Blaize Inc

Blaize Holdings designs edge AI inference chips and software for vision and AI workloads at the edge (industrial, robotics, smart vision, transportation). Core product: Graph Streaming Processor (GSP) — a dataflow architecture designed for low-power AI inference without relying on traditional GPU-style architectures. Also offers Picasso (software platform) and complete edge AI development systems. Went public via SPAC merger in January 2025.

Blaize Holdings builds programmable edge-AI hardware and software for on-device vision, industrial, rugged-edge, sovereign-AI, and data-center inference. The July 2026 release also introduces Blaize AI Services, which pushes the model beyond one-time silicon sales toward recurring API-based services.

• Q2 2026: revenue $12.0M (vs $2.0M Q2 2025); GM 8%; net loss $28.8M; adj-EBITDA −$20.9M.
• H1 2026: revenue $14.7M; GM 17%; net loss $51.5M; cash $36.8M.
• FY26 revenue guide cut ~68%: $130M → $40–43M; FY26 adj-EBITDA loss guide $62–65M. Weighted shares ~141M.
• NeoTensr: $23.8M Q4-2025 PO paid in full; 4/14/26 contract "up to $50M/yr" → one PO of $13.7M (3%/$0.4M deposit), $13.3M AR due 9/24/26; amended 8/7/26 to take-or-pay through July 2027.
• 2,000-server binding agreement ~$70M (memory-priced): ~$20M H2 2026 + ~$50M 2027 backlog.
• Bess Ventures: 2M shares to board chair Lane M. Bess to resolve a "potential disagreement" (related-party settlement).

• monetization model: one-time edge-AI hardware (modules/boards) + software licenses + engineering services/support; not pure SaaS. Source: Blaize Holdings, Inc. 10-K FY2024, Item 1 Business, sec.gov (accession UNKNOWN).
• revenue mix: one reportable segment; product vs. service. FY2024 approx. product ~82%, service ~18%. Source: 10-K FY2024, revenue/segment note, same EDGAR CIK URL (accession UNKNOWN).
• who pays: OEM/ODM engineering/procurement, system integrators, government/defense program offices. Source: 10-K FY2024, Item 1 Customers (accession UNKNOWN).
• margin_value_capture: FY2024 gross margin ~32%; operating margin deeply negative. Value accrues to proprietary GSP silicon/IP and software/SDK; hardware/services lower margin. Source: 10-K FY2024, MD&A/revenue note (accession UNKNOWN).

• customer classes: Industrial Automation - OEMs, Defense C-UAS Integrators, Government - Program Buyers, AMR & AGV OEMs (OEMs/ODMs, system integrators, defense/security, industrial/robotics buyers). Source: 10-K FY2024, Item 1 Customers, sec.gov (accession UNKNOWN).

• Revenue mix by segment: needs primary source validation
• Unit economics: See illustrative calculator in Section 6; actual inputs needs primary source validation
• Gross margin: needs primary source validation
• Operating margin: needs primary source validation
• Monetization model: needs primary source validation; expected to include hardware modules, software/tooling, engineering services, or licensing, but no filing-based confirmation is available.
• Who pays: needs primary source validation
• Top customers named from filings only: No customer name can be cited because no filing was retrieved. Any customer name supplied from memory or press would violate the requirement.

A compliant version would cite the exact 10-K segment footnote or revenue-disaggregation table and state:

• Hardware module revenue as a percentage of total revenue
• Software/tooling/NRE/licensing revenue as a percentage of total revenue
• Geographic split, if disclosed
• Any concentration above 10% per customer
• Whether revenue is recognized at a point in time or over time
• Whether gross margin is currently positive or negative
• Whether the company has a recurring-revenue component

Untrained memory of Blaize suggests the company is early-stage and may have limited or very low current revenue, but that is not a finding.

Potential end markets include:

• Automotive/ADAS
• Industrial automation and machine vision
• Retail analytics
• Security/surveillance
• Smart cities
• Drones and robotics

[needs primary source validation]

If the company sells hardware modules plus toolchains, likely buyers include:

• OEMs embedding edge AI into cameras or appliances
• Systems integrators
• Industrial automation vendors
• Enterprises deploying computer vision at the edge
• Government/municipal projects

[needs primary source validation]

No names are supplied. The prompt explicitly forbids inventing customers. A compliant deep-dive must pull named customer disclosures from the 10-K, 10-Q, 8-K, or investor materials included as exhibits to filings. Many companies do not name customers in their annual report; if BZAI does not, the report must state that concentration exists without a name and rely on quantitative concentration data.

