Blaize Holdings, Inc. (BZAI) Earnings

Management highlights

- Market Position & Strategic Alignment * Management identifies a key industry shift: value in AI is moving from training large models to efficient inference, which aligns directly with Blaze's designed architecture * Two high-growth converging trends Blaze is positioned to win: 1) Physical AI (AI deployed on edge field devices, vehicles, and industrial equipment, where cloud cannot deliver required response speed and data control) and 2) New purpose-built AI data centers focused on inference rather than just training, with multi-vendor, hybrid infrastructure * Management sees long-term value in the software layer that schedules and optimizes inference workloads, which is Blaze's core focus - Product & Go-To-Market Updates * Blaze has expanded geographically, securing its first European purchase order for thousands of units, growing activity across Asia-Pacific, and beginning pipeline buildout in the U.S., creating a more diversified customer base than ever before * Aerorobotics and ruggedized platforms segment is a scalable, predictable revenue engine: Blaze is selected for edge deployments with tight size, weight, and power constraints, and generates recurring revenue after initial solution qualification * Hybrid AI platform, launched two quarters ago, supports large national-scale industry programs (including facility supervision, manufacturing quality grading, and fuel retail monitoring) that are progressing through qualification; Blaze is also in active discussions for two sovereign national data center build programs (one 150 megawatt facility) * Development is ongoing for a next-generation AI inference product for production environments, which will add confidential computing capabilities for sovereign customers and support higher-performance workloads; development will be paced against customer demand and business capacity * AI services (currently including facial recognition, with document processing, video analytics, and model orchestration in development) will be priced per rack, per megawatt, or per fleet, and are expected to become an increasingly important margin contributor over time - Operational & Financial Updates * Q2 2026 operating expenses totaled $31.5 million, including a $7.1 million bad debt provision for the Starshine receivable, $1 million additional R&D investment for the next-generation chip, and a $2.8 million one-time non-cash charge for a related party settlement * Blaze ended Q2 with $36.8 million in cash, up $3.6 million from Q1, supported by $9.4 million in customer payments and $32.8 million in net proceeds from a Q2 equity offering

Guidance

- Full year 2026 revenue guidance is a large downward revision from the prior $130 million to a new range of $40 million to $43 million; 90% of the orders supporting this guidance are already in-house, with high confidence in the revised target - Full year 2026 adjusted EBITDA loss is projected to be between $62 million and $65 million, widened from prior guidance due to lower revenue, the Starshine receivable provision, and supply chain inflation, partially offset by operating expense cuts and shifted next-gen chip payment timing - Second half 2026 gross margin is expected to be between 17% and 19%, up from the 8% Q2 2026 gross margin as mix shifts to higher-margin offerings - Q3 2026 revenue is expected to be similar to Q2 2026, with Q4 2020 revenue slightly higher; most upside from unguided opportunities would likely hit in Q4 - Blaze projects ending 2026 with $50 million in backlog (binding committed orders not yet fulfilled), with strong expectations for 2027 revenue to be 2.5x to 3x the 2026 guided midpoint - Most large unguided national hybrid AI platform programs are expected to start generating revenue in Q1 2027

Segment performance

Blaze Holdings reports two core product/revenue segments: 1) Silicon NSDK (designed into OEM products for autonomous systems, robotics, and ruggedized equipment): Revenue contribution for Q2 2026 was not separately disclosed in absolute terms; this segment delivers scalable, recurring revenue once design wins are secured, with customers providing monthly forecasts and deposits to support inventory planning. 2) Hybrid AI Platform (vertical integrated stack for AI inference, supporting managed AI services): The majority of Q2 2026 total revenue came from this segment's third-party server hardware, which drove the overall low quarterly gross margin. For Q2 2026, Blaze posted total company revenue of $12 million, up from $2.7 million in Q1 2026 and $3 million in H1 2025 (H1 2026 total revenue hit $14.7 million, up 390% year-over-year). Total company gross profit was $0.9 million, for an overall gross margin of 8% (down from 58% in Q1 2026 due to the mix shift to low-margin third-party hardware).

Risks & headwinds

- Multiple late-stage pilot opportunities did not convert to purchase orders as expected, and the Starshine customer has an unpaid outstanding balance that has been fully reserved, with meaningful uncertainty about the future of that commercial relationship - Customer deployment scaling and procurement timelines have been slower than forecast: cloud and data center customers have extended technology qualification timelines, government programs have longer procurement cycles, and a Middle East smart city opportunity has been pushed to an extended field trial due to regional uncertainty - Material memory price inflation has occurred this year, as industry capacity shifts to high bandwidth memory, and suppliers now require advanced payments; these constrained supply conditions are expected to persist, and Blaze may not be able to immediately pass through all cost increases to customers - Over-reliance on a small number of large deals has created material quarterly revenue volatility, as individual delayed projects can significantly impact full-year results - The company will need to manage cash burn carefully; while current cash position is adequate, there is uncertainty around securing non-dilutive debt financing or customer advanced payments to meet future working capital requirements

Analyst Q&A

• Q: Why are so many customer purchase orders being delayed? Is it weak end demand, AI ROI concerns, or memory price increases? /

  A: Most delays are tied to the sequential nature of end customer deployments. Many projects are linked to physical camera rollouts in the field, which must be completed before Blaze hardware and software are ordered. For cloud and data center customers, delays stem from their own need to ramp end-user demand for their AI services before they purchase additional capacity from Blaze. Management confirmed underlying end demand for Blaze's solutions remains strong, and delays are purely timing-related, not demand-related.

• Q: Based on current backlog, what is your early outlook for 2027 revenue growth? /

  A: The $50 million projected end-of-2026 backlog is almost entirely committed from the company's largest customer, Neotensor, as an amendment to the previously announced $50 million contract. Management expects 2027 full-year revenue to be 2.5x to 3x the 2026 guided revenue midpoint.

• Q: Are all delayed opportunities still in the pipeline, or have they been lost? Do memory prices need to fall for these opportunities to reaccelerate? /

  A: All delayed opportunities remain active in the pipeline, and the overall pipeline continues to grow; management has just pulled forward the most certain near-term deals into 2026 guidance, pushing slower-moving opportunities to 2027. Blaze is already managing memory costs via forward component buying (it has already invested $8-$9 million in pre-purchased components) and reserves the right to adjust customer pricing to offset ongoing memory inflation, so price decreases are not required to reaccelerate deal closing.

• Q: What is the expected size and geographic focus of the large national hybrid AI programs currently in the sales process? /

  A: Most of these large opportunities are located across Southeast Asia, where Blaze's partner-led go-to-market model (including existing partnerships like Nokia) has accelerated sales cycles. The scale of thousands of servers for these programs, referenced in prior disclosures, is a reasonable estimate of their potential size.