen in foreign exchange gains.
  • Net profit: 448 million yen, up 30.3% year-over-year.
• Customer Base Performance
  • Repeat order revenue share remained strong at 85.5% for the first half, with repeat existing customers as the core target of top-line recovery efforts.
  • Nearly 50% of total customers have placed orders for 4 consecutive periods as of Q2, demonstrating a strong stable customer base.
• Operational Strategy
  • Early recovery measures were implemented from the start of Q1, shifting focused sales resources to the company's core strength of small-to-medium sized projects ranging from several million to several tens of millions of yen. This shift improved sales mix and delivered strong profitability that offset the top-line decline.
  • The company maintains a focus on proof-of-concept (PoC) projects for next-generation AI, offering pre-built solutions including Local RAG Starter Box for agent AI and ROBODEV for Physical AI, to seed future large upsell opportunities.
  • SG&A increased 25.5% year-over-year to 271 million yen, driven by a 28.1% increase in personnel costs from prior year headcount expansion and a 123.8% increase in depreciation to 340 million yen from capital investments. The company will continue pursuing capital and human investment to support future growth.

By industry segment: Cloud vendor revenue decreased significantly, offset by growth from the university, manufacturing, and R&D segments. The university segment saw particularly strong weight growth driven by the delivery and construction of the SEIRAN AI supercomputer for Tokyo University of Technology. Automotive and manufacturing segments underperformed relative to budget and continued to face uncertain, challenging conditions. Domestic-focused segments including education, information and communications, AI startups, finance, and construction delivered solid results. By service type: DX Services revenue decreased 25.6% year-over-year, while high-margin Service & Support revenue increased 26.5% year-over-year. The strong growth of high-margin Service & Support was the primary driver of overall gross margin improvement.

• Full-year operating profit progress is 65.7%, ordinary profit progress is 69.5%, and net profit progress is 72.6% against the full-year guidance, which management considers to be on track.
  • The full-year revenue forecast has not been revised despite the Q2 miss. The 650 million yen revenue shortfall from Q2 is expected to be recovered in the second half, with most of the recovery occurring in Q3 (December-February) by February, with the remainder booked in Q4. Management expects roughly a little over 300 million yen in additional revenue to be added to each of Q3 and Q4 to hit the full-year target, which is required to meet the top-line plan.
  • Operating profit is the company's most important KPI, and management is focused on achieving 10 consecutive years of profit growth. The company will balance achieving the stated full-year operating profit target with continued strategic investment for future fiscal years, with upside to the full-year profit dependent on the pace of investment execution.
  • The dividend forecast is maintained unchanged from the original guidance.
  • The Rubin GPU project is included in next fiscal year's planning, with customer delivery conservatively targeted for early 2027, and additional clarity expected at NVIDIA's GTC 2026 event in March.

• Project delays: Several approximately 100 million yen projects in automotive and manufacturing, impacted by Trump tariffs, have been delayed, with some pushed to the next fiscal year or later. The top-line shortfall from these delays was larger than management originally expected.
  • IT component price inflation and supply constraints: Global memory and storage prices have surged 3x to 5x compared to last fall, with general DRAM and NAND both up approximately 4x. Storage (including NAND flash and hard disk drives) has also seen significant price increases and supply shortages, with lead times extending to 24-36 weeks, and in some cases 52 weeks. Securing components at reasonable prices from the second half of this fiscal year through next fiscal year is an urgent priority.
  • Extended lead times for high-end GPUs: Lead times for high-end Blackwell-series data center GPUs have doubled compared to the first half, increasing from 8-12 weeks to 20-24 weeks for both B200 and B300 models. There is risk that already received orders cannot be delivered within the current fiscal period due to extended lead times.
  • Procurement strategy challenge: The company historically used case-by-case spot procurement rather than long-term contracts, due to prior stable pricing and supply. The company is now considering shifting to long-term contracts, but signing long-term contracts at current peak prices creates margin risk. The custom, flexible nature of the company's PoC solutions (which require customized component procurement per customer order rather than bulk standard orders) makes long-term contracting more challenging.

Q: Operating income for the second quarter came in below the prior sales forecast, but there is no operating income plan figure disclosed. Can we assume operating income was in line with plan?

A: The company only discloses quarterly top-line targets, and benchmarks operating income against prior year results, so there is no explicit plan figure to compare against. Management confirms there is no material deviation from expectations for operating income.

Q: There is a roughly 650 million yen revenue shortfall in Q2 against plan, and full-year guidance is unchanged. How do you expect to allocate this recovery between Q3 and Q4?

A: We have received confirmation from customers that delayed projects will move into the second half, with most final investment decisions coming close to fiscal year-end. We expect to recover a portion in February and March, with the balance coming in late March to April. Most of the recovery will be booked in Q3 (December-January-February), with recovery completed by February.

Q: Does that mean the bulk of the 650 million yen shortfall will be added to Q3?

A: That is correct. However, the exact split between Q3 and Q4 depends on customer delivery schedules and internal engineer resource allocation, so we cannot give a precise fixed breakdown. To hit the full-year top-line target, we will need to add roughly a little over 300 million yen to each quarter, which aligns with the 27 billion yen to 28 billion yen Q3 and 15 billion yen to 16 billion yen Q4 you outlined.

