Back to catalogue
№ 018 24 filings · 2021-09-30 → 2026-06-30

AGORA, INC.

AGORA
Software and IT Services Share price $4.35 Market cap $368m Overall fit 665 /1000

Strong fit on three of four dimensions — genuine AI-infrastructure receiver, real operating leverage, valuation floor at net cash — offset by material China exposure and ongoing HQ capex drain that keep it out of the top band.

Fair value range $5.50–$8.50 Mid case · $592m
Absolute upside +60.8% vs current market cap
Conviction 3/5 confidence in undervalued call
Supports the call
  • Cash + deposits ($362m) essentially equal to market cap — floor is well-anchored
  • Seven consecutive quarters of GAAP profitability and reaccelerating revenue (Q2 2026 +18% YoY, above guidance)
  • Aggressive share buyback (44.6m ADSs / 26% of shares retired since 2022)
Limits the call
  • Large Chinese Shengwang division carries ADR/geopolitical discount that is difficult to quantify
  • Voice AI Agent revenue not separately disclosed — AI ramp is still narrative, not measured
Methodology

Sum-of-parts: net cash/investments + 1.5-3.0x EV/sales on operating business

In one line · bull case

Real-time-engagement infrastructure company trading at net cash while its Voice AI Agent product is in early commercial ramp and high operating leverage means any revenue upside drops disproportionately to profit.

In one line · biggest risk

Roughly half of revenue is generated by a China-based subsidiary under a Cayman Islands ADR structure, exposing shareholders to regulatory, macro and delisting risk that could compress fair value regardless of operating performance.

Drivers
AI beneficiary 68 /100
Real-time infrastructure for conversational voice AI; sells directly into the AI stack rather than spending on it, though revenue contribution still small and unquantified.
Operating leverage 75 /100
Fixed ~$27m/qtr opex base with 64% gross margin; 10-20% revenue upside would swing operating result by multiples.
Earnings vs expectations 72 /100
Beat guidance in each of the last 5-6 quarters with sequential margin expansion.
Growth momentum 68 /100
Revenue growth reaccelerated from +3.6% (Q4 2024) to +18.0% (Q2 2026); Q3 2026 guidance implies continued momentum.
Moat 42 /100
SD-RTN network and developer ecosystem provide some moat, but LLM vendors (OpenAI, Google) and CPaaS peers compete directly.
Earnings quality 62 /100
GAAP profits are real but small; operating cash flow inconsistent and flattered by high interest income on cash pile.
Management quality 58 /100
Consistent buyback execution and returned to profitability, but historical growth misses and large HQ capex commitment temper score.
Cyclicality 40 /100
SaaS-like recurring usage; some sensitivity to Chinese consumer/enterprise cycles and live-shopping demand.
Leverage 20 /100
$361.7m cash/deposits vs. $99m HQ construction borrowings — meaningful net cash position (~$263m).
Value-trap signals · 4
  • Revenue still below 2021 peak; multi-year decline before recent recovery
  • Ongoing ~$5-15m/qtr capex on Chinese headquarters construction (~$270m committed) while buying back stock
  • China/Cayman Islands ADR structure carries geopolitical overhang
  • Gross margin compressing as lower-margin conversational-AI mix grows

Agora, Inc. (AGORA / NASDAQ: API) — Research Note

Executive summary

Agora is a real-time engagement (RTE) Platform-as-a-Service that provides APIs for embedding voice, video, chat, live streaming and — increasingly — conversational voice-AI agents into other developers' applications, split across a US/International "Agora" division and a Shanghai-based "Shengwang" division. After a brutal 2021-2024 (revenue peaked in 2021 at $168m, fell to $133m in 2024, cumulative net losses of ~$285m), the business has now printed seven consecutive quarters of GAAP net income, reaccelerated to +18% YoY revenue growth in Q2 2026, and is landing early commercial deployments of its Voice AI Agents in call centers. The single most important point for valuation: the company's ~$362m of cash, deposits and bank products (as of Q2 2026) is essentially equal to its entire $343m market cap — you are getting a growing, profitable, AI-adjacent software business for free, offset by real China/geopolitical risk and a large ongoing headquarters construction project consuming that cash cushion.

Fair value estimate

  • Fair value range: $5.50 – $8.50 per ADS, implying a market cap of ~$465m – $720m.
  • Methodology: sum-of-parts / cash-adjusted revenue multiple. Cash + bank deposits + financial products of $361.7m less long-term construction borrowings of ~$99m = ~$263m net financial assets (about $3.11/ADS). At Q3 2026 guidance run-rate revenue is $164m annualised. Applying a 1.5x – 3.0x EV/sales multiple to the operating business (justified by high gross margin, 15-18% growth, recent profitability, but discounted for competitive intensity and China exposure) gives an operating-business EV of **$245m – $490m**; adding net financial assets gives $508m – $753m market cap = $6.00 – $8.90 per ADS. Anchoring conservatively at the lower end for China risk: $5.50 – $8.50.
  • Current market cap: $342.8m ($4.05/ADS).
  • Absolute upside/downside: mid-point $7.00 vs. $4.05 = ~+73% upside.