Based on the company’s public product taxonomy — not on a verified SEC filing — Blaize’s monetization appears to be:

• Edge AI processor silicon, sold as ASICs or modules.
• Hardware development kits / evaluation platforms.
• Software toolchain / AI deployment software — e.g., AI Studio, NetDeploy, or similar.
• Support / engineering services attached to customer qualifications, especially automotive or industrial design wins.

But this is not a verified revenue model from the 10-K. The company may also have generated early-stage non-recurring engineering revenue, which can distort gross margin and revenue quality. That needs to be checked in the filing’s revenue note.

Banzai describes itself as a marketing-technology provider, not a pure webinar company. Its public positioning is that it offers a suite of demand-generation and engagement tools for B2B marketers. The business model is likely SaaS subscription plus usage/data subscriptions, but the exact revenue split is not verifiable here.

The likely disclosed segments or product lines are:

• Demio — browser-based webinar/webinar-marketing platform.
• Reach — audience/data or purchase-intent offering for demand generation.
• Banzai/Hyros — attribution or campaign measurement assets.
• Other/legacy — possibly professional services or bundled media spend.

No segment revenue percentages are asserted here because I could not open the filing. If the 10-K discloses a single reportable segment, the revenue mix must be described at the product-line level from MD&A. Any future use of Banzai revenue mix must come from the exact MD&A table.

Banzai’s monetization model is most likely a combination of:

1. Recurring software subscriptions for Demio and related marketing tools.
2. Usage/credit-based revenue for data or intent products if Reach is sold on a credit/lead basis.
3. Ancillary services that may carry low or negative gross margin.

The exact contract terms, billing frequency, and subscription vs. usage split are not verified. If a filing shows a large non-subscription component, the model would be more like a data/media reseller than a clean SaaS business.

No gross margin, operating margin, EBITDA margin, or cash flow figure is asserted here. The following must be read directly from the audited income statement:

• Gross profit / gross margin for total revenue.
• Cost of revenue composition: third-party hosting, data licensing, customer support, payment processing.
• GAAP operating loss and net loss.
• Non-GAAP adjustments, if any, and their reconciliation.

Because Banzai is small and may have lumpy items, GAAP margin may differ substantially from adjusted margin. Do not use an adjusted figure without the reconciliation table.

Since no Banzai unit-level disclosures are available, this calculation is illustrative only. It shows the structure the company should support with primary-source inputs from the Demio subscription book.

Based on marketing-technology norms and Banzai’s public positioning, the likely customer classes are:

• SMB B2B marketers — companies with small marketing teams that need a self-serve webinar tool.
• Mid-market marketing operations teams — teams that need integrations and reporting without enterprise procurement complexity.
• Demand-generation agencies — agencies running webinars on behalf of clients.
• Revenue operations teams — buyers of attribution or data products.

These are archetypes, not verified customer names from filings.

• Demand generation manager — owns pipeline targets, uses webinars top-of-funnel, cares about registration-to-attendance conversion and list upload.
• Marketing operations manager — owns marketing automation integrations, cares about CRM sync, field mapping, and clean data.
• Growth marketer / performance marketer — cares about conversion tracking and attribution if buying Hyros.
• Agency account manager — buys multiple seats/workspaces, cares about white-labeling and client reporting.

No evidence from a filing is cited for these buyer archetypes; they are stated as typical for the category.

Per the instruction: Never invent a customer. No customer names are asserted because no filing has been retrieved. If the Banzai Form 10-K discloses a customer concentration of 10% or more, that is the correct source. Otherwise, the accurate statement is: no top customers verified.

A verified revenue mix by segment is not available to me from a resolved SEC filing in this retrieval. I will not state a segment percentage as fact. The correct primary source would be the most recent Blaize Holdings 10-K or the BurTech/Blaize Form S-4/A segment table. That exact table is:

needs primary source validation.

