Why Amazon might pay billions for Decart, a 3-year-old AI startup that hasn't disclosed its revenue

Amazon is reportedly in acquisition talks with AI software startup Decart at a valuation of about $7 billion, though neither company has confirmed the discussions. Decart’s software helps AI models run faster across chips and makes switching between them easier, which could support Amazon’s effort to expand adoption of its Trainium chips and reduce reliance on Nvidia hardware. Decart was valued at about $4 billion in a $300 million funding round in May; reported prior talks with Anthropic discussed a roughly $6 billion price.

Analysis

The strategic value is not Decart’s standalone revenue so much as lowering the engineering and utilization barriers that keep workloads on Nvidia. If its tooling makes migration repeatable, AWS could improve Trainium adoption and cloud retention; the second-order threat to Nvidia is gradual pressure on software-driven pricing power, not an immediate loss of chip demand. Microsoft’s Maia push and Google’s TPUs reinforce that direction, while Broadcom may benefit if custom-chip programs expand.

But the asset’s value depends on remaining credible across chip platforms. Restricting Nvidia or Google support after an Amazon acquisition could weaken Decart’s developer appeal; preserving neutrality may limit how much advantage AWS captures. The reported $7B discussion is especially hard to underwrite given the reported $4B funding valuation and a prior Anthropic process that ended after diligence. Those are not proof of overpayment, but they raise execution and valuation risk. AWS’s reported performance gains are company claims, not yet evidence of broad, independently verified workload migration.

Near term, this is a rumor-driven catalyst with binary deal risk and little basis for a directional position in NVDA. Over 1–3 months, verify terms, retention of cross-chip support, and customer conversion evidence. Over 6–18 months, the key test is whether Trainium wins recurring production workloads—not demos—without AWS sacrificing Decart’s ecosystem reach. The thesis weakens if talks fail, or if AWS adoption and utilization do not improve despite the tooling.

Key Decisions for Investors

• AMZN: Treat any pullback on acquisition-price concerns as a watch opportunity, not an automatic buy. Consider a small relative long versus NVDA only after a signed deal and evidence of production workload migration; the trade thesis is invalidated if AWS does not show improved Trainium adoption or utilization over subsequent reporting periods.
• NVDA: Do not short solely on the rumor. Monitor whether Decart remains available across Nvidia, Google, and AWS after any transaction; broad support would validate the software portability threat, while AWS-only alignment would reduce Decart’s ecosystem value and limit the read-through.
• Before assigning strategic value, verify final consideration and deal structure, Decart’s revenue and profitability, independent benchmarks versus comparable GPU workloads, and customer retention—especially whether existing non-AWS users can continue using the software.
• If talks collapse or the price approaches the reported valuation without verifiable commercial traction, avoid buying the acquisition narrative; the key downside catalyst is evidence that the asset’s demonstrated performance does not translate into durable cloud workload share.