Radware Grew 12% and Reported Less — the Shekel Ate the Quarter

A record top line, a first Xploit Shield carrier win, and a CEO who dates the AI payoff to 2027

RDWR · Earnings Call · 2026-10-28

The quarter the currency ate

Radware's third quarter was, on the operating line, its best in years: record revenue of $84.4 million, up 12%, the eighth double-digit quarter in the last nine, with cloud ARR at $107 million (+21%) now 41% of a $266 million total ARR base that grew 11%. Then you read down the income statement and it looks like a stall — non-GAAP EPS of $0.32 versus $0.33 a year ago, net income down 6%. The gap between those two sentences is one number: the shekel. Guy Avidan put it plainly: “Excluding this currency impact, operating income would have been $16.5 million, representing 37% growth compared with the third quarter of 2025” — Guy Avidan, Chief Financial Officer · 2026-10-28. And on the bottom line, ex-FX non-GAAP EPS would have been $0.41, up 24%. So the honest description of Q3 2026 is a 37%-growth operating quarter wearing a 6%-decline costume. What makes this a story rather than an excuse is the keyword history. Israeli Shekel has been in Radware's transcript vocabulary for at least three quarters — it ranked high in the spring — yet the dollar damage has grown even as its share of the conversation has quietly slipped. Management is now absorbing a bigger hit with less airtime. The same wedge shows up in gross margin: 81.4% versus 82.4%, blamed on FX, product mix and the supply-chain cost pressure that traces back to the memory inflation Radware was pricing around earlier this year, when Roy Zisapel noted they “did increase price lists by 5% to 8%” — Roy Zisapel, President and Chief Executive Officer · 2026-05-07 on memory-exposed hardware. Underneath that, the balance sheet keeps doing work: $421.5 million of cash and securities against a ~$1.56 billion market cap, with $62 million of buybacks year-to-date. That capital return is now visibly self-taxing — financial income fell to $4.5 million from $5.2 million on lower rates and a smaller cash pile, a modest headwind management says persists into year-end.

From AI narrative to AI product names

The most interesting evolution in the transcript is linguistic. For roughly six quarters, Radware's AI keywords were conceptual — agentic AI, AI security, "Fight AI with AI." This quarter's top of the keyword table is suddenly product: security platform, Xploit Shield, vulnerability discovery, API protection. That is what a slide turning into a pipeline looks like in the data. The proof point: Xploit Shield, launched last quarter as payload runtime protection for freshly discovered exploits, produced its first large win with a leading Asian telecom carrier. “The carrier selected Radware as its primary cloud application security platform following a competitive evaluation with Xploit Shield serving as a key differentiator” — Roy Zisapel, President and Chief Executive Officer · 2026-10-28. Layered on top are MCP Server Protection (with production customers) and Agent Trust Management — the emerging question of whether an AI agent asking for access is who it claims to be. Here is where the contrast beats the consensus. The narrative is hot; the revenue is not yet. Zisapel refused to sell the timeline:

I do think it is a 2027 opportunity, but we are definitely seeing a rise there.
Roy Zisapel, President and Chief Executive Officer · 2026-10-28

That is an unusually disciplined answer for a company whose stock story now rests on Agentic AI. It also echoes what he said three months ago about AI security generally — “It is very early... sales cycles in enterprise are 6 to 9 months on the short side” — Roy Zisapel, President and Chief Executive Officer · 2026-05-07 — while noting API security, by contrast, is immediate because it is one more module on the same platform to the same buyer. API and WAF consolidation, not agentic security, is what is actually in the ARR today.

The unglamorous engine carrying the next four quarters

The refresh cycle is the part of this call that deserves more attention than it got. DefensePro X upgrades of end-of-sale DDoS estates: “With less than one-third of the end-of-sale install base refreshed to date and a strong pipeline ahead, we see meaningful opportunity from this cycle over the next several quarters” — Roy Zisapel, President and Chief Executive Officer · 2026-10-28. Compare that to February, when the framing was “we haven't crossed half of the way. So we still have a long runway here in terms of refresh” — Guy Avidan, Chief Financial Officer · 2026-02-11. Both statements are true, but together they say something uncomfortable: after four or five quarters of pitching a refresh cycle, only a third of the base has moved. That is either a runway with years left in it or a conversion problem wearing a runway's clothes — and the answer differentiates Radware from every AI-security peer story in the market right now. The counterweight arrived yesterday: an agreement with a major global cloud provider to modernize and expand its DDoS protection infrastructure after a competitive evaluation, exactly the kind of CapEx-flavored deal that sits outside recurring ARR. The reason this business can keep winning price is structural, and Zisapel was happy to spell it out — own mitigators, own silicon, AppSec POPs opened like storefronts. “It's very capital efficient, what we're doing” — Roy Zisapel, President and Chief Executive Officer · 2026-10-28. Avidan added the tell: free cash flow of $12.4 million sat close to operating cash flow of $15.5 million despite CapEx tripling year-over-year to $3.1 million. A DDoS network that consumes capex the way a software company consumes servers is a genuine competitive moat.

What the room and the tape are actually saying

Notice what the analysts asked. Not one question was about the size of the agentic AI TAM. Joe Gallo asked whether the AI products land in Q4 or next year; George Notter asked about returns on the sales-and-marketing build and about scrubbing-center capex; Jeff Hopson asked whether AI is a genuinely different sale. Zisapel conceded the fragmentation — “There's plenty of startups, I think over 20, and of course, some of the large players, so there's still a lot of noise” — Roy Zisapel, President and Chief Executive Officer · 2026-10-28. The sell side is underwriting execution and expense discipline, not the AI option. The broader tape agrees with that caution, if indirectly. Global keyword flow for the quarter is bank-heavy (IRB models, core deposits, Fed hike language), while the price tape shows the AI-infrastructure cluster being sold — AI data center and data center capacity names down mid-to-high single digits over 30 and 90 days — as the market rotates toward energy, biotech and rate-sensitive financials. Radware is not selling compute; it is selling defense against the consequences of compute, which is a better place to sit in a de-rating hardware tape, but it also means it cannot rent a sector multiple. And the ambient theme the market keeps flagging, AI risk, is precisely Radware's demand driver — attacks that arrive faster because AI shortens the path from disclosure to exploit, the PHP2Shell and Node2Shell pattern Zisapel used as his hook. No price tape was available for RDWR, so we cannot say how the market scored a 37%-growth quarter dressed as a 6% decline. That omission is itself the open question: the operating story accelerated, the reported story did not, and the largest new growth vector has been given a 2027 delivery date by its own CEO. What Radware bought this quarter was time — and with $421 million of cash and a buyback at 4% of the float, it can afford to spend it.