Sector context

  • Sector: Software and IT Services (Technology). Correctly classified.
  • Agora's growth (13-18% YoY) is now above the median for legacy CPaaS peers but its scale is a fraction of theirs. Gross margin (63-66%) is below best-in-class SaaS (75%+) because of bandwidth costs. Balance sheet quality is well above peer average — net cash vs. typical peers with meaningful net debt.
  • Listed peers: Twilio (TWLO), Vonage (private after Ericsson deal), Zoom Communications (ZM) for videoconferencing-adjacent; more closely, the conversational-AI angle puts it alongside SoundHound AI (SOUN), LivePerson (LPSN), and infrastructure vendors like Nice (NICE).

Investment thesis (3 bullets)

  1. Genuine AI-receiver at the infrastructure layer, not an AI spender. Agora sells the real-time voice/video pipes that make LLM-based agents actually usable — sub-second latency, noise suppression, interruption handling, global SD-RTN network. Management is now landing paying Voice AI Agent deployments in outbound marketing, market surveying and customer service, with pricing "an order of magnitude higher than our standard RTC audio product" 2025-06 Q2 prepared remarks, and Convo AI Engine usage more than doubled sequentially every quarter since March 2025 launch 2025-12 Q4 prepared remarks.
  2. Trading essentially at cash with reaccelerating growth and seven straight profitable quarters. $361.7m cash/deposits vs. $342.8m market cap — the operating business has a negative implied enterprise value even before netting the ~$99m HQ borrowings 2026-06 Q2 earnings release. Growth has re-accelerated from +3.6% (Q4 2024) → +12.1% (Q1 2025) → +18.0% (Q2 2026), and Q3 2026 guidance implies another 16-19% growth quarter 2026-06 Q2 earnings release.
  3. Aggressive buybacks + operating leverage. Company has repurchased 44.6m ADSs for ~$160m since Feb 2022, shrinking share count from 112.4m to 83.8m (–26%) 2026-06 Q2 earnings release. With a largely fixed ~$27m/quarter opex base, gross margin of 64%, and DBNR back above 100% (104% in Q2 2026), incremental revenue drops disproportionately to profit — CFO's explicit goal is quarterly GAAP operating profitability by end-2026 2026-06 Q2 prepared remarks.

Key risks (3 bullets)

  1. China / VIE / US-China dual-listing risk. Shengwang (~46% of 2025 revenue) is fully China-based, HQ in Shanghai, exposed to Chinese regulation, macro slowdown, and PCAOB / delisting overhang typical of US-listed Chinese ADRs — the "Cayman Islands holding company of two independent divisions" structure means this is legally a Cayman ADS, and the 2021 K12 tutoring regulation already destroyed a large chunk of Shengwang revenue 2024-12 Q4 earnings release.
  2. Headquarters construction is a major cash drain masking otherwise-clean balance sheet. Construction in progress is now $104.5m plus $165m of land use right, funded by $99m of long-term borrowings; another ~$5-15m/quarter of construction capex is being spent 2026-06 Q2 earnings release. If Chinese property/economic conditions worsen, this asset could impair or divert capital that would otherwise return to shareholders.
  3. Competitive intensity and small scale. Competes against Twilio (multi-billion-$ scale), hyperscalers' native voice AI stacks (OpenAI Realtime, Google Gemini Live), and open-source frameworks — including Agora's own sponsored TEN Framework. Gross margin has actually compressed from 68% (Q1 2025) to 63.7% (Q2 2026) as conversational AI products are "at a sub-scale stage" with lower margins 2026-06 Q2 earnings release; if scale doesn't improve margins, the AI thesis is weaker than it looks.

Operating leverage

Agora exhibits classic high-fixed-cost software operating leverage. Total operating expenses have been remarkably flat at ~$26-27m per quarter for the last six quarters (R&D $14-15m, S&M $6-7m, G&A $5-6m) while revenue has grown from $33.3m (Q1 2025) to $40.4m (Q2 2026), a 21% increase 2026-06 Q2 earnings release. Gross margins are ~64%. This means each incremental $1m of revenue drops ~$640k to operating profit before any variable opex. If Q3 2026 guidance of $41-42m holds and opex stays disciplined, operating loss narrows to near-zero; a 10-15% revenue upside surprise (say $46m quarterly) would generate ~$29m gross profit against ~$27m opex, flipping to $2-3m of operating profit vs. today's ~$1m loss — a swing of multiples of profit on modest revenue upside. Voice AI Agent deployments could accelerate this materially given their higher price point. The offset: bandwidth/server costs (COGS) do scale with usage, and management has flagged that conversational AI products currently run at below-corporate gross margin.

