Patrick Colville

Leighton said the CrowdStrike win closed recently and validates Akamai's security and reliability; CrowdStrike was dissatisfied with its prior provider, and the two also partner.

Tom couldn't discuss the specific deal but emphasized Akamai's distributed architecture as the core value proposition. Use cases involving video processing, generation, and agentic AI need scale and low latency that only distributed infrastructure provides. Ed explained the potential additional CapEx is tied to the broader GPU pipeline exceeding current inventory, not the $1.8B deal specifically. Wants another quarter before committing to additional orders.

Tom explained edge inference is different from edge CPU because AI applications are bandwidth-intensive. Generative media, personalized video processing, and live sports require distributed infrastructure, core data centers lack the bandwidth capacity for millions of concurrent personalized streams. Speech/avatar applications need real-time responses that only proximity enables. New GPUs are now fast enough (few tens of milliseconds) to make distributed inference viable.

Tom said no special software changes needed for GPUs, the cloud platform works as Linode always has. Customer pool is broader than traditional Akamai base, including companies without significant delivery or web app firewall needs. New customers signing who never used Akamai before. Selling both the traditional VM-hour/token model and the new cluster model with long-term contracts.

Tom said the partnership is much deeper with Blackwell 6000 representing a huge capability leap over prior 4000 series. Much broader deployment and a world of difference that enables Inference Cloud.

Ed cautioned against reading negative delivery signal from the wide guidance range. Q4 delivery is hard to call as seasonal factors do not manifest until Thanksgiving. The slightly below 15% compute timing bodes well for next year.

Ed said Edgio had no security or compute products. First priority is completing customer migration and onboarding. Expects 6 to 9 months to build pipeline before seeing potential upsell.

Ed confirmed government business is not a huge portion and they haven't modeled anything material from cost-cutting efforts. Not anticipating significant impact.

Ed explained cloud infrastructure services will accelerate but legacy applications (net storage, transcoding, image/video management) are being migrated to ISV partners, causing near-term flattening.

Ken emphasized disciplined M&A focused on technology and talent rather than market share. Acknowledged more reasonable multiples create more opportunities, especially in security operations. Maintained commitment to Rule of 45 and healthy margins as constraints.

Christiane confirmed 5-20% target depending on the appliance. Ken added Fortinet can adjust prices monthly with 30-day distributor notification and maintains inventory buffer for flexibility.

Christiane positioned upgrade activity as one of multiple growth drivers, not the primary one. John Whittle noted the normal 4-year refresh cycle aligns well with new ASIC and OS technology coming next year.

Ken positioned SASE firewall as the next growth driver beyond refresh. Christiane added COVID-era devices (5-7 years out from end of support) will create additional future tailwinds. John Whittle highlighted cloud services as incremental TAM.

Keith confirmed the leap year and Q4-to-Q1 day count effect was discussed in 2021 as well. Also noted caution in guidance-setting while the upgrade cycle signals are night and day better than 6 months ago.

George pointed to digital natives, AI-native companies and frontier labs turning to CrowdStrike, citing NVIDIA integration spanning chips to applications and AI inference, and called the AI ecosystem opportunity large.

Kurtz indicated discounting is in line with normal enterprise software sales, pointed to 81% gross margin as evidence of price discipline while maintaining competitive share gains.

Podbere explained margin headwinds from CCP amortization, marketing reignition, platform resiliency investments, and AI tooling, but guided to margin expansion in the second half.

George explained Charlotte AI was designed from the start as an autonomous agent, citing a customer reducing situational report creation from four days to one hour.

Olivier said it is both traditional observability and newer GPU monitoring. Hyperscalers use a broad product set but are particularly interested in GPU monitoring for training workloads, seen as a bellwether for the next wave of training adopters.

David confirmed they apply higher conservatism to the largest customer while noting no change in overall methodology. Olivier gently cautioned against overweighting specific word choices in the prepared remarks.

Pomel argued competitive landscape is unchanged, with Datadog winning through integrated platform breadth, higher R&D investment capacity, faster adjacency coverage, and an economic model competitors cannot match.

Prince said Cloudflare won't sell commodity compute or lend its balance sheet for GPU buildouts, focusing on extracting up to 10x more utilization per CapEx dollar.

Prince contrasted Cloudflare's organic R&D approach with Palo Alto's acquisition strategy. Agents already flow through Cloudflare's network; guardrails and rule enforcement already in market.

Prince explained AI companies start with security (fraud protection on expensive queries) then move to inference. Not yet the right place for massive LLMs but investing in larger model support.

CJ confirmed frontier labs is plural and deliberate but declined specifics; use cases vary by lab and remain early.

Dev argued Postgres is only an OLTP database - a false comparison to MongoDB's document plus search plus vector stack - and a weaker OLTP engine, citing very high win rates against Postgres.