Velaura AI

1.1 Identity, Product, and Business Model

Velaura AI, Inc. is a Silicon Valley-based semiconductor company developing ultra-low-power silicon and software technologies for AI compute infrastructure. The company was previously known as Auradine, a U.S.-based bitcoin-mining ASIC maker, before rebranding to Velaura AI in March 2026 as part of a strategic pivot toward AI compute. Velaura's flagship offering, Titan Core, is not a standalone chip but a silicon design and IP platform: customers provide their own RTL processor design, and Velaura applies proprietary ultra-energy-efficient circuit and physical-design libraries to deliver an optimized, drop-in physical layout at advanced process nodes such as 3nm and 2nm. The company claims Titan Core delivers a 2-4x improvement in performance per watt for the matrix-multiplication operations that dominate AI training and inference energy use, translating into up to 500W of savings on a typical 1000W-class GPU or XPU. Velaura monetizes this IP through an upfront licensing fee plus royalties tied to the power savings a customer actually achieves, a structure CEO Rajiv Khemani has explicitly compared to Arm's processor-licensing model. The underlying low-power design methodology has already been validated at commercial scale, with more than 30 million ASICs in production across leading semiconductor nodes, giving Velaura a real-world reliability and yield track record that most first-time AI-silicon entrants lack. The company targets three adjacent markets: hyperscale AI data centers seeking to cut electricity and cooling costs, Physical and Embodied AI systems (robots, drones, autonomous machines) operating under strict power and thermal budgets, and Edge AI deployments requiring energy-efficient local processing.[CO001, CO002, CO003, CO004, CO005, CO006]

1.2 Leadership, Founders, and Governance

Rajiv Khemani, co-founder and CEO, is the public face of Velaura AI across every funding and product announcement reviewed for this chapter, and Mayfield's Navin Chaddha describes the 2026 round as his firm's fourth partnership with Khemani specifically. Manu Gulati serves as Chief Development Officer and is named alongside Khemani as the technical core of the leadership team; Mayfield notes this is its second partnership with Gulati. Beyond the founding duo, Velaura's Team page lists Sanjay Gupta (President, Strategy & GTM), YJ Kim (President, Products), Tim Vehling (VP, Product), Brian Campbell (VP of Engineering Operations), and Atul Dhablania (Chief Systems Operations Officer), rounding out a leadership bench the company and press coverage describe as combining alumni of Apple, NVIDIA, Google, Qualcomm, and Marvell. Governance is anchored by a board that mixes venture and strategic interests: Navin Chaddha (Mayfield), Umesh Padval (Seligman Ventures, which holds the lead Series A board seat), Dipender Saluja (Capricorn Investment Group), Fred Thiel (CEO, MARA - both an investor and a former Teraflux hardware customer), Sriram Viswanathan (Celesta Capital), and Intel CEO Lip-Bu Tan, who sits on the board via Walden International and lends significant semiconductor-industry credibility. Two academic/technical advisors, Aditya Grover (CTO, Inception AI; UCLA professor) and Magnus Egerstedt (Provost, UNC Chapel Hill; roboticist), extend the company's Physical AI and machine-learning expertise. The concentration of public commentary on Khemani and Gulati, and the near-total absence of independent biographical detail on the rest of the executive bench, signals meaningful key-person dependence on the founding duo.[CO012, CO013, CO014, CO015, CO025, CO029]

1.3 Funding History, Valuation, and Investors

Velaura AI's capitalization history spans four disclosed institutional rounds under two corporate identities. As Auradine, the company raised an $81 million Series A in May 2023 (led by Celesta Capital and Mayfield) to launch its Teraflux bitcoin-mining ASIC line, an $80 million oversubscribed Series B in April 2024 that also produced $80 million in customer bookings, and a $153 million Series C in April 2025 led by StepStone Group with Maverick Silicon, Premji Invest, Samsung Catalyst Fund, Qualcomm Ventures, Mayfield, and MARA participating, explicitly to expand beyond bitcoin mining into blockchain and AI infrastructure. On August 18, 2026, newly rebranded Velaura AI announced a $110 million Series A round led by Seligman Ventures, with new investors Capricorn Investment Group and Prosperity7 Ventures joining repeat backers Mayfield, Maverick Silicon, MARA, Premji Invest, Samsung Catalyst Fund, and StepStone Group; the round pushed Velaura's valuation above $1 billion, achieving unicorn status. Cumulative disclosed capital raised across all four rounds exceeds $424 million, though no public source has audited or reconciled this total. MARA's dual role as both a repeat financial investor and a historical customer for Teraflux mining hardware is a governance and independence consideration worth separate diligence. Seligman Ventures itself is a newly launched $500 million venture vehicle of Seligman Investments (founded November 2025), and this Series A marks the firm's first disclosed investment in Physical AI per Managing Partner Umesh Padval.[CO007, CO008, CO009, CO010, CO011, CO021]

1.4 Traction, Deployment Scale, and Adverse Signals

Velaura AI's central traction claim is that its ultra-low-power design methodology is already proven at scale: more than 30 million ASICs manufactured on leading-edge process nodes are cited across the company's own Titan Core launch release and its Series A announcement alike, and this volume is used to substantiate manufacturing-yield and reliability claims for the newer AI-focused product. On the customer side, the company says it has ongoing engagements with three of the four largest cloud providers; CEO Rajiv Khemani reportedly confirmed this to Reuters per Heise Online's coverage, but he declined to name any of the hyperscalers involved, and no customer has independently confirmed a Velaura or Titan Core relationship. This lack of named-customer verification, combined with the absence of any independent, peer-reviewed benchmark of the claimed 2-4x performance-per-watt improvement, is the most consequential evidence gap identified in this chapter. Trade press coverage is broadly positive on the underlying power-constraint thesis but flags real execution risk: TechBooky explicitly cautions that 'a valuation above $1 billion does not guarantee commercial adoption,' and notes that chip startups require deep engineering, manufacturing partnerships, and customer validation to convert design wins into recurring royalty revenue. Independent analyst Patrick Moorhead of Moor Insights & Strategy has offered a more constructive read, saying Velaura's approach 'has the potential' to reduce total cost of ownership and ease thermal limits for customers, a hedged endorsement rather than independent verification.[CO006, CO017, CO018, CO019, CO026, CO033]

