Amazon's $25B Bond, SambaNova's $11B Raise, Meta's Iris Chip

A busy Wednesday and Thursday produced three distinct storylines for AI investors: Amazon going deep into debt markets to fund its buildout, a billion-dollar vote of confidence for an inference-focused chip startup, and Meta moving its custom silicon from announcement to production schedule.

Key points

• Amazon priced a $25 billion multi-tranche bond sale to finance AI infrastructure, contrasting sharply with Microsoft’s approach of leveraging its OpenAI stake and contracted backlog.
• Inference chip startup SambaNova raised $1 billion in a Series F led by General Atlantic, valuing the company at $11 billion post-money, with T. Rowe Price, Capital Group, and Seligman Ventures also participating.
• Meta plans to put its custom “Iris” AI chip into production in September, giving the company its own silicon for platform compute.
• SpaceX joined the Nasdaq-100, with analysts estimating the rebalance could force roughly $4.3 billion in passive buying from index funds.
• A looming US semiconductor worker shortage could leave a gap of 157,000 skilled workers by 2030, threatening TSMC, Samsung, and other fab expansion plans.

What does Amazon’s bond sale tell investors about the AI race?

Amazon chose debt, not equity, to fund its next infrastructure wave. The $25 billion bond tranche signals that Amazon sees the AWS acceleration as durable enough to service substantial new obligations. The company’s strategy leans on custom silicon rather than third-party model partnerships. Microsoft, which filed earnings the same day on April 29, 2026, is taking the opposite route: monetizing its OpenAI relationship and leaning into a contracted backlog rather than a hardware-heavy balance sheet.

For investors, the divergence matters. Amazon is making a capital-intensive bet that owning the infrastructure layer, including its own chips, produces better long-run margins than licensing or partnering. Microsoft is betting that a sticky enterprise base and AI software services compound faster than raw compute spending. Both are playing the same tailwind. The balance sheet consequences, and the risks attached, are very different.

Is the SambaNova raise a signal about where inference spending is heading?

SambaNova’s $1 billion Series F at an $11 billion valuation is the largest private AI chip raise in recent memory from a company that doesn’t make training chips. Its entire product line, custom chips, hardware systems, and cloud services, is aimed at inference: the moment a deployed model actually responds to a user query.

That focus is increasingly where enterprise AI spending concentrates. Training happens once or occasionally; inference happens billions of times per day at scale. General Atlantic led the round, with institutional backing from T. Rowe Price Associates and Capital Group suggesting this is not purely venture-stage risk appetite. For public market investors, the round is worth tracking as a pricing benchmark. If a private inference chip specialist commands $11 billion, it clarifies where the market sees durable value in the AI stack beyond Nvidia.

Meta’s Iris chip: what September production means

Meta’s plan to move its “Iris” custom AI chip into production in September puts it on a timeline similar to other hyperscalers who have spent years reducing their dependence on merchant silicon. Google has TPUs, Amazon has Trainium and Inferentia, and Microsoft has Maia. Meta joining that group with Iris in volume production this year changes its compute cost structure over time.

The immediate read-through for Nvidia is limited; hyperscaler custom chips tend to handle specific workloads while Nvidia GPUs remain the workhorse for training and general inference. The longer-term implication is that each custom chip generation chips away at the total addressable market Nvidia would otherwise own.

Broader signals: SpaceX index mechanics and the worker shortage

The SpaceX Nasdaq-100 inclusion is primarily a passive-flow story. Index funds that track the Nasdaq-100 have no discretion; they must buy SPCX at the rebalance weight. CNBC’s Leslie Picker put the forced buying figure at approximately $4.3 billion. Baron’s RONB ETF already holds a 19% position in SpaceX as a core active holding, so active managers had a head start. The index inclusion now drags passive capital in behind them.

Separately, a report on the US semiconductor worker shortage puts a structural ceiling on how fast domestic fab capacity can actually be built. A projected gap of 157,000 workers by 2030 spans manufacturing and engineering roles, and without sustained government and education investment, the federal funding flowing to companies like TSMC and Samsung faces real execution risk. That is a macro constraint worth watching as investