Buying Claude Direct from Anthropic or Through AWS Bedrock? Two Thai Market Cases With Diametrically Opposite Conclusions ซื้อ Claude ตรงจาก Anthropic หรือผ่าน AWS Bedrock ดี? สองกรณีจากตลาดไทยที่ให้คำตอบต่างกันสุดขั้ว

This question has been coming up more and more over the past year — from clients I advise, and from conversations in Thai tech circles: "If we're going to deploy Claude within our organisation, is it better to buy directly from Anthropic, or go through AWS Bedrock?"

The short answer is: it depends on who you are.

But the longer answer is far more interesting — because two cases that recently played out in the Thai market show how organisations with very different profiles can arrive at diametrically opposite conclusions, and both can be entirely correct for their own situations.

A large corporate conglomerate in Thailand recently signed with AWS Thailand to use Claude via Amazon Bedrock. The initial deployment focuses on Cowork and exploring other use cases across their operations.

A source on the procurement side said the deciding factor wasn't the model itself, but what came bundled with it — the AWS contract packaged training and technical support together as a single deal.

This is the point many people overlook. For an organisation with thousands of employees, training, onboarding, and ongoing support are not trivial costs if arranged separately — they're more expensive than people typically assume. Bundling everything into one contract genuinely reduces both procurement overhead and internal IT burden.

It's a familiar pattern in enterprise procurement: software price is not the total cost. The implementation costs invisible at the outset are what lead to disappointment later.

A Bangkok-based tech startup building an event management platform chose to purchase the Enterprise Plan directly from Anthropic via self-service, with no intermediary.

Management's logic was direct: the Enterprise Plan's $20 per person per month is a predictable fixed cost, whereas paying per token on Bedrock creates budget uncertainty.

For a team with stable headcount and fairly predictable AI usage, flat-rate billing delivers a certainty that consumption billing cannot.

I agree with their reasoning. Variance is risk. For a startup that needs to control its burn rate, a fixed cost is far easier to plan around.

Before continuing, there's something that needs to be stated clearly — because I see confusion about this regularly.

AWS Bedrock prices Claude models identically to the direct Anthropic API. There is no token discount for going through AWS.

Bedrock's value lies in ecosystem integration, bundled services, and existing commercial relationships — not cheaper tokens. If you go in expecting AWS to undercut Anthropic on pricing, you will be disappointed.

The startup in Case 2 already understood this — which is exactly why they had no compelling reason to accept the uncertainty of token billing.

To broaden the picture, here are some interesting numbers at the global level.

Today, over 100,000 customers run Claude on Amazon Bedrock, and Bedrock revenue grew more than 3× year-over-year per AWS's Q4 2025 earnings call, with the Claude family as the primary consumption driver.

Anthropic's revenue has reached a $30 billion run-rate, up from approximately $9 billion at the end of 2025, and approximately 70–75% of that revenue comes from API consumption — whether direct API, Bedrock, or Google Vertex AI.

One major development worth watching: Anthropic and AWS recently announced that the full Claude platform will be accessible directly through AWS accounts — using existing credentials, security controls, and billing, with no separate contract required. This means Cowork and Claude Code can run entirely within an organisation's existing AWS perimeter.

For organisations in regulated industries (finance, healthcare, government) that frequently run into data residency requirements — this changes the equation significantly.

Interesting international cases: Lyft used Claude via Bedrock to automate customer support, reducing resolution time by 87%. Pfizer used Claude to improve research workflows, saving scientists thousands of hours per year. Both already had AWS infrastructure in place — choosing Bedrock wasn't a new decision, it was an extension of existing systems.

This is what I believe matters most, and what tends to be overlooked in most analyses.

In the US or EU, large enterprises typically have internal AI teams capable of evaluating token economics themselves, and the leverage to negotiate with Anthropic directly at an account-led level — very different from self-service. Anthropic is currently signing multi-year contracts in banking, insurance, and pharma — with organisations that have the legal and compliance teams to negotiate SLAs directly with Anthropic.

Thailand's market isn't at that point for most organisations. That's not something to be embarrassed about — it simply means knowing where you stand.

The contexts where Thailand clearly differs:

AWS relationship comes first — Many large Thai organisations already have AWS contracts in place. If an Enterprise Discount Programme (EDP) exists, Bedrock usage may draw down against an existing commitment, which changes the economics significantly.

Internal IT capability still relies on SIs — Most large Thai organisations still depend on system integrators for infrastructure management. Bedrock's bundled training and support carry more value than the price figure alone suggests.

Still in the explore phase — Unlike Fortune 500 companies simultaneously piloting multiple AI models, most Thai organisations are still in the first phase. Having a vendor that provides guidance from the start carries significant weight in the decision.

AWS continues to invest in Thai infrastructure — The launch of the AWS Asia Pacific (Thailand) Region in 2025, and Claude's availability via Global Cross-Region Inference in Thailand, means Bedrock is no longer merely a theoretical option.

There is no universally correct answer. Both cases above reflect different profiles that arrived at the right conclusion for their own situations.

To summarise simply:

• Large enterprises already using AWS → evaluate Bedrock in the broader context of your AWS contract. The bundle value is real.
• SMEs, startups, self-sufficient teams → try Enterprise direct first. Fixed costs, and you can deploy far faster.

One factor people often leave out of the equation is time-to-value. A procurement process that takes three months — even if it secures marginally better token pricing — may be more expensive in terms of the productivity lost while waiting. The startup in Case 2 deployed within days. That value doesn't appear in any Excel spreadsheet.