1.5 Cover Metrics and Diligence Gaps

Several cover metrics that later chapters will rely on remain undisclosed by the company and unavailable in public sources as of the run date: current headcount, revenue or run-rate specific to the AI compute business (as distinct from legacy Teraflux bookings), and a named customer list. This chapter records those as explicit nulls rather than guesses, with concrete diligence paths - for example, triangulating headcount from professional-network employee counts or requesting disclosure directly in a management interview, and requesting audited or third-party-verified power-savings benchmarks before relying on the 2-4x claim in valuation work. The unresolved and partial-status research questions in this chapter's evidence ledger (customer names, headcount, AI-specific revenue, and independent benchmark validation) should be treated as open items that later chapters (financials, customers, risks, valuation) must either close with new evidence or carry forward as named gaps.[CO038, CO019, CO034]

1.6 Exhibits

2.1 Market Boundary and Definition

Velaura AI does not sell a market-ready chip; it licenses ultra-low-power silicon design IP that other companies integrate into their own AI accelerators. That structural fact means the relevant market boundary is not simply 'the AI chip market' but the subset of AI-silicon design decisions where a third-party power-efficiency IP vendor can plausibly be licensed in: hyperscale data-center accelerators, Physical and Embodied AI systems (robots, drones, autonomous machines), and Edge AI devices, per Velaura's own market framing. The data center AI chip market itself is tracked narrowly by some publishers (SemiconductorInsight sizes it at $13.8 billion in 2026, reaching $45.2 billion by 2034) and far more broadly by others once merchant GPUs, custom ASICs, memory, and networking silicon are included (Deloitte and Gartner both place the overall AI semiconductor market at $300-500 billion in 2026, while IDC's narrower 'intelligent datacenter' segment sits at $281 billion within a $477.1 billion data-center-semiconductor total). NVIDIA alone captured roughly 87.4% of Q1 2026 merchant data-center AI chip revenue among NVIDIA, AMD, and Intel, underscoring that most of this headline market value is currently captured by GPU vendors rather than IP licensors like Velaura. Explicitly excluded from Velaura's addressable market are general-purpose CPU/server silicon, non-AI networking chips, and the finished-chip revenue hyperscalers capture internally when they design their own custom ASICs (Google TPU, AWS Trainium, Microsoft Maia, Meta MTIA) - though those same internal programs are potential IP-licensing customers for Velaura's underlying power-efficiency technology rather than direct substitutes for it.[CM025, CM001, CM002, CM004, CM019]

2.2 TAM/SAM/SOM and Contradictory Sizing Lenses

No single, evidence-constrained TAM/SAM/SOM figure exists for Velaura's specific niche of ultra-low-power AI silicon IP licensing; every available public figure describes an adjacent, broader category. On the data-center side, three independent 2026 methodologies diverge by nearly 2x even before scope differences are considered: IDC's segment-level $281 billion 'intelligent datacenter' figure, an implied ~$390 billion AI share of Gartner's $1.3 trillion total semiconductor forecast, and Deloitte's revised ~$500 billion full-year estimate. On the Physical AI side - relevant to Velaura's robotics/embodied-AI ambitions - three publishers' 2026 estimates span more than a 7x range ($0.89 billion per MarketsandMarkets' narrower 2025 base, $6.93 billion per SNS Insider, and $7.11 billion per Mordor Intelligence), reflecting materially different definitions of what counts as 'physical AI' hardware and software rather than a converging consensus. Because Velaura's revenue model is royalty-based (a license fee plus a share of measured power savings) rather than direct chip sales, even a well-defined SAM for the underlying accelerator market would overstate Velaura's own obtainable revenue; the analytically correct SOM would be measured in royalty dollars per XPU shipped, and no public source provides a bottom-up estimate of that figure. This chapter therefore preserves the sizing contradiction explicitly rather than collapsing it into one headline number, and records the missing SOM estimate as a named evidence gap.[CM002, CM008, CM005, CM006, CM007, CM027]

2.3 Buyer Segmentation and Adoption Path

Velaura's three target buyer segments have structurally different budget owners and adoption triggers. In hyperscale data centers, the buyer is the cloud infrastructure/silicon team, the payer is the hyperscaler capex budget - which is projected at $600-725 billion in 2026 for the Big Five hyperscalers, with roughly 75% directed at AI infrastructure - and the adoption trigger is power/cooling constraints on data-center expansion. In Physical and Embodied AI, the buyer is robotics/drone hardware engineering, the payer is OEM R&D and bill-of-materials budget, and the adoption trigger is battery-life and thermal-envelope requirements for always-on autonomous systems; on-device compute already accounts for 71.43% of physical-AI deployments, directly aligned with Velaura's on-chip efficiency pitch. In Edge AI, device OEM procurement teams buy chips to extend battery life and enable local inference. A fourth, cross-cutting sub-segment - internal hyperscaler custom-ASIC programs such as Google TPU, AWS Trainium, Microsoft Maia, and Meta MTIA - draws on the same capex budget as the broader data-center segment but represents a distinct adoption path: these teams already design their own silicon and would license Velaura's IP specifically to close a performance-per-watt gap versus merchant GPU competitors, not to replace an existing chip-purchase decision. Broadcom and Marvell, which together provide ASIC design services behind more than 80% of hyperscaler custom silicon, are a further complication: they could be channel partners that embed Velaura's IP into their design-service offerings, or competitors offering their own power-optimization capabilities.[CM012, CM011, CM019, CM031, CM035]

