SiFive
B2B IP licensing: upfront license fees + per-unit royalties on shipped chips + custom design services. Fabless and capital-light - customers integrate SiFive cores into their own SoCs; SiFive manufactures nothing.
Earlier rounds (Series A-E, ~$366M total pre-Series F per Sacra; Series E $61M led by SK hynix, Aug 2020) had no publicly disclosed valuations, so the trail starts at the Intel offer reference point. Total raised ~$766M per Tracxn including Series G; not officially disclosed. The Intel $2B+ figure is an offer price, not a priced round - included as a market reference only.
Earnings, margins, COGS & capex
SiFive is private and publishes no financials. Internal documents reported by Bloomberg (Mar 2024) showed 2023 revenue of $38.2M, a 94% gross margin, a $113.1M net operating loss, $186M of lifetime deals signed, and a company projection of ~$60M licensing revenue in 2024. Sacra's estimates corroborate the revenue level and put the 2023 mix at roughly 60% IP licensing / 30% custom design services / 10% boards and tools. Royalty economics lag design wins by years, so recognized revenue understates contracted value across 500+ design wins and 10B+ cores shipped (company, Apr 2026). The Apr 9 2026 $400M Series G (oversubscribed, led by Atreides Management with NVIDIA, Apollo, T. Rowe Price, Point72 Turion, D1 Capital, Capital Group, Prosperity7, Sutter Hill) at $3.65B - up from $2.5B+ at the Mar 2022 Series F - funds the data center CPU push and was described by CEO Patrick Little in press interviews as the final private round before an IPO.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~6¢ is cost of goods and ~0¢ operating expense, leaving ~94¢ of operating profit.
Revenue trend
Margins
IP licensing structurally high-margin; mix shift away from lower-margin design services should help
R&D-heavy scale-up phase funded by successive private rounds; later years not disclosed
COGS structure
Minimal true COGS - core deliverable is RTL/IP (94% gross margin in 2023 per Bloomberg-reported documents). Cost base is dominated by processor design engineering (R&D), verification, EDA tooling, and customer support engineering; design-services revenue carries meaningfully lower margin than license/royalty revenue.
Capex
Negligible physical capex (no fabs, no manufacturing). Investment flows to R&D on scalar/vector/matrix core roadmaps, software ecosystem work (Linux, compilers, CUDA/NVLink Fusion enablement), and hyperscaler customer enablement.
Latest earnings
n/a
None. CEO Patrick Little told press the Series G is the final private round before an IPO (statement made to media, not in the press release; no exchange or timeline confirmed as of 2026-07)
- Last priced round
- $400M Series G at $3.65B, Apr 9 2026 (Atreides Management lead; NVIDIA, Apollo Global Management, T. Rowe Price, Point72 Turion, D1 Capital, Capital Group, Prosperity7 Ventures, Sutter Hill Ventures)
- Prior round
- $175M Series F at $2.5B+ post (Coatue lead, Mar 2022; Intel Capital, Qualcomm Ventures, SK hynix, Samsung Ventures, AMD/Xilinx Ventures, Western Digital Capital, Prosperity7 among backers)
- Total raised
- ~$766M including Series G (Tracxn tally; Sacra counted ~$366M pre-Series G; some 2026 press cites ~$970M) - not officially disclosed
- Cores shipped (cumulative)
- 10B+ (company, Apr 2026)
- Design wins
- 500+ designs featuring SiFive IP; 8 of top 10 semiconductor companies are customers (both company claims)
Growth drivers
- AI data center CPU pivot — high-performance RISC-V cores targeting hyperscaler custom silicon, backed by NVLink Fusion integration (announced Jan 15, 2026) letting RISC-V CPUs attach directly to NVIDIA GPU infrastructure; first such designs expected in market around 2027
- 2nd Gen Intelligence family (Sep 8, 2025) — new X160/X180 Gen 2 plus upgraded X280/X390/XM Gen 2 combining scalar, vector, and matrix compute for AI from far-edge IoT to the data center; two US tier-1 semiconductor licensees signed pre-announcement
- Royalty ramp from 500+ design wins and 10B+ cumulative cores shipped (company, Apr 2026) - royalties compound as licensed designs reach volume
- RISC-V share gains — RISC-V International announced 25% market penetration reached in 2025, ahead of Omdia's 2024 forecast of 25% by 2030; SHD Group projects 21B+ RISC-V chips shipped annually by 2031
- Arm licensing-fee inflation and Arm's move toward selling its own chips push licensees toward the royalty-free open ISA
- Design wins in aerospace/gov (NASA HPSC flight computer built by Microchip on SiFive X280) and hyperscaler accelerators (Google has used X280 alongside TPU infrastructure)
Bull & bear
SiFive is the Arm of RISC-V at the exact moment the industry wants an open alternative for AI data centers - with NVIDIA's blessing, hyperscaler pull, 94% gross margins, and a royalty model that compounds off 500+ design wins already signed.