Value-trap signals

  • Multi-year revenue decline from $168m (2021) → $133m (2024) driven by China regulatory reset — only partially recovered, still below 2021 peak.
  • Ongoing heavy capex on the Chinese headquarters project ($270m+ committed in land + construction) at a time when the company is buying back stock — capital-allocation coherence is questionable.
  • US-listed China ADR structure carries inherent overhang; peers (DIDI, others) have shown how quickly this can go wrong.
  • Gross margin compression as core RTE mix shifts toward lower-margin conversational-AI products.

Earnings vs. expectations

Recent track record is a clear beat-and-raise pattern. Q4 2025 revenue of $38.2m beat guidance high-end of $37-38m; Q1 2026 revenue of $37.7m beat guidance high-end of $36-37m; Q2 2026 revenue of $40.4m beat guidance high-end of $39-40m. Net income has also grown sequentially every quarter for the last six quarters. Prior to 2025 the pattern was more mixed — significant guidance disappointments in 2022-2024 as China regulatory changes and K12 tutoring collapse hit. Summary: 5-6 consecutive beats with reacceleration; the recent operating cadence is credible, but the 2022-2024 misses show the business is vulnerable to macro/regulatory shocks it cannot control.

Conviction

Conviction: 3 (moderate).

Anchors: (1) balance sheet is clean and unambiguous — cash exceeds market cap and the company is buying back ~$15-20m of stock per quarter; (2) revenue trajectory is well-disclosed with clear sequential acceleration and management guidance being consistently beaten; (3) multiple valuation approaches (EV/Sales, sum-of-parts, cash-per-share floor) converge on a range meaningfully above $4.05.

Limits: (1) the biggest variable — whether the Chinese subsidiary Shengwang can be freely valued alongside the US business — is genuinely uncertain and could compress fair value by 30-40% if a discount for China/ADR risk is warranted; (2) the conversational AI revenue ramp is real but early — management gives no specific dollar disclosure of Voice AI Agent revenue, so the AI-beneficiary thesis is more narrative than measured today.

Filings consulted · 29

Every document the LLM read for this note. Click any row to open the source.

  1. 2026-06-30Prepared Remarks2026-06-30_820_prepared-remarks.md0.90
  2. 2026-06-30Investor Presentation2026-06-30_827_investor-presentation.md0.70
  3. 2026-06-30Earnings Release2026-06-30_828_earnings-release.md0.95
  4. 2026-03-31Prepared Remarks2026-03-31_810_prepared-remarks.md0.90
  5. 2026-03-31Earnings Release2026-03-31_818_earnings-release.md0.95
  6. 2025-12-31Prepared Remarks2025-12-31_740_prepared-remarks.md0.77
  7. 2025-12-31Earnings Release2025-12-31_748_earnings-release.md0.81
  8. 2025-09-30Prepared Remarks2025-09-30_730_prepared-remarks.md0.77
  9. 2025-09-30Earnings Release2025-09-30_738_earnings-release.md0.81
  10. 2025-06-30Prepared Remarks2025-06-30_720_prepared-remarks.md0.58
  11. 2025-06-30Earnings Release2025-06-30_728_earnings-release.md0.62
  12. 2025-03-31Prepared Remarks2025-03-31_710_prepared-remarks.md0.58
  13. 2025-03-31Earnings Release2025-03-31_718_earnings-release.md0.62
  14. 2024-12-31Prepared Remarks2024-12-31_640_prepared-remarks.md0.58
  15. 2024-12-31Earnings Release2024-12-31_648_earnings-release.md0.62
  16. 2024-09-30Prepared Remarks2024-09-30_630_prepared-remarks.md0.58
  17. 2024-09-30Earnings Release2024-09-30_638_earnings-release.md0.62
  18. 2024-06-30Earnings Release2024-06-30_628_earnings-release.md0.43
  19. 2024-03-31Earnings Release2024-03-31_618_earnings-release.md0.43
  20. 2023-12-31Earnings Release2023-12-31_548_earnings-release.md0.43
  21. 2023-09-30Earnings Release2023-09-30_538_earnings-release.md0.43
  22. 2023-06-30Earnings Release2023-06-30_528_earnings-release.md0.24
  23. 2023-03-31Earnings Release2023-03-31_518_earnings-release.md0.24
  24. 2022-12-31Earnings Release2022-12-31_448_earnings-release.md0.24
  25. 2022-09-30Earnings Release2022-09-30_438_earnings-release.md0.24
  26. 2022-06-30Earnings Release2022-06-30_428_earnings-release.md0.24
  27. 2022-03-31Earnings Release2022-03-31_418_earnings-release.md0.24
  28. 2021-12-31Earnings Release2021-12-31_348_earnings-release.md0.24
  29. 2021-09-30Earnings Release2021-09-30_338_earnings-release.md0.24

This research note was authored by a large language model after reading 24 regulatory filings published between 2021-09-30 and 2026-06-30. Each citation refers to a specific RNS announcement in the underlying data set. The note is an opinion, not advice. Do your own work before risking capital.