2.4 Growth Drivers and Adoption Constraints

The dominant growth driver for Velaura's thesis is that AI data-center electricity demand is on track to nearly double, from approximately 485 TWh in 2025 to 950 TWh by 2030, while the US alone faces an estimated 9.3 GW structural power shortfall in 2026 - a physical constraint that increases the value of any technology that extracts more useful compute per watt. Hyperscaler capex growth (projected at $600-725 billion in 2026, up roughly 75% year over year) expands the pool of budget that could eventually fund IP-licensing fees, and the faster growth of custom-ASIC shipments (44.6% CAGR) versus merchant GPUs (16.1% CAGR) expands the set of in-house silicon programs that are natural licensing prospects. Working against these drivers are several real adoption constraints: US-China export controls and Chinese rare-earth retaliation add compliance cost and supply-chain uncertainty; the high capital intensity of advanced-node chip design and manufacturing raises barriers to entry for competitors but does not eliminate Velaura's own execution risk; and typical 12-24 month SoC integration and foundry-qualification cycles mean that even a signed design win delays royalty cash flow by one to two years. A further constraint worth flagging for valuation purposes is that the broader industry is attacking the same power problem from multiple angles simultaneously - liquid-cooling adoption in AI servers rose from 15% in 2024 to a projected 76% in 2026 - meaning system-level cooling improvements could partially substitute for chip-level efficiency gains over time and are not solely dependent on vendors like Velaura.[CM015, CM016, CM012, CM017, CM022, CM034]

2.5 Diligence Gaps and Preserved Contradictions

This chapter deliberately preserves two categories of open item rather than resolving them with a single confident number. First, no public source sizes Velaura's specific royalty-licensing SOM; every available figure describes a broader adjacent market (data-center AI chips, physical AI hardware, or overall AI semiconductors), and blending them (as in the illustrative SAM figure in this chapter's market-sizing pyramid) is explicitly flagged as an approximation rather than a verified estimate. Second, published physical-AI market-size estimates for materially the same time window diverge by more than 7x across three independent, reputable-tier publishers, which should be read as a signal that 'physical AI' is not yet a standardized market category rather than as evidence that any one estimate is wrong. Diligence should treat any single-number TAM claim from Velaura's own materials or from any one third-party report with corresponding skepticism until a bottom-up, source-triangulated estimate is available.[CM028, CM008, CM027]

2.6 Exhibits

3.1 Competitive Landscape: Direct, Incumbent, Adjacent, and Substitute

Velaura AI competes across four distinct competitive categories rather than a single well-defined peer set. Among direct low-power AI chip peers, the field has consolidated sharply in the last year: Untether AI, a near-memory-compute competitor, shut down in June 2025 and filed for bankruptcy in October 2025 with over $128 million in liabilities, its engineering team acqui-hired by AMD; Hailo, an edge-AI chip maker with roughly 100 customers, was acquired by Microchip Technology in July 2026; and Mythic required a full architecture and leadership overhaul before raising a $125 million Series D in December 2025. Groq illustrates how quickly comparables can re-rate: after a reported $20 billion NVIDIA deal absorbed its founding technical team and IP in December 2025, Groq pivoted business models entirely and raised new capital at a $3.5 billion valuation, down from $6.9 billion a year earlier. Cerebras, by contrast, validated the segment's upside by going public in May 2026 at roughly a $56 billion fully diluted valuation. NVIDIA and AMD remain the dominant incumbents, together with Intel capturing the overwhelming majority of merchant data-center AI chip revenue (NVIDIA alone at approximately 87.4%). The most structurally important substitute, however, is internal build: hyperscalers' own custom-silicon programs (Google TPU, AWS Trainium, Microsoft Maia, Meta MTIA) are growing shipments nearly three times faster than merchant GPUs, and any of these teams could in principle achieve comparable power-efficiency gains in-house using commercial EDA tools from Cadence or Synopsys rather than licensing Velaura's IP.[CP001, CP002, CP004, CP005, CP006, CP007]

3.2 Capability, Pricing, and Business-Model Comparison

On performance-per-watt claims alone, Velaura's 2-4x improvement is not the largest headline figure in the segment - Mythic claims up to 100x efficiency gains from its analog compute-in-memory architecture - but Velaura is unusual in pairing its efficiency claim with 30 million-plus already-shipped production ASICs, a scale proof neither Mythic nor the now-defunct Untether AI could match at a comparable stage. Business models diverge sharply across the set: Velaura and its closest structural analog, Arm's historical processor-licensing business, both charge an upfront fee plus a usage-based royalty, whereas NVIDIA, AMD, Hailo, and Mythic all sell finished silicon directly and capture full unit economics (and full inventory/manufacturing risk) rather than a royalty stream. Notably, Arm itself broke from three decades of pure IP licensing in 2026 to begin selling finished AI chips directly, illustrating that even an established, successful royalty-model incumbent can face pressure to compete directly with its own licensees - a precedent risk worth tracking for Velaura's own model. On disclosure and independent verification, Velaura compares unfavorably to some peers: Hailo publicly named roughly 100 customers before its acquisition, while Velaura has not named any of the three hyperscalers it claims to be engaged with, and no independent, head-to-head benchmark of Titan Core against Mythic's or Hailo's architectures exists in the sources reviewed for this chapter.[CP018, CP014, CP015, CP016, CP029, CP020]

