Akamai Technologies locked in a seven-year, $11.6 billion infrastructure agreement with Anthropic on September 24, 2026, and the ripple effects are still moving through edge cloud stocks nearly three weeks later. The deal, which could grow to as much as $20 billion if Anthropic exercises additional spending options, is the largest contract in Akamai’s history and a clear signal that AI labs are running out of GPU and CPU capacity fast enough to sign multi-year commitments with CDN companies that, until recently, were known mainly for caching web pages and blocking DDoS traffic.
The timing matters. Two weeks after the Akamai-Anthropic announcement, Fastly stock spiked as much as 17% on October 9 after Oppenheimer upgraded the stock to Outperform with a $35 price target, citing the exact same dynamic: AI agents and model providers need distributed compute closer to users, and traffic-metered edge vendors are positioned to capture that demand. Cloudflare and Datadog moved too, though by smaller amounts. Wall Street has apparently decided that edge infrastructure, not just hyperscaler data centers, is where the next leg of the AI capacity crunch plays out.
What the Akamai-Anthropic deal actually covers
According to Akamai’s official announcement, the agreement runs for seven years and commits Anthropic to $11.6 billion in contractual spending on Akamai Cloud infrastructure. The company behind Claude will tap Akamai’s distributed network, which spans thousands of points of presence, for what Akamai describes as compute running “from core to edge.” Importantly, the deal is built primarily around CPU capacity rather than GPU training clusters, which separates it from the headline-grabbing Nvidia GPU megadeals that have dominated AI infrastructure news for the past two years.
A separate report cited by Enterprise DNA put the potential ceiling at $20 billion if Anthropic exercises additional options tied to future spending tiers. To meet that demand, Akamai is expected to spend roughly $5.5 billion in capital expenditure of its own, and the company reportedly raised its planned 2026 capex by $1.7 billion, partly to lock in memory and other hardware components ahead of an industry-wide supply squeeze. That capex increase is notable on its own: it suggests Akamai is buying hardware now rather than waiting, a sign the company expects demand from Anthropic and other AI customers to keep climbing through 2027.
What the deal does not confirm is just as important as what it does. The available disclosures do not specify what percentage of Claude’s inference traffic will actually run at Akamai’s edge nodes versus centralized regions, and neither company has published a technical breakdown of latency targets or workload types. The $11.6 billion figure describes a capacity commitment, not a guaranteed deployment architecture, so some of the loudest “AI at the edge” framing in early coverage outran what Akamai and Anthropic have actually confirmed in writing.
Why Anthropic needs CPU capacity, not just GPUs
Most AI infrastructure headlines in 2025 and 2026 have centered on GPU scarcity: Nvidia’s B200 and Vera Rubin lineups, cloud rental prices climbing as high as $8 an hour, and chip export controls shaping who gets access to what hardware. The Akamai deal points to a quieter but equally real bottleneck: the CPU-side work that surrounds every model inference call. Request routing, authentication, API rate limiting, data preprocessing, caching, and the thousands of small orchestration tasks that sit between a user’s prompt and a GPU cluster all run on CPUs, and at Anthropic’s current scale, that overhead has become its own capacity problem.
Akamai’s pitch here is specific: its network already handles enormous volumes of CPU-bound traffic for its existing CDN and security customers, and extending that infrastructure to Anthropic’s orchestration layer lets Claude’s user-facing requests get processed closer to where users actually are, without competing for the same GPU racks that hyperscalers are rationing. It is a complementary deal to Anthropic’s existing GPU commitments with cloud providers, not a replacement for them, and that distinction is why several outlets initially mischaracterized the agreement as a GPU-capacity deal before the CPU framing became clear.
The Fastly rally: Oppenheimer’s $35 call
Fastly shares jumped sharply on October 9, with figures ranging from a 15% intraday move to a reported 17% spike depending on the measurement window, according to 24/7 Wall St. and a separate Yahoo Finance report. The catalyst was an Oppenheimer upgrade from Neutral to Outperform, with analyst Param Singh setting a $35 price target, roughly 38% above Fastly’s October 8 closing price.