- Royalty flywheel: 10B+ cores shipped and 500+ design wins mean royalty revenue mechanically ramps for years even without new license wins; Bloomberg-reported documents showed $186M of lifetime deals signed by 2023 against only $38M recognized
- NVIDIA investment + NVLink Fusion + CUDA enablement collapse the biggest historical objection to RISC-V in the data center - software and interconnect
- Arm's own actions (royalty hikes, suing its largest licensee, competing with customers) are the best salesperson for an open ISA; SiFive is the highest-credibility commercial vendor to capture that defection
- Crossover investors (T. Rowe Price, Apollo, Point72 Turion, D1) at $3.65B with a stated pre-IPO framing suggests institutional diligence supports a near-term public path
- RISC-V International announced 25% market penetration in 2025, ahead of Omdia's 25%-by-2030 forecast; if the high-value data center tail follows the embedded base, the leading independent IP vendor grows into and beyond the valuation - Arm trades at a premium multiple on ~$4B FY2025 revenue as the comp ceiling
A $3.65B price on a ~$38M-revenue, ~$113M-operating-loss IP vendor whose core market is being commoditized by the very openness it champions, whose data center ambition is unproven, and whose biggest prospective customers are also its likeliest competitors.
- Valuation-to-revenue gap is extreme (~60-95x the last known 2023 revenue, even allowing for growth since); the royalty ramp must be nearly flawless to grow into it
- Last known P&L (2023, Bloomberg-reported) showed a $113.1M net operating loss on $38.2M revenue - a ~3x opex-to-revenue burn that successive rounds have had to fund
- Open-ISA paradox: RISC-V eliminates the ISA moat - hyperscalers can hire teams or use open cores (or Tenstorrent's licensable Ascalon) instead of paying SiFive
- No shipped, benchmarked, hyperscale-volume server CPU exists on SiFive IP today; NVLink Fusion silicon is a 2027 story, while Arm Neoverse (Graviton, Axion, Grace) has a multi-generation head start
- Semiconductor IP is a winner-take-most, long-sales-cycle business with lumpy license revenue - the 2023 restructuring showed how quickly the model strains; if the IPO window closes or S-1 numbers disappoint, late-round preference stacks could sit above employee/secondary holders
- Geopolitics cuts both ways - export controls on RISC-V collaboration would shrink the global ecosystem that makes the ISA viable against Arm
What it is worth
Last priced round + comparable-company framing (no public financials to model)
$1.5-2.5B or a postponed IPO
if disclosed revenue disappoints (2023's $113M operating loss on $38M revenue repriced), the data center push stalls against Arm Neoverse, or open-core commoditization compresses pricing - reverting toward the Series F mark / Intel-offer zone
$3-4.5B
IPO near or modestly above the Series G mark; royalty ramp on 500+ design wins continues, data center wins still early (first NVLink Fusion silicon ~2027)
$6-10B+ at IPO
if the S-1 shows revenue scaled well past $100M with a marquee hyperscaler data center win and clean royalty growth - priced as 'the open-ISA Arm'
The Apr 9 2026 Series G set $3.65B with crossover investors, a 46% step-up from the $2.5B+ Series F (Mar 2022). Against the last known revenue print ($38.2M in 2023, Bloomberg-reported internal documents) this is an extreme multiple, so the round is priced on the royalty-ramp option and data center TAM, not trailing financials. Arm (ARM) is the public comp: it trades at a premium multiple on ~$4B FY2025 revenue with 95%+ gross margin - the bull case is SiFive capturing a single-digit slice of that profit pool at similar 94% gross margins. Reference point: Intel reportedly offered $2B+ to acquire SiFive in Jun 2021 (Bloomberg) and SiFive stayed independent. All scenarios are illustrative, not price targets, and hinge on undisclosed financials that only an S-1 will reveal.