3.3 Switching Costs, Lock-In, and Distribution Power

Once a hyperscaler commits to integrating Velaura's Titan Core IP into a chip design, switching costs are likely meaningful: SoC integration and foundry-qualification cycles for a new silicon IP block at advanced process nodes typically span 12 to 24 months, creating real lock-in once a design win is secured but also meaning Velaura's own revenue recognition lags any announced engagement by a year or more. Distribution power in the broader AI-chip value chain is concentrated: Broadcom and Marvell together provide ASIC design services behind more than 80% of hyperscaler custom silicon, giving them significant influence over whether third-party IP like Velaura's gets incorporated into a given design - they could act as a channel partner that embeds Velaura's libraries into their design-service offerings, or as a direct competitor if they build equivalent power-optimization capability internally. NVIDIA is responding to the broader custom-silicon shift not by competing purely on standalone chip performance but by selling more fully integrated racks and systems and opening its NVLink interconnect to third-party ASICs, a strategy that could either create new integration opportunities for Velaura's IP or further entrench NVIDIA's ecosystem lock-in depending on how it plays out.[CP026, CP027, CP028]

3.4 Moat Durability and Adverse Competitor Evidence

Velaura's clearest moat is production-proven manufacturing history: 30 million-plus shipped ASICs give hyperscaler partners a real yield and reliability track record that a from-scratch competitor cannot replicate quickly. That moat is qualified, however, by the fact that this volume was built primarily on Auradine's bitcoin-mining chips rather than on AI-specific Titan Core silicon, so it derisks manufacturing execution more than it validates the AI-specific performance claim. The clearest threat to durability is commoditization risk: Cadence's Cerebrus and Synopsys's DSO.ai already offer AI-driven low-power design automation that any competitor or customer could use to pursue similar power gains in-house, and if a foundry partner (Samsung, TSMC) chose to embed equivalent libraries directly into its standard process design kits, much of Velaura's differentiation could erode. The competitive-density trend in the sector cuts both ways for this assessment: Untether AI's shutdown and Hailo's acquisition reduce the number of independent low-power AI chip vendors Velaura must compete against directly, but they could equally be read as evidence that the segment cannot yet support many independent vendors at all - a risk that applies to Velaura as much as to any peer that has already exited.[CP035, CP034, CP030, CP031, CP017, CP033]

3.5 Exhibits

4.1 Revenue Model, Pricing, and Recognition

Velaura AI's core AI-silicon revenue mechanism is a two-part structure: an upfront IP license fee plus a royalty tied to the power savings a customer measurably achieves using Titan Core, a model the company explicitly likens to Arm's historical processor-licensing business. The company's own materials illustrate the customer-side savings basis at approximately $1,300 per XPU over three years, but neither the royalty percentage, the measurement methodology for 'power savings achieved,' nor any minimum license fee is disclosed in any source reviewed for this chapter - a material gap given that the entire royalty mechanic depends on a savings measurement neither party has described in public. Because royalty revenue is tied to realized customer savings rather than unit shipment alone, revenue recognition likely lags physical chip production by however long a customer's own qualification and deployment cycle takes, plausibly the same 12-24 month window identified in the Competitors chapter for SoC integration generally. Velaura's legacy Teraflux bitcoin-mining hardware business, by contrast, used a conventional direct hardware-sale model; MARA's own SEC filing shows this legacy business operated on a prepayment basis, with MARA advancing $22.3 million for product purchases in the quarter ended March 31, 2025 against a $57.2 million outstanding balance, fully settled by the most recent filing period with no new advances. No source discloses whether or how much of Velaura's current revenue mix still derives from this legacy hardware business versus the newer Titan Core licensing model.[CI007, CI008, CI009, CI028, CI030, CI005]

4.2 Cost Structure and Unit Economics

No source reviewed discloses Velaura's actual gross margin, cost-of-goods structure, R&D spend, or customer acquisition cost. Industry benchmarks provide only a rough comparable frame: fabless semiconductor companies typically report gross margins of 50-65% (with leaders like NVIDIA exceeding 70%), and early-stage fabless startups typically spend 20-40% of revenue on R&D (spiking above 50% pre-revenue) while burning $500,000 to $2,000,000 per month, with 12-18 months of runway considered prudent. None of these figures is Velaura-specific, and applying them directly risks understating the unique economics of a royalty-based IP licensor versus a unit-sale fabless chip vendor: Velaura's model requires far less manufacturing capex and inventory risk than a typical fabless comparable, since it is not the party paying for wafer starts, but it also means its revenue per design win is a small fraction of the underlying chip's total value, similar to Arm's historically estimated 1-2%-of-ASP royalty rate. The only independently corroborated cash-flow-adjacent data point available for any part of Velaura's business is MARA's SEC-filed disclosure of its product-purchase advances and investment carrying value, which relates to the legacy Teraflux hardware business rather than the new Titan Core IP licensing model this valuation thesis primarily rests on.[CI011, CI012, CI013, CI025, CI027, CI029]

4.3 Public Traction and Headcount

Velaura AI's only publicly disclosed traction figure with a specific dollar amount is the $80 million in customer bookings Auradine reported alongside its April 2024 Series B raise, a figure that predates the AI pivot by roughly two years and says nothing about Titan Core-specific traction. Headcount estimates diverge meaningfully across sources: Tracxn's tracked employee-count trend shows 62 employees as of March 26, 2026, while other analyst databases (PitchBook, AI Market Watch) suggest a broader 100-150 employee range, and LinkedIn-style company-size bands show 101-500 employees. None of these figures is company-confirmed, and the spread itself (roughly 2-3x between the low and high estimates) is a useful signal that headcount tracking for a recently rebranded private company is unreliable without a direct company disclosure. No source discloses current ARR, run-rate revenue, active customer count with names, or any AI-specific bookings figure post-pivot.[CI014, CI015, CI016, CI017, CI018]