Q: The full-year plan forecasts a slight decline in H2 operating profit year-over-year to 324 million yen, compared to 335 million yen in H1 last year. What is your current outlook for H2 operating profit after the strong H1 recovery?

A: Operating profit is our most important KPI, and we are focused on achieving 10 consecutive years of profit growth as we work through this period.

Q: Is it possible to achieve year-over-year profit growth in H2 this fiscal year?

A: The outcome depends on how we manage the Q4 landing, and we balance hitting near-term targets with planting seeds for future growth via continued investment for next and subsequent fiscal years. We can confirm that we will work as one team to achieve the stated operating profit target, which is our commitment. Upside to the target will depend on how much we invest in future growth, and we believe balancing profit delivery with future investment is the best approach for long-term success, as lack of investment would leave us behind market trends.

Q: How different is the level of project delays in automotive and manufacturing compared to your original expectations, and can the gap be filled by small and medium projects outside these sectors?

A: On a top-line basis, the deviation from original expectations has been larger than we anticipated; we expected several more large projects to be booked in the first half or early second half, and I recognize this is an area we need to improve on. For smaller projects ranging from millions to tens of millions of yen, the key driver of fast close is the level of approval required, not the industry. These smaller projects do not require time-consuming processes like board approval or lengthy purchasing review, so they can be closed much more quickly, and they have high profitability. We are on track with profit expectations because we have filled the gap with these high-margin small projects across industries. We have already seeded next-generation AI PoC projects that can convert to larger upsell opportunities in future periods.

Q: Has progress on Physical AI and Agent AI fallen behind the pace expected one year ago?

A: No, progress is actually ahead of our original expectations. We did not expect Physical AI to gain this much market recognition this year; we originally expected broad adoption to start in the second half of next year, but many enterprises are already starting work in this area. We also expect future areas like Scientific AI and Swarm Intelligence will progress faster than originally expected.

Q: Is it correct that most Physical AI PoC wins so far have been in the university segment?

A: No, university projects are the large ticket size ones, but the overwhelming majority of Physical AI inquiries come from enterprise companies. We have been able to fill the revenue gap across manufacturing, startups, finance, automotive, healthcare, and information/communications via these PoC projects for Physical AI and Agent AI, and we will continue to focus on automotive and manufacturing as Japan's core industries.

Q: Is the current situation that you can still procure 100% of the required components, but just at much higher prices and with longer lead times?

A: It varies by component, but procurement is still possible, but prices have become extraordinarily high. We have an ongoing sense of urgency because costs are rising week-over-week for already ordered projects. Memory has the largest price impact, followed by storage: memory is up more than 4x compared to last fall, and flash storage has seen similar increases. The bigger issue for storage is extended lead times, with some lead times stretching to 52 weeks, and it is clear that supply-demand balance has broken down significantly.

Q: Is it correct that generic DRAM is up ~4x year-over-year from last fall, and NAND is up by a similar amount?

A: That is correct. Last fall's price was not anomalously low, so the increase compared to the prior quarter is similar, at 4x to 5x for memory. The impact is amplified because GPU memory sizes increase each generation, which requires proportionally more main memory for swapping, so the overall cost impact is very large. We historically did not rely heavily on long-term contracts, as pricing and supply were stable, and we used case-by-case procurement. We are now considering shifting to long-term contracts, but we are aware that locking in long-term contracts at current peak prices carries risk. Our business model relies on high flexibility to customize hardware specs for each customer PoC, rather than selling standardized products from bulk inventory, so customized per-order procurement is often required, which limits the benefit of long-term bulk contracts.

Q: Are there any other bottlenecks for procurement besides main memory, NAND, and hard disks?

A: High-end GPUs for data centers also have much longer lead times now compared to the first half. Lead times for high-end Blackwell GPUs have doubled from 8-12 weeks to 20-24 weeks, for both B200 and B300 models. If we do not manage this carefully, there is risk we cannot deliver already received orders within the current fiscal year. NVIDIA Hopper does not have significant demand, so this issue is isolated to Blackwell-series products.

Q: What is the status and expected delivery timeline for the Rubin GPU project?

A: Rubin is on our radar for next fiscal year projects, but delivery timelines and final hardware specifications are not yet finalized, so customers are currently evaluating it in parallel with Blackwell. We conservatively expect customer availability in early 2027, so we advise customers to plan for that timeline and evaluate both Rubin and Blackwell in parallel. Blackwell Ultra does not support double-precision (FP64) compute, and many Japanese customers moving from HPC to AI research value FP64 performance, which creates ongoing demand for NVIDIA DGX B200. We will get more clarity on whether Rubin will support FP64 at GTC 2026 in March.

Q: Do customers tend to downgrade memory specifications due to the high cost and procurement issues?

A: The ideal configuration has main memory equal to or double the GPU memory size. Depending on customer budget, we can flexibly propose starting with main memory equal to GPU memory size and scaling up incrementally later, to align with customer budget and delivery timelines.