SWOT
Strengths
- Founding pedigree — created by RISC-V ISA inventors (Krste Asanovic, Yunsup Lee, Andrew Waterman, UC Berkeley) - unmatched technical credibility in the ecosystem
- Broadest commercial RISC-V IP portfolio — Essential (embedded), Performance (application), Intelligence (AI/vector-matrix), Automotive lines
- Capital-light licensing model with 94% gross margin (2023) and compounding royalty streams from 500+ design wins
- Strategic cap table — NVIDIA, Apollo, T. Rowe Price, Atreides - crossover investors signal IPO-readiness and NVIDIA alignment de-risks the data center software story
- $400M fresh capital (Apr 2026) funds a multi-year data center roadmap without near-term financing pressure
Weaknesses
- Revenue ($38.2M in 2023 per Bloomberg-reported documents) is tiny relative to a $3.65B valuation - the price is a bet on future royalties, not current financials
- Deeply loss-making at last known print ($113.1M net operating loss in 2023); no proven high-end silicon at hyperscale volume yet - the data center CPU pivot is early and unproven against Arm Neoverse and x86 incumbents
- Customer concentration among large semiconductor firms; royalty timing is outside SiFive's control
- 2023 restructuring (layoffs, exit from pre-packaged custom-silicon business) showed strategy volatility
- RISC-V software ecosystem still trails Arm/x86 for server workloads despite RHEL/Ubuntu/CUDA progress
Opportunities
- Hyperscaler custom CPUs — every major cloud (AWS, Google, Microsoft, Meta) builds custom silicon - RISC-V removes Arm license cost and control risk
- NVLink Fusion partnership makes SiFive the first RISC-V vendor in NVIDIA's NVLink Fusion ecosystem (after Arm, Intel, and AWS) - a credible wedge into AI-rack infrastructure from 2027
- Arm's litigation posture (Qualcomm dispute) and its move to compete with its own licensees push the industry toward ISA neutrality
- Automotive and aerospace/defense RISC-V adoption (NASA HPSC halo effect, safety-certified roadmaps, Quintauris ecosystem pull)
- IPO window: a public listing would give currency for talent and M&A and make it the only pure-play RISC-V IP stock on a US exchange
Threats
- Arm Holdings (ARM) defends its server beachhead aggressively with Neoverse, pricing leverage, and the deepest software ecosystem
- Free/open competition — RISC-V's openness means customers can use open-source cores or in-house designs, capping pricing power
- Well-funded RISC-V rivals — Tenstorrent ($693M Series D at $2.6B post, Dec 2024), Andes Technology, Ventana Micro, plus China's ecosystem (Alibaba T-Head XuanTie, Nuclei) accelerating the commoditization of cores
- US-China export-control expansion around RISC-V could constrain the addressable market or complicate the ecosystem SiFive depends on
- Hyperscalers may internalize RISC-V design entirely rather than license, following their Arm-CPU playbook
Moats, dependencies & bottlenecks
Moats
Credibility and architectural depth are real but people-dependent
Once a core ships in silicon, replacing it requires a respin; royalties persist for product lifetimes
Widest commercial RISC-V lineup, but Andes and Tenstorrent are closing gaps
RHEL/Ubuntu enablement, RISC-V International founding member) The most differentiated new moat if data center wins materialize
Deliberately absent - RISC-V is open; SiFive must win on implementation quality, not lock-in
Dependencies
strategic partner + investor NVLink Fusion/CUDA alignment is the data center wedge; NVIDIA's NVLink Fusion ecosystem also includes Arm, Intel, and AWS, so the relationship is not exclusive
Google/GOOGL, Microsoft/MSFT, Meta/META) The $3.65B thesis requires at least one marquee RISC-V server/accelerator-host win reaching volume
compilers, cloud-native stack) Server viability depends on community + commercial software SiFive only partially controls
Standard fabless-IP dependencies; EDA vendors also sell competing CPU IP (Synopsys ARC, Cadence Tensilica)
Congressional pressure to restrict RISC-V collaboration with China could fragment the ecosystem
Advantages