4.4 Capital Adequacy and Financing Dependency

Velaura AI/Auradine's cumulative disclosed equity capital raised across four rounds - the $81 million Series A (2023), $80 million Series B (2024), $153 million Series C (2025), and $110 million 2026 Series A - totals approximately $424 million, though this is a simple sum of separately announced figures rather than an audited or company-confirmed total. The August 2026 round's stated use of funds is to expand engineering and customer-facing teams and deepen strategic partnerships, per both the company's own release and lead investor commentary from Mayfield, but no source discloses current cash-on-hand, monthly burn, or resulting runway. No source discloses any debt facility or project-finance arrangement separate from the disclosed equity rounds. The single independently filed (SEC-sourced) financial data point available for any part of Velaura's business is MARA Holdings' related-party disclosure: an $85.4 million investment carrying value as of March 31, 2026, one board seat, and $11.9 million plus $2.7 million in ASC 321 fair-value gains recorded when Velaura's later financing rounds established a higher observable price for MARA's existing preferred and common stock holdings - itself indirect evidence that Velaura's valuation has risen across financing rounds, corroborating the company's own unicorn-valuation claim from an independent, audited source rather than press coverage alone.[CI033, CI010, CI024, CI034, CI001, CI002]

4.5 Financial Verdict: Revenue Quality, Margin Path, and Diligence Blockers

Velaura AI's financial profile as of the run date is best described as evidence-thin but directionally corroborated: the one independently filed data point available (MARA's SEC-disclosed investment and its associated fair-value gains) supports the company's rising-valuation narrative, but every operating metric that would let a diligence team assess revenue quality, margin path, or capital efficiency - royalty rate, gross margin, burn, runway, ARR, and named customer concentration - remains undisclosed. Trade press coverage independently flags two caveats directly relevant to this verdict: TechBooky cautions that a unicorn valuation does not guarantee commercial adoption, and Heise Online notes that Velaura's refusal to name its chips or customers obscures the very revenue base against which any future royalty stream would need to be measured. Given the capital-intensive comparable signal from Groq's need for a second large raise in the same sector, and given that Velaura's royalty economics likely resemble Arm's historically thin per-unit take (roughly 1-2% of ASP) rather than a full hardware-sale margin, the central financial diligence blocker is not whether Velaura's technology works, but whether its licensing revenue can scale fast enough, and transparently enough, to justify a $1 billion-plus valuation before the next financing round is needed.[CI031, CI032, CI035, CI026, CI025, CI021]

4.6 Exhibits

5.1 Product Definition and Architecture

Velaura AI does not sell a finished chip; it delivers a physical-design IP and toolflow product through a three-stage engagement model described on its own site: the customer submits its RTL design plus priorities for area, power, and other factors; Velaura applies proprietary low-voltage IP libraries and a proprietary toolflow built for yield and reliability; the customer receives an optimized, node-specific GDS layout or chiplet as output. Architecturally, Titan Core targets the matrix-multiplication (MATMUL) operations that dominate AI training and inference energy use, claiming a 2-4x reduction in the energy those operations require through proprietary circuit and library technology validated at advanced 3nm and 2nm process nodes. Integration is designed to preserve full functional equivalence at a lower operating voltage without requiring software changes, positioning Titan Core as a drop-in physical-design optimization rather than a new architecture the customer must adopt wholesale. Velaura maintains two distinct product/solution lines reflecting its corporate history: Ultra Low Power AI Compute (the Titan Core IP platform for AI accelerators) and Blockchain (its legacy air, immersion, and hydro-cooled Bitcoin-mining hardware), with the underlying low-voltage engineering discipline shared across both.[CE001, CE005, CE006, CE007, CE008, CE030]

5.2 Deployment, Workflow, and Critical Dependencies

For a hyperscaler silicon team, the alternative to licensing Titan Core is pursuing similar power-efficiency gains in-house using commercial EDA tools such as Cadence's Cerebrus AI Studio or Synopsys's DSO.ai, both of which provide AI-driven low-power design automation as of 2026 - meaning Velaura's value proposition rests on its pre-built libraries and engineering expertise outperforming what a well-resourced in-house team could achieve with commodity tooling. Once a customer engages, Velaura's delivery depends on several external dependencies outside its direct control: the customer must be willing to share proprietary RTL under NDA, advanced-node foundry capacity (at partners such as TSMC or Samsung Foundry) must be available to physically implement the ultra-low-voltage design, and Velaura's own specialized low-voltage circuit engineering team - a comparatively small, hard-to-replicate group - must execute the physical-design toolflow. Notably, Samsung Catalyst Fund's position as both a Velaura investor and Samsung's foundry-linked venture arm creates a plausible (though undisclosed) strategic foundry relationship worth separate diligence. On the customer side, Velaura's own materials estimate up to $100 million per year in power savings for a typical data center, explicitly built on stated assumptions (10 cents/kWh electricity pricing and roughly $433.30 per chip per year across 100,000 to 250,000 units) rather than a customer-specific measured result, and this transparency about assumptions is notably greater than the company's undisclosed royalty-rate methodology covered in the Financials chapter.[CE003, CE023, CE026, CE022]

5.3 Maturity, Evidence Strength, and Differentiation Across Segments

Evidence strength varies sharply across Velaura's three target segments. The data-center product line is best evidenced: the underlying low-voltage IP and toolflow has been validated across more than 30 million production ASICs (primarily from the legacy Teraflux bitcoin-mining business), giving Titan Core a real manufacturing-yield and reliability foundation that independent analyst Patrick Moorhead of Moor Insights & Strategy has reviewed favorably, saying the approach has the potential to reduce total cost of ownership and ease thermal limits for customers. By contrast, the Physical AI and Edge AI segments remain marketing-stage: neither Velaura's Physical AI page nor its Solutions page names a specific robotics or edge-device customer, discloses a deployed unit count, or provides technical specifications distinct from the data-center pitch. The legacy Teraflux hardware line, meanwhile, has the strongest named-customer evidence of all three areas (MARA), with disclosed technical specifications (9.8 J/TH efficiency, up to 600 TH/s, 4nm process, multi-cooling-method variants) - though it is one process generation behind the 3nm/2nm nodes Titan Core now targets, and its strategic priority following the 2026 AI pivot is unclear. Independent, peer-reviewed benchmarking of the specific 2-4x MATMUL efficiency claim or the $100 million/year savings estimate does not exist in any source reviewed for this chapter.[CE004, CE014, CE018, CE019, CE020, CE024]