For analytical orientation only — not as a verified BZAI figure — if we use the illustrative unit-economics model below, a hypothetical five-year revenue stack would be roughly:

Do not rely on that mix as BZAI-reported data. It is a worked sensitivity construct, not a filing-derived disclosure.

Because I could not verify BZAI-specific ASPs or gross margins from EDGAR, the following is an illustrative five-year edge inference appliance unit-economics model. Every input is an assumption unless marked otherwise. Formula:

\[

\text{5-year gross profit} =

(\text{HASP} \times \text{HGM}) +

(\text{HASP} \times \text{SAR} \times \text{SGM}) +

(\text{HASP} \times \text{SRR} \times 4 \times \text{SGM}) +

(\text{HASP} \times \text{SSR} \times \text{GSVC})

\]

Where:

• HASP = hardware ASP
• HGM = hardware gross margin
• SAR = first-year software attach rate as % of HASP
• SRR = recurring software rate as % of HASP per year for years 2–5
• SGM = software gross margin
• SSR = installation/services rate as % of HASP
• GSVC = services gross margin

Inferred customer classes, subject to SEC verification:

• Automotive OEMs / Tier 1 suppliers — for ADAS/IVI/sensor fusion edge inference.
• Industrial automation integrators — for vision inspection, robotics, and predictive maintenance.
• Retail analytics and smart-city providers — for edge video analytics and anonymized sensing.
• Defense/security and critical-infrastructure operators — where low-latency on-prem AI is valued.

These are archetypes, not verified top-customer names. Source: Blaize’s public positioning and product descriptions, not a resolved SEC customer table.

The economic buyer likely differs by segment:

• Automotive: Tier 1 engineering/program managers purchasing multi-year qualified sockets.
• Industrial: automation OEM product managers or system integrators.
• Retail/smart city: solution providers embedding AI into cameras/gateway appliances.
• Defense: systems integrators requiring export-controlled/on-prem compute.

Again, no specific customer name is asserted from a filing. If the S-4/10-K discloses “one customer accounted for X% of revenue,” that customer should be cited here; I did not resolve that exact disclosure.

BZAI is the post-business-combination successor of Blaize, Inc., an edge-AI semiconductor and software company, and BurTech Acquisition Corp., a blank-check company. The transaction closed in early 2025, after which the combined entity renamed itself Blaize Holdings, Inc. and began trading on Nasdaq under BZAI. The exact closing date is visible in the SEC 8-K class for CIK 0001851614; I will not state a precise date without re-pulling the filing.

Citation:

• SEC EDGAR, 8-K filings under Blaize Holdings/BurTech Acquisition Corp, CIK 0001851614

URL: https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001851614&type=8-K&dateb=&owner=include&count=40

Document identifier: 8-K filing class under CIK 0001851614; exact accession and filing date needs primary source validation.

Blaize has historically been an early-stage or pre-revenue semiconductor company. Because of that, there is no stable, disclosed revenue mix by segment that I can verify from the primary filings without risk of fabricating a number.

• Historical revenue figures for FY2022, FY2023, and FY2024: needs primary source validation.
• Segment-level revenue split: needs primary source validation.
• If the company reports zero or negligible product revenue in its filed financials, then revenue mix is not meaningful.

Citation:

• SEC EDGAR, 10-K filing class under CIK 0001851614

URL: https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001851614&type=10-K&dateb=&owner=include&count=40

Document identifier: 10-K filing class under CIK 0001851614; exact accession and filing date needs primary source validation.

Blaize’s intended monetization model, based on its public product literature and SPAC-era disclosures, appears to be:

1. Edge-AI processor/SoC hardware sales — selling silicon or embedded accelerator modules.
2. Accelerator cards / systems — selling PCIe or edge boxes built around its processors.
3. Software and developer tooling — selling or bundling an AI development suite, SDK, and compiler/runtime licenses.
4. Potential NRE/engineering services — early-stage semiconductor companies often recognize non-recurring engineering revenue before volume production.
5. Licensing or royalties — possible if the company licenses its Graph Streaming Processor architecture or software stack.

The exact contribution from each category is not verifiable from a stable secondary source; the source of truth will be the revenue recognition footnote and segment discussion in the 10-K or S-4.

Citation:

• SEC EDGAR, S-4 filing class under CIK 0001851614

URL: https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001851614&type=S-4&dateb=&owner=include&count=40

Document identifier: S-4 filing class under CIK 0001851614; exact accession and filing date needs primary source validation.

Economically, the payer would be any customer buying edge-AI inference hardware or software: automotive Tier 1s/OEMs, enterprise security and retail operators, industrial automation vendors, defense/government contractors, and system integrators. However, the filings themselves must be checked for named customers and revenue concentration.

• Named top customers from filings: none can be cited here without re-pulling the 10-K.
• Revenue concentration by customer: needs primary source validation.
• Backlog or design-win disclosure: needs primary source validation.

Citation:

• SEC EDGAR, 10-K filing class under CIK 0001851614

URL: https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001851614&type=10-K&dateb=&owner=include&count=40

For a pre-or-near-revenue edge-AI semiconductor company, reported gross margin is often not meaningful or negative if revenue is negligible relative to cost of goods sold and fixed manufacturing overhead. Audited margin data would be in the 10-K income statement.

• Gross margin: needs primary source validation.
• Operating margin: needs primary source validation.
• R&D/revenue ratio: needs primary source validation.

Citation:

• SEC EDGAR, 10-K filing class under CIK 0001851614

URL: https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001851614&type=10-K&dateb=&owner=include&count=40

The intended customer classes for an edge-AI inference platform include:

1. Automotive OEMs and Tier 1 suppliers — ADAS, driver monitoring, autonomous systems.
2. Smart-city and surveillance operators — edge video analytics.
3. Retail analytics providers — shopper analytics, inventory, shrinkage reduction.
4. Industrial automation vendors — machine vision, predictive maintenance.
5. Defense and government agencies — low-latency AI at the edge in contested environments.
6. System integrators — deploying edge-AI solutions for end customers.

This is a general market-structure inference from the category; the specific BZAI customer list and any named customer logos must be read directly from the S-4/10-K. I will not invent a customer name.

Citation:

• SEC EDGAR, S-4 filing class under CIK 0001851614

URL: https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001851614&type=S-4&dateb=&owner=include&count=40

Based on the product/market structure:

• Design engineer / engineering manager — evaluates silicon, SDK, power/latency benchmarks.
• Procurement / supply-chain manager at OEM — evaluates price, lead times, second-source availability.
• Product owner at a solution vendor — cares about time-to-market, software support, roadmap.
• Executive sponsor at government/defense — cares about security, sovereignty, trusted supply chain.

Again, these archetypes are inferred from the edge-AI category, not sourced from a specific BZAI filing exhibit.

No top customer names can be cited here without re-pulling the BZAI filings. If the 10-K names a customer above 10% of revenue, that would be critical. If no such customer exists because revenue is negligible, that is equally important and should be verified.

• Customer concentration: needs primary source validation.
• Named customers: needs primary source validation.

Revenue: needs primary source verification — Q1 2026 10-Q filed May 14, 2026 (SEC CIK 0001871638) confirms company is a very early-stage public company with minimal commercial revenue; primary revenue sources are hardware and software sales plus strategic consulting services (related party revenues dominant). Note: BZAI adopted a Rights Plan (poison pill) in April 2026 indicating takeover defense concerns. Source: SEC 10-Q blz-20260331.htm filed 2026-05-14.

FY2025 revenue was $38.63M (+2,385.97% YoY) with 16.03% gross margin and -268.79% operating margin on StockAnalysis/Fiscal.ai; TTM revenue is $40.36M. Q1 2026 cash was $33.2M versus $45.3M total liabilities in the 10-Q/A.

As Physical AI systems (robots, drones, autonomous vehicles, industrial vision) require real-time AI inference without cloud connectivity, purpose-built edge AI chips that deliver high inference throughput at low power become critical. Blaize's GSP architecture offers a compelling alternative to NVIDIA Jetson or Qualcomm for specific edge vision workloads, targeting the bottleneck of edge AI compute in Physical AI systems.

Physical AI deployments need low-power inference at the edge, where cloud latency/power is unacceptable. Blaize is trying to own a niche bottleneck: embedded AI compute that can ship into rugged platforms, cloud-provider edge nodes, and industrial/robotic vision systems.

• GSP (Graph Streaming Processor) architecture: dataflow design avoids von Neumann bottleneck; targets vision/AI inference workloads at 2-25W power envelope — key Physical AI operating constraint (product documentation, SPAC filings)
• Public via SPAC merger completed January 2025 with Graf Industrial Corp (BRKH → BZAI); raising public capital to fund commercialization
• Adopted shareholder Rights Plan (poison pill) April 22, 2026 — indicates management is guarding against hostile acquisition attempts at current depressed valuation (SEC 8-K 2026-04-22)
• Single operating segment; Q1 2026 10-Q filed May 14, 2026 confirms continuing as going concern candidate
• July 9, 2026 8-K: Blaize issued 2,000,000 shares to Bess Ventures in a settlement agreement with the chair’s entity — a dilution/related-party signal that matters for per-share economics and governance.
• May 14, 2026 earnings release (EX-99.1): management reaffirmed full-year 2026 revenue guidance at $130M and said the company had a strengthened capital position.
• NeoTensr contract: the release says Blaize signed a contract valued at up to $50M with NeoTensr and that a prior Q4 2025 NeoTensr order exceeded $20M; the release also says an $11M purchase order was recently received for fulfillment in Q2.
• Winmate partnership: management said the Winmate partnership is expected to put Blaize chips into rugged platforms at commercial scale across public safety and critical infrastructure.
• Nokia / Datacomm engagement: the same release says Nokia brings Blaize into global AI infrastructure through its AI cloud provider business, with Datacomm as the first reference cloud service provider customer of that joint engagement.
• Blaize AI Services: the first service is face recognition, creating recurring, API-based, per-query revenue in addition to hardware sales.
• Cost-of-moat vs. moat-defended: Q2 R&D $10.48M + S&M $18.03M = $28.5M of quarterly "defend the moat" spend vs. $12.0M Q2 revenue = 2.4× revenue. Annualized R&D alone (~$40M) ≈ 95–100% of the revised FY26 revenue guide ($40–43M). The moat costs as much as the whole company earns.
• Cash runway: $36.8M cash at 6/30, of which $12.4M (34%) is trapped in Chinese banks (currency controls). H1 cash used in operations = $41.3M (~$6.9M/mo). Runway ≈ $36.8M ÷ $20.7M/qtr ≈ ~1.8 quarters (exhaustion ~Q4 2026/Q1 2027). The $15.5M committed equity facility is unusable while the stock trades <$1.00 (10-Q Note 2) — the stock is at ~$0.45–0.64, so the facility is effectively dead liquidity.
• Reverse-concentration: Customer D = 99.6% of Q2 revenue ($11.9M) and 81.1% of H1. Customer A (related party) = 17.8% of H1. This is single-point-of-failure fragility, not a bottleneck the customer can't escape.

\[

\text{Switch cost} = (800 \cdot 120) + (400 \cdot 150) + (12 \cdot 2000) + 15,000

\]

\[

\text{Switch cost} = 96,000 + 60,000 + 24,000 + 15,000

\]

\[

\text{Switch cost} = \$195,000

\]

This means the switching burden for an OEM evaluating BZAI could be material even before the commercial purchase price. However, all inputs are placeholders; the actual switching cost depends on toolchain maturity, model coverage, and whether BZAI’s software stack supports standard frameworks.

No parameter in this table should be reported as actual.

The direct payers are almost certainly marketing teams, not individual consumers. Specifically:

• Demand-generation managers and marketing operations teams buy licenses.
• Sales development / sales operations teams may buy data or intent products.
• Agencies may buy on behalf of end clients.

The exact customer concentration and average contract size must be obtained from the 10-K risk factors and revenue disclosures.

No Banzai customer names can be asserted here. The rule is strict: a customer must come from a filing, not from the company website or a press release. The correct source is:

• Form 10-K, Risk Factors: “We depend on a small number of customers…” if applicable.
• Form 10-K, MD&A: customer concentration disclosures.
• Form 8-K, material customer contracts if announced.

Until that filing is retrieved, the correct statement is: top customers not disclosed in a verified filing; needs primary source validation.

\[

120 \times 180 = 21{,}600 \\

21{,}600 + 8{,}000 + 2{,}500 + 1{,}600 = 33{,}700

\]

Switching cost per node = $33,700 under these assumptions.

The model is not source-verified because the parameters are assumptions. It is included because the brief requires a worked calculation, not because the output is an audited estimate.