The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →The $38 billion figure was not an AWS investment in OpenAI. It was OpenAI’s seven-year commitment to buy AWS computing capacity, including access to large Nvidia GPU clusters for ChatGPT inference, future model training and AI agents. By August 2026, the relationship had expanded into a broader commercial partnership involving AWS Trainium, OpenAI products on Amazon Bedrock and a separately announced Amazon investment.
What the original $38 billion agreement covered
On November 3, 2025, Amazon Web Services and OpenAI announced a seven-year agreement under which OpenAI would purchase AWS cloud capacity for advanced AI workloads. OpenAI said it began using AWS compute immediately, with a larger capacity buildout targeted for completion before the end of 2026. The companies described the arrangement as capable of growing further over time. OpenAI’s announcement characterized the scale as hundreds of thousands of Nvidia GPUs and the ability to scale to tens of millions of CPUs.
The $38 billion is a multiyear commitment, not $38 billion of cash paid up front, AWS revenue recognized immediately, or profit. Public announcements do not establish how much capacity was already available, the detailed payment schedule, or the economics AWS will earn on it.
Why the Nvidia systems matter
AWS said the infrastructure would use Nvidia GB200 and GB300 systems connected through Amazon EC2 UltraServers. The point is not simply to assemble many GPUs: large AI workloads require accelerators to exchange data rapidly, and networking can constrain training or inference when computation is spread across many devices. UltraServer configurations are intended to provide tightly connected systems with low-latency communication.
#1 Best Overall
The GB200 and GB300 are Nvidia Grace Blackwell-generation platforms. Their inclusion indicates the kind of infrastructure planned, but GPU names and counts alone do not establish ChatGPT’s speed, capacity or model quality. Those outcomes also depend on networking, software, utilization, model design, storage, power and workload scheduling; the companies have not published an independent benchmark of OpenAI workloads on the systems.
AWS announced general availability of EC2 P6e-GB300 UltraServers on December 2, 2025, and said they offered 1.5 times the GPU memory and 1.5 times the FP4 compute of P6e-GB200 under its stated comparison. Those are AWS product claims, not measured results for OpenAI’s deployment. AWS’s announcement provides the product details.
Rank #2
- Ideal for Gifting
- Ideal for a bookworm
- Compact for travelling
How AWS compute could support ChatGPT
The November agreement covered several kinds of work, not a wholesale move of ChatGPT to AWS:
- Inference: computing needed to generate responses for ChatGPT users.
- Training: developing and training future OpenAI models.
- Agent workloads: running systems that use tools and carry out multistep tasks at scale.
OpenAI said the capacity would support these workloads; it did not say every ChatGPT request would run on AWS or that AWS would become its exclusive infrastructure provider. Nor did the agreement promise a particular consumer-facing improvement, such as faster responses, lower prices or a specific new feature.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWhat AWS CEO Matt Garman meant by “trust”
In an interview with CRN, AWS CEO Matt Garman described the deal as a “powerful reminder” of customer trust. His argument was that organizations choose AWS for the ability to deliver scale, security, performance and operational reliability, including when they need specialized infrastructure quickly. CRN’s interview reports his comments.
That is AWS’s positioning, not an independent finding that the contract proves every part of the claim. The deal does demonstrate that OpenAI selected AWS for a major capacity commitment. Whether the promised infrastructure arrives on schedule and performs reliably at OpenAI’s workloads is a separate operational question.
Rank #4
Why OpenAI added another major cloud partner
Frontier AI development depends on more than model ideas: it requires large and growing amounts of compute, suitable GPUs, networking, data-center power and the ability to bring capacity online. A second hyperscale provider can give OpenAI more options and reduce concentration in one infrastructure relationship. It does not, by itself, mean OpenAI is abandoning existing cloud arrangements.
The November deal also gave AWS a high-profile customer for its AI infrastructure and strengthened its position in competition with Azure and Google Cloud. But the public announcements do not disclose enough to determine how much new data-center capacity AWS must build, who carries particular power or utilization risks, or how the economics compare with OpenAI’s other infrastructure commitments.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Best Value
- It can be a gift option
- Comes with secure packaging
- Helpful in various ways
How the relationship expanded in 2026
A separate capacity expansion and Amazon investment
On February 27, 2026, OpenAI and Amazon announced a further partnership. OpenAI said it would expand its AWS capacity commitment by $100 billion over eight years and consume approximately 2 gigawatts of AWS Trainium capacity. This is a later expansion with its own term and structure; it should not be collapsed into the original seven-year $38 billion agreement as though the two figures represented a single upfront payment.
The February announcement also included a planned $50 billion Amazon investment in OpenAI: an initial $15 billion followed by $35 billion subject to conditions. That investment is distinct from OpenAI’s cloud-capacity purchases. AWS was also named the exclusive third-party cloud distribution provider for OpenAI Frontier. That exclusivity concerns Frontier distribution; it does not make AWS OpenAI’s exclusive cloud provider. The companies said they would jointly develop a Stateful Runtime Environment powered by OpenAI models. OpenAI’s February announcement describes the terms.
From infrastructure to enterprise products
By April through July 2026, the relationship had also become a route for enterprises to access OpenAI products through AWS. OpenAI described an April 28 launch that included OpenAI models, Codex and managed-agent capabilities on AWS; availability varied by service and model as the rollout progressed. By July, Amazon said GPT-5.6 Sol, Terra and Luna were generally available on Bedrock. OpenAI’s overview, Amazon’s Bedrock coverage and AWS’s GPT-5.6 announcement describe the evolving availability.
This adds a third layer to the relationship: alongside AWS infrastructure used by OpenAI, AWS customers can access OpenAI models and tools through AWS services. Amazon said pricing for GPT-5.5 and GPT-5.4 on Bedrock matched OpenAI’s first-party rates and involved no additional fees; later GPT-5.6 coverage also said pricing matched OpenAI rates. Those statements concern the named model offerings and do not establish that Bedrock is cheaper than direct OpenAI access.
What the deal could mean for customers and the cloud market
- For OpenAI: access to another major source of compute, initially including Nvidia systems and later a substantial Trainium commitment, plus an AWS channel for enterprise products.
- For AWS: a prominent AI customer, demand for large-scale infrastructure and a closer connection between Bedrock and OpenAI products.
- For enterprise buyers: a way to procure some OpenAI capabilities through AWS environments and controls. That may be convenient for organizations already using AWS, but it does not automatically make it the best route for every team.
- For Nvidia: the original agreement named GB200 and GB300 systems, while the later Trainium commitment shows the partnership is not Nvidia-only.
The main execution risks are familiar to large infrastructure projects: hardware supply, power availability, networking, data-center construction, deployment timing and whether capacity is well utilized. The end-of-2026 target for the initial buildout was a company target, not proof that all planned capacity had been delivered.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