Singh’s reasoning, as reported by Investing.com, rested on three pillars: industry checks pointing to larger contract sizes, stronger security cross-selling across Fastly’s product line, and early monetization of traffic generated by agentic AI applications. Because Fastly bills customers based on metered network traffic, rising request volume from AI agents translates more directly into revenue than it does for flat-rate competitors. Oppenheimer applied roughly 7x estimated 2027 enterprise value-to-sales to Fastly, a discount to the 10.6x peer median, implying the stock still had room to re-rate even after the jump.
Singh also projected that Fastly’s third-quarter revenue would land at the top of management’s $184 million to $190 million guidance range, and predicted the company would raise full-year 2026 revenue growth guidance to roughly 20% to 21%. Separately, Fastly’s AI Runtime, AI Firewall, and API Enforcement products reached general availability in September 2026, giving the company new, as-yet unpriced products that Oppenheimer flagged as a source of upside the market had not fully modeled.
Cloudflare and Datadog: smaller moves, same narrative
Cloudflare and Datadog both rose on the same trading day, though by much smaller amounts than Fastly. Reported figures vary by source: 24/7 Wall St. put Datadog’s gain at roughly 4% and Cloudflare’s at about 2%, while other coverage cited 4.18% and 3.51% respectively. The size of the discrepancy itself is a reminder that single-day percentage moves depend heavily on the exact timestamp used, and conservative readers should treat the smaller, more frequently cited figures as the safer reference point.
Oppenheimer’s commentary on Cloudflare was separate from the Fastly call but reinforced the same theme: the firm raised its Cloudflare price target to $430 from $380 while keeping an Outperform rating, ahead of Cloudflare’s expected October 29 earnings report. The analyst distinction drawn between the two companies is worth noting. Cloudflare was described as having less direct, metered exposure to AI-agent traffic growth than Fastly, because more of Cloudflare’s revenue comes from flat-rate plans and bundled security products rather than pay-as-you-go bandwidth. That is part of why Fastly’s percentage move dwarfed Cloudflare’s even though both companies compete in the same edge compute category.
Market impact: a repricing of edge infrastructure stocks
Put together, the Akamai-Anthropic deal and the Fastly, Cloudflare, and Datadog moves describe a market that is actively repricing edge and CDN infrastructure as AI-adjacent businesses rather than pure web-performance plays. For most of the last decade, CDN vendors competed on latency, uptime, and DDoS mitigation. The 2026 pitch is different: these companies own distributed points of presence that are difficult and capital-intensive to replicate, and AI labs that are already capacity-constrained on GPUs are now also competing for CPU, network, and colocation capacity that CDN operators have spent twenty years building out.
That repricing has a ceiling, though. Akamai’s own stock reaction to its own deal was not clearly reported in the same news cycle that sent Fastly up double digits, which suggests investors are still working out which company benefits most from AI-driven edge demand and by how much. A single $11.6 billion contract with one customer is a meaningful de-risking event for Akamai’s multi-year revenue visibility, but it is a different kind of catalyst than Fastly’s broader, traffic-metered exposure to AI agent growth across its entire customer base.
Competitive comparison: Akamai, Cloudflare, and Fastly’s different AI edge bets
The three companies are not pursuing identical strategies, and conflating them risks missing how differently each is positioned for AI-driven edge demand. Akamai’s approach centers on long-term, large-customer infrastructure contracts, exemplified by the Anthropic deal. Cloudflare has built out Workers AI and a broader developer platform, including recent launches like Basin for serverless analytics and K2 for event streaming, betting that developers will build AI-native applications directly on its edge network rather than renting raw capacity. Fastly, by contrast, is leaning on its traffic-based billing model and newly shipped AI Runtime, AI Firewall, and API Enforcement suite to capture revenue from the sheer volume of requests that AI agents generate as they browse, scrape, and call APIs on users’ behalf.