5.4 Trust, Compliance, and Developer-Signal Gaps

Velaura's public trust surface is thin: the company publishes a general privacy notice (dated August 2024, still under its prior Auradine, Inc. legal name in the version reviewed) but no dedicated security certification (such as ISO 27001 or SOC 2) or safety/quality certification (such as AEC-Q100 or ISO 26262, which would be particularly relevant given its Physical AI/robotics ambitions) is disclosed in any source reviewed. As a hardware IP company, Velaura also lacks the kind of public developer surface a software company would have: no GitHub repository, Hacker News thread, Stack Overflow tag, or public developer forum discussion specific to Titan Core was identified. The company does maintain an active careers presence on Lever, confirming ongoing engineering hiring, though the retrieved content did not disclose specific role-level technical stack detail. Following this skill's guidance for companies without a public developer surface, the closest available practitioner-community proxies are broader semiconductor-industry venues: SEMI's August 2026 'AI Techniques in Semiconductor Manufacturing' workshop and the wider 2026 semiconductor conference calendar (DAC, SEMICON WEST) show active practitioner engagement with the general AI-in-silicon topic Velaura operates in, even though neither venue provides Velaura-specific technical discussion. Company IP protection is described only as 'patented' technology in marketing language, with no specific patent number confirmed via USPTO Patent Public Search or Google Patents in this chapter's research.[CE009, CE010, CE011, CE012, CE013, CE027]

5.5 Exhibits

6.1 Customer Base Segmentation

Velaura AI's disclosed customer base splits cleanly into two eras with very different evidence quality. The legacy Teraflux bitcoin-mining hardware business claims 30 or more industrial mining operators and 40 or more data center operators cumulatively, of which exactly two - MARA Holdings and Genesis Digital Assets Limited (GDA) - are individually named and independently corroborated in public sources. The new Titan Core AI-silicon business, by contrast, claims engagement with three of the four largest cloud providers but names none of them, and two additional target segments - Physical AI/robotics OEMs and Edge AI device makers - have no disclosed customer, deployed unit, or case study of any kind. Buyer, user, and payer roles differ meaningfully across segments: Bitcoin-mining customers combine buyer and payer in mining-operations finance teams, while hyperscaler Titan Core engagements would involve separate infrastructure/capex budget owners and cloud silicon design-team users. This segmentation pattern - strong named evidence for a legacy, strategically de-prioritized business and zero named evidence for the new business the valuation thesis actually depends on - is the central customer-diligence finding of this chapter.[CU007, CU008, CU009, CU019, CU028, CU034]

6.2 Named Customer Proof: Production vs. Pilot

MARA Holdings provides Velaura's strongest customer evidence: an SEC 10-Q filing independently confirms MARA advanced $22.3 million to Velaura (then Auradine) for product purchases in the quarter ended March 31, 2025, against a $57.2 million outstanding balance, and independent press reporting separately confirms MARA's total H1 2025 Teraflux purchase reached $73.3 million and was fully delivered by June 30, 2025 - genuine production deployment corroborated by a regulatory filing rather than only a press release. However, MARA is simultaneously a Velaura investor holding a board seat and $85.4 million in investment carrying value, which means this best-evidenced relationship is not an independent, arms-length reference. Genesis Digital Assets (GDA), by contrast, is a genuinely independent customer with no disclosed investor relationship to Velaura: in June 2025 it agreed to purchase 1,000 Teraflux AT2880-277 air-cooled miners for its 40 MW Glasscock County, Texas facility, with a named, on-the-record testimonial from Executive President Abdumalik Mirakhmedov and confirmed participation in ERCOT's demand-response program - though whether that specific deployment has reached full production at the facility is not independently confirmed as of the run date. For the AI-silicon business specifically, no named customer exists at all: Velaura's own materials and GamesBeat's interview coverage describe the three unnamed hyperscaler engagements as being in technical evaluation and RTL-benchmark stages, not confirmed production deployment, and Heise Online's independent reporting explicitly flags that Velaura's CEO declined to name any of them even when asked directly by Reuters.[CU001, CU002, CU003, CU004, CU005, CU006]

6.3 Retention, Durability, and Repeat Usage

No public source discloses a net revenue retention, gross revenue retention, churn, renewal rate, or customer satisfaction metric for any Velaura customer relationship, legacy or new. The only available repeat-engagement signal is qualitative: MARA's relationship with Velaura progressed from a Series B/C financial investor to a 2026 Series A investor to a disclosed $73.3 million hardware purchaser, illustrating a form of land-and-expand behavior across relationship types even though it is not a formal retention metric and is confounded by the investor relationship itself. Following the March 2026 rebrand, Velaura's own statements indicate existing Teraflux customers retain warranty support even as the company shifted unsold inventory to in-house mining rather than continuing external sales - a signal of continuity of obligation to existing customers, but also of declining new-customer acquisition in that product line specifically. No source provides a durability signal (cohort, renewal, or retention data) for the hyperscaler Titan Core engagements, which remain at an evaluation stage without any customer relationship old enough to measure retention against.[CU016, CU030, CU033, CU013, CU029]