- Only scaled, US-based, pure-play commercial RISC-V IP vendor - the default call for any team leaving Arm
- Capital-light model: 94% gross margin (2023) and a $400M raise fund years of R&D without fabs or inventory
- NASA HPSC (via Microchip) and aerospace/defense wins provide a high-trust, long-duration revenue halo
- NVIDIA cap-table alignment no RISC-V rival can currently match
- First-mover brand in RISC-V: 'SiFive' is shorthand for commercial RISC-V the way Arm was for licensed cores
Weaknesses
- Revenue ($38.2M in 2023 per Bloomberg-reported documents) is tiny relative to a $3.65B valuation - the price is a bet on future royalties, not current financials
- Deeply loss-making at last known print ($113.1M net operating loss in 2023); no proven high-end silicon at hyperscale volume yet - the data center CPU pivot is early and unproven against Arm Neoverse and x86 incumbents
- Customer concentration among large semiconductor firms; royalty timing is outside SiFive's control
- 2023 restructuring (layoffs, exit from pre-packaged custom-silicon business) showed strategy volatility
- RISC-V software ecosystem still trails Arm/x86 for server workloads despite RHEL/Ubuntu/CUDA progress
Bottlenecks
- Proof-point gap — needs a publicly benchmarked, volume-shipping high-performance CPU on SiFive IP to convert data center interest into licenses; first NVLink Fusion designs are not expected in market before 2027
- Royalty lag: 2-4 years from license signing to meaningful royalty flow constrains near-term revenue growth
- Server-class software maturity (JIT runtimes, databases, virtualization) still behind Arm64
- Senior CPU-architect talent scarcity — competing for the same people as Apple, NVIDIA, Tenstorrent, and hyperscalers
Top signals & trends
Top signals
46% valuation step-up over 4 years; T. Rowe/Point72 Turion/Apollo/D1 presence signals IPO prep
First RISC-V vendor in NVLink Fusion - strongest external validation of the data center thesis to date
Watch for S-1 filing; would force first real financial disclosure
The thesis-critical milestone remains outstanding as of 2026-07
Core-IP commoditization pressure rising
Directionally positive but unverifiable until IPO disclosures; last hard numbers are the 2023 Bloomberg-reported documents
Trends
Every custom accelerator needs host/management CPUs - SiFive X-series' beachhead (e.g., Google's use of X280 alongside TPU infrastructure)
Structural tailwind for open-ISA alternatives
Mostly embedded/edge-AI first; data center is the high-value tail
RISC-V's global openness is both its engine and its regulatory exposure
Creates license customers today, in-house competitors tomorrow
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
EDA tools for core design/verification (also a competitor via ARC IP)
EDA/verification tooling
Foundry where most customer silicon using SiFive IP is fabbed (indirect)
Has used SiFive X280 alongside TPU infrastructure (disclosed 2022)
NASA HPSC spaceflight processor built on SiFive X280; PolarFire SoC uses SiFive cores
Licenses SiFive cores for automotive/embedded lines
Investor (Qualcomm Ventures, Series F) and RISC-V collaborator (wearables JV context)
Series F investor; memory-adjacent compute interest
Company claim (repeated 2022-2026); most names undisclosed
Dominant CPU IP incumbent; Neoverse owns the non-x86 data center. The company SiFive must displace
Private (Jim Keller); licenses Ascalon RISC-V cores + AI accelerators; $693M Series D at $2.6B post (Dec 2024), led by Samsung Securities and AFW Partners with Bezos Expeditions participating
Publicly listed Taiwanese RISC-V IP vendor, strong in embedded/consumer; the volume-share rival
ARC processor IP plus EDA leverage bundled into customer flows
Tensilica DSP/processor IP; same bundling dynamic as Synopsys
Defend the data center CPU socket SiFive ultimately targets
AWS Annapurna, Google, Microsoft, Meta can build RISC-V without licensing SiFive
Mainland-China RISC-V vendors accelerating open-core commoditization - competitive-landscape context only, not a buy/own call