| Company | AI edge strategy | Revenue model | Recent catalyst |
|---|---|---|---|
| Akamai | Large-customer infrastructure contracts (core-to-edge CPU capacity) | Multi-year committed contracts | $11.6B Anthropic deal, Sept. 24, 2026 |
| Fastly | Metered traffic capture from agentic AI requests, new AI security products | Usage-based network billing | Oppenheimer upgrade to Outperform, $35 target, Oct. 9 |
| Cloudflare | Developer platform (Workers AI, Basin, K2) for AI-native apps | Mixed flat-rate and usage billing | Price target raised to $430 ahead of Oct. 29 earnings |
| Datadog | Observability and monitoring layered onto AI infrastructure growth | Subscription plus usage-based observability | Shared in the Oct. 9 sector rally (~4%) |
Historical context: how we got here
Edge compute and CDN companies have been angling for a bigger role in AI infrastructure for several years, mostly by cutting prices on object storage and compute to lure developers away from hyperscaler lock-in. Cloudflare’s long-running push against AWS S3 egress fees with R2, and its workers vs Lambda@Edge comparisons, were early examples of that strategy. What changed in 2026 is the customer type. Instead of courting individual developers and mid-size SaaS companies, CDN vendors are now negotiating directly with frontier AI labs that need enough committed capacity to plan infrastructure years in advance.
Anthropic itself has been on an infrastructure-buying spree throughout 2026, layering commitments across multiple cloud and compute providers rather than concentrating spend with a single hyperscaler. The Akamai deal fits that pattern: it diversifies Anthropic’s CPU and distributed-compute supply chain away from AWS, Google Cloud, and Microsoft Azure, giving the company a hedge against capacity shortfalls or pricing pressure from any single vendor. For an industry that watched GPU shortages dictate which labs could ship new models on schedule, locking in CPU and edge capacity years ahead of need is a lesson learned the hard way.
Data table: the numbers behind the deal and the rally
| Metric | Figure | Source |
|---|---|---|
| Akamai-Anthropic contract value | $11.6 billion over 7 years | Akamai press release |
| Potential maximum with options | Up to $20 billion | Enterprise DNA |
| Akamai’s expected capex to fulfill deal | ~$5.5 billion | Enterprise DNA |
| Akamai’s 2026 capex increase | +$1.7 billion | Enterprise DNA |
| Fastly stock move, Oct. 9 | 15%-17% (range by source) | 24/7 Wall St., Yahoo Finance |
| Oppenheimer Fastly price target | $35 (from prior rating) | Investing.com |
| Target premium vs Oct. 8 close | ~38% | Investing.com |
| Fastly Q3 2026 revenue guidance range | $184M-$190M | Investing.com |
| Projected FY2026 revenue growth guide | ~20%-21% | Investing.com |
| Cloudflare price target raise | $380 to $430 | Market reports |
Why the CPU framing changes how to read this deal
It is worth dwelling on the CPU versus GPU distinction because it changes who should pay attention to this story. GPU megadeals, like Amazon’s chip leaseback arrangements with Nvidia or the various multi-gigawatt data center buildouts announced across the industry this year, are primarily about training and running the largest models. CPU and edge infrastructure deals like Akamai’s are about serving those models efficiently once they are trained, at the scale Anthropic needs to run Claude for enterprise and consumer customers simultaneously, worldwide, with acceptable latency.
That makes the Akamai deal a leading indicator for a different kind of AI infrastructure spending than markets have mostly priced in. If Anthropic’s user base and API call volume keep growing, the orchestration layer, not just the model layer, becomes a cost center large enough to justify an $11.6 billion, seven-year commitment. Other AI labs serving comparable traffic volumes, including OpenAI and Google, are likely running into similar CPU and edge bottlenecks even if they have not announced comparably sized contracts publicly.
What this means for enterprise cloud buyers
For engineering teams evaluating where to run AI-adjacent workloads, the Akamai deal is a reminder that the edge-versus-cloud decision is no longer a simple cost trade-off. Companies building on top of Claude, or any model provider leaning more heavily on distributed infrastructure, may see latency and reliability improvements as that provider’s own edge footprint expands, without needing to change their own architecture. At the same time, enterprise buyers negotiating their own cloud contracts should expect vendors across the CDN and edge space to point to deals like this one as justification for price increases on committed-use contracts, since capacity that AI labs are reserving is capacity that is no longer available to everyone else at the old price.
The broader FinOps implication is that cloud cost forecasting now has to account for AI-provider-driven capacity competition even for teams that never touch a GPU directly. If Anthropic, OpenAI, and other labs are absorbing meaningful shares of CPU, memory, and edge point-of-presence capacity at the same vendors mid-size companies rely on for ordinary web traffic, pricing pressure can show up in contracts that have nothing to do with AI workloads at all.