6.4 Expansion, Concentration, and Adverse Market Signals

The single largest concentration risk this chapter identifies is that Velaura's most valuable and best-evidenced customer transaction (MARA's $73.3 million H1 2025 purchase, corroborated by SEC filing) comes from a related party rather than an independent buyer, while its one genuinely independent named customer (GDA) represents a single transaction with no confirmed repeat business. Compounding this, the addressable market for Velaura's only currently-revenue-evidenced business - legacy Teraflux hardware - is structurally shrinking: major public Bitcoin miners including Cipher Mining, Core Scientific, TeraWulf, and Bitdeer are redirecting capital and power capacity toward AI/HPC colocation deals rather than expanding Bitcoin-mining hardware fleets, and public miners collectively shed 21% of Bitcoin hashrate in 2026 as AI revenue accelerated industry-wide. This creates a difficult transition window for Velaura: its legacy customer base is contracting just as its new AI-silicon customer base (the three unnamed hyperscalers) remains entirely unconfirmed. No evidence of a failed deployment, customer complaint, or negative review was found for either business line in the sources reviewed, which is a mild positive signal but should be read against the broader absence of any independent customer review platform (G2, Gartner Peer Insights) coverage of Velaura at all.[CU006, CU020, CU014, CU015, CU035, CU026]

6.5 Exhibits

7.1 Severity-Ranked Risk Overview

This chapter identifies five risk categories with material investment implications for Velaura AI. Ranked by combined likelihood and severity, the highest-priority risk is customer/revenue concentration: the $1B+ Series A valuation depends almost entirely on an AI-silicon licensing business with zero named customers, while the only independently corroborated customer relationship (MARA) is simultaneously a related-party investor. Close behind is commoditization risk from commercial EDA tooling (Cadence, Synopsys) that could let hyperscaler customers replicate Velaura's power-efficiency gains in-house. Regulatory risk from evolving US-China semiconductor export controls is material but indirect, transmitted through Velaura's undisclosed foundry and hyperscaler relationships rather than a direct, confirmed exposure. Operational risk centers on the structural contraction of Velaura's legacy Bitcoin-mining hardware customer base as the industry pivots to AI/HPC colocation, removing the company's only currently revenue-evidenced business just as its new business remains unproven. Financial/model risk stems from total non-disclosure of cash-on-hand, burn, runway, and royalty-measurement methodology. Each of these risks carries a distinct investment implication: diligence should not treat Velaura's unicorn valuation as proof of de-risked commercial execution, given how much of the underlying evidence traces back to company claims rather than independent confirmation.[CR030, CR016, CR002, CR013, CR019, CR009]

7.2 Regulatory and Legal Risk

As of January 15, 2026, US export control policy for advanced computing chips shifted from a presumption of denial to case-by-case license review for exports to China and Macau, adding certification and third-party testing requirements for covered products. Velaura's own disclosures do not address its exposure to this regime, but the transmission channel is real: if any of its three undisclosed hyperscaler engagements involve China-serving infrastructure or China-linked foundry capacity, these rules would directly affect deal economics and timelines. Separately, AI-related patent litigation has surged industry-wide, with legal-industry analysis counting over a thousand AI patent lawsuits globally in the past five years; no lawsuit or patent dispute specifically names Auradine or Velaura AI as of the run date, but the company's own unverified claim of 'patented' technology - without a specific patent number confirmed via USPTO or Google Patents - is itself a diligence gap that leaves IP-strength unverifiable. Independent confirmation of Velaura AI, Inc.'s Delaware legal-entity status was not completed in this chapter (the state registry requires a fee-gated direct query), and no sanctions-list or denied-party exposure was identified for the company or its named leadership.[CR001, CR002, CR003, CR004, CR005, CR006]

7.3 Operational and Partner/Dependency Risk

Velaura's most acute operational risk is commoditization: Cadence's Cerebrus AI Studio and Synopsys's DSO.ai both provide AI-driven low-power design automation as of 2026, meaning any hyperscaler customer could in principle attempt to replicate Titan Core's efficiency gains in-house using commodity tooling rather than licensing Velaura's IP. A second operational risk is the structural contraction of Velaura's legacy Teraflux hardware customer base: public Bitcoin miners collectively shed 21% of hashrate in 2026 as AI revenue accelerated, and roughly 70% of top miners are reportedly using AI-related income just to survive the current bear market, directly shrinking demand for new mining-hardware purchases even as Velaura's own post-rebrand decision to shift unsold inventory to in-house mining signals it recognizes this decline. On the partner-dependency side, two relationships carry outsized concentration risk: MARA Holdings is simultaneously Velaura's largest independently corroborated customer transaction and a related-party investor with a board seat, and Samsung Catalyst Fund is both an investor and the venture arm of a company whose foundry business could be either a manufacturing partner or a competitor for Velaura's power-efficiency IP. The entire hyperscaler engagement pipeline - the source of any future AI-silicon revenue - remains unnamed and unconfirmed, the single highest-severity dependency identified in this chapter.[CR016, CR013, CR014, CR015, CR011, CR012]

7.4 Financial/Model Risk and People/Execution Risk

No public source discloses Velaura's cash-on-hand, monthly burn, or runway following its August 2026 Series A, nor the measurement methodology underlying its royalty revenue model - a financial/model risk compounded by the fact that Groq, a well-funded independent AI-silicon competitor, needed a second large capital raise in the same month despite a prior $750 million round, illustrating persistent capital intensity across the sector. This chapter finds no evidence of fraud, restatement, or accounting irregularity at Velaura or in MARA's disclosure of its Velaura investment; the ASC 321 fair-value treatment MARA applies follows standard equity-investment accounting practice, a mitigating signal on financial-integrity risk specifically. On people/execution risk, Velaura's public narrative concentrates heavily on CEO Rajiv Khemani and CDO Manu Gulati, with minimal independent biographical detail on the rest of the executive bench, creating meaningful key-person dependence; no executive departure or resignation was identified in 2025-2026, a mitigating continuity signal, and board oversight (including Intel CEO Lip-Bu Tan and MARA CEO Fred Thiel) provides governance credibility that partially offsets operating-level key-person concentration.[CR019, CR027, CR035, CR038, CR024, CR025]