Predictions: where this goes next
- Expect at least one more large AI lab to announce a comparable CPU or edge capacity deal with a CDN vendor (Cloudflare, Fastly, or a regional provider) within the next two quarters, as labs diversify away from hyperscaler-only infrastructure.
- Akamai’s Q4 2026 earnings call will likely include more detail on how much of the $11.6 billion has been recognized versus deferred, and analysts will press for a clearer edge-versus-core traffic split for the Anthropic workload.
- Fastly’s next earnings report, if it lands at or above the top of its $184M-$190M guidance range as Oppenheimer predicted, will trigger a second wave of analyst upgrades; a miss would likely erase a meaningful share of the October rally.
- Cloudflare’s October 29 earnings report becomes a referendum on whether its developer-platform strategy (Workers AI, Basin, K2) can capture AI-driven revenue as directly as Fastly’s metered-traffic model, given the $430 price target Oppenheimer has already set.
- Expect continued divergence between companies with usage-based billing and those with flat-rate plans, as AI agent traffic volume becomes a measurable, monetizable line item rather than background noise in network bills.
The open questions nobody has answered yet
Several details remain genuinely unconfirmed, and readers should treat any claim beyond them with caution. Neither Akamai nor Anthropic has published named executive quotes explaining the strategic rationale for the deal in enough detail to settle whether this is primarily a hedge against hyperscaler capacity constraints, a cost play, or a latency-driven product decision. No source has quantified what share of Claude’s live traffic will actually route through Akamai’s edge nodes versus centralized data centers, and no analyst note reviewed for this story has put a confirmed, same-day Akamai stock price reaction on record, which is unusual for a contract this size and suggests the market is still digesting what the deal means for Akamai’s own margins given the roughly $5.5 billion in capex required to deliver it.
Frequently asked questions
How much is the Akamai-Anthropic deal worth?
Akamai and Anthropic confirmed a $11.6 billion contract spanning seven years, announced September 24, 2026. A separate report indicates the total could reach $20 billion if Anthropic exercises additional spending options built into the agreement.
Is this deal about GPUs or CPUs?
The agreement is primarily built around CPU computing capacity to support Anthropic’s distributed infrastructure needs, not GPU training clusters. That distinguishes it from the GPU-focused megadeals that have dominated AI infrastructure headlines over the past two years.
Why did Fastly stock jump in October 2026?
Oppenheimer upgraded Fastly to Outperform with a $35 price target on October 9, 2026, citing larger contract sizes, stronger security cross-selling, and early monetization of AI-agent traffic through Fastly’s usage-based billing model. Shares rose between 15% and 17% depending on the measurement window.
Did Cloudflare and Datadog benefit from the same news?
Both stocks rose on October 9, 2026, though by smaller amounts than Fastly, with reported figures ranging from roughly 2% to 4% for Cloudflare and about 4% for Datadog depending on the source. Oppenheimer separately raised its Cloudflare price target from $380 to $430 ahead of the company’s October 29 earnings report.
How much will Akamai have to spend to deliver the deal?
Akamai is expected to invest approximately $5.5 billion in capital expenditure to build out the infrastructure needed to fulfill the Anthropic agreement. The company reportedly raised its total planned 2026 capex by $1.7 billion, partly to secure components such as memory ahead of expected shortages.
Does this mean Claude runs at the edge now?
Not entirely confirmed. Akamai describes its infrastructure as supporting compute “from core to edge,” but neither company has published a specific breakdown of what percentage of Claude’s inference or orchestration traffic actually runs on edge nodes versus centralized regions.
How does this compare to Cloudflare’s AI strategy?
Cloudflare has focused on building a developer platform, including Workers AI, Basin, and K2, aimed at letting developers build AI-native applications on its network. Akamai’s approach is centered on signing large, long-term infrastructure contracts directly with AI labs rather than courting individual developers.
What should enterprise cloud buyers take away from this?
Buyers negotiating cloud or CDN contracts should expect vendors to cite AI-driven capacity demand, including deals like Akamai’s, as justification for price increases on committed-use agreements. Capacity reserved for AI labs is capacity no longer available to other customers at prior pricing.