7.5 Mitigations, Monitoring Indicators, and Thesis-Break Triggers

Velaura's strongest available mitigant against pure execution-risk skepticism is its 30 million-plus ASIC production history, which addresses manufacturing-yield concerns but does not resolve the customer-concentration, disclosure, or commoditization risks identified separately in this chapter. This chapter defines six monitorable thesis-break triggers diligence teams should track: the absence of a named hyperscaler design win within 12-18 months of the Series A close; any competing low-power IP library release from a major EDA vendor or foundry partner; continued year-over-year decline in legacy hardware bookings without offsetting AI-silicon revenue; any deterioration in MARA's disclosed investment or purchasing relationship; another capital raise within 12-18 months without a disclosed revenue inflection; and departure of either founding co-founder. None of these triggers has fired as of the run date, but each represents a specific, source-grounded event diligence should monitor rather than treating the current valuation as a settled outcome.[CR029, CR031, CR032, CR030, CR028]

7.6 Exhibits

8.1 Investment Thesis and Anti-Thesis

Velaura AI's investment thesis rests on three pillars: production-proven ultra-low-power silicon IP (30 million-plus ASICs shipped), a credible leadership and board bench (including Intel CEO Lip-Bu Tan and MARA CEO Fred Thiel), and favorable macro tailwinds - a 9.3 GW US power shortfall and $600-725 billion in 2026 hyperscaler AI infrastructure capex - that increase the theoretical value of performance-per-watt technology. The anti-thesis is equally concrete: zero named customers exist for the Titan Core AI-silicon business the valuation is actually built on, no royalty rate or revenue figure of any kind is disclosed, and Heise Online's independent reporting confirms Velaura's CEO declined to name any of the three claimed hyperscaler engagements even under direct questioning. Compounding this, Arm - Velaura's own explicit business-model comparable - broke from three decades of pure IP licensing in 2026 to begin selling chips directly, a precedent that raises long-term doubts about the durability of royalty-based economics as a standalone model. This chapter's overall judgment is that both the thesis and anti-thesis are evidence-supported, which is precisely why a confident buy or avoid recommendation is not warranted with current information.[CV001, CV016, CV023, CV020, CV018]

8.2 Current Financing and Valuation Context

Velaura AI closed a $110 million Series A on August 18, 2026, at a valuation exceeding $1 billion, led by Seligman Ventures with participation from Capricorn Investment Group, Prosperity7 Ventures, Mayfield, Maverick Silicon, MARA, Premji Invest, Samsung Catalyst Fund, and StepStone Group. No source discloses the exact post-money valuation figure, share price, or preference structure beyond the 'more than $1 billion' headline, nor any liquidation-preference, anti-dilution, or option-pool detail that would let diligence distinguish common-equity value from preferred-stock-inflated structuring. The single independently filed corroboration of a rising valuation mark comes from MARA Holdings' SEC 10-Q, which recorded $11.9 million and $2.7 million in ASC 321 fair-value gains on its existing Velaura holdings when a later financing round established a higher observable price - real, audited evidence that Velaura's valuation has risen across rounds, even though it does not independently confirm the specific $1B+ figure. Velaura's cap table mixes financial VCs with strategic and related-party investors (Samsung Catalyst Fund, MARA), a structure that adds interpretive complexity: MARA's rising fair-value marks are simultaneously the best evidence of Velaura's increasing valuation and a related-party data point that should not be treated as fully independent.[CV001, CV002, CV007, CV014, CV015]

8.3 Comparable Set and Bull/Base/Bear Scenarios

No comparable in this chapter's set is a precise stage-and-business-model match for Velaura. Arm Holdings, the closest royalty-model public comparable, trades at a $270.57 billion market cap against $4.92 billion in FY2026 revenue - a roughly 52-55x price-to-sales ratio reflecting its mature, diversified scale rather than a transferable benchmark for an early-stage licensor. The direct AI-silicon sector comparables show extraordinary volatility within a single year: Cerebras validated the segment's upside with a ~$56 billion IPO in May 2026, while Groq's valuation was reset from $6.9 billion to $3.5 billion following a reported $20 billion NVIDIA deal that absorbed its founding team, and Untether AI - a directly comparable low-power AI chip startup - went bankrupt within roughly sixteen months of still operating. Applying even the conservative end of typical semiconductor IP licensing multiples (8x revenue) to Velaura's $1B+ valuation would imply roughly $125 million in annual revenue is needed to justify the price - a figure with no public evidence of support. Against this backdrop, the bull case requires at least one hyperscaler engagement to convert to a named design win within 18 months; the base case assumes gradual, partial conversion consistent with typical 12-24 month qualification cycles; and the bear case - a real tail risk given the comparable set's demonstrated volatility - assumes no conversion and continued legacy-business decline, potentially forcing a down round or worse.[CV003, CV004, CV005, CV009, CV008, CV010]

8.4 Recommendation, Exit Readiness, and Final Diligence Asks

This chapter's recommendation is 'track' (research-more) with medium confidence, a high risk rating, and an unresolved valuation stance - not a generic company-quality score, but a direct consequence of the near-total absence of revenue, royalty-rate, and named-customer evidence needed to underwrite a confident price-sensitive call. This is consistent with preferring research-more over false precision when core valuation inputs are missing. Exit readiness is entirely unproven: no secondary-market transaction, tender offer, or IPO/M&A timeline has been disclosed, though Velaura's board composition (Intel's Lip-Bu Tan, MARA's Fred Thiel, and multiple experienced venture partners) provides governance credibility comparable to what facilitated Cerebras's IPO and Groq's acquisition-adjacent outcome. Five diligence asks would most efficiently resolve this chapter's uncertainty, in priority order: (1) actual revenue and royalty rate, (2) at least one named hyperscaler customer reference, (3) cap-table and preference terms, (4) current cash position and runway, and (5) an independent technical benchmark of the claimed 2-4x efficiency improvement. Resolving even two or three of these would likely move this chapter's recommendation decisively toward either a buy or an avoid call.[CV033, CV028, CV036, CV037, CV038, CV039]

8.5 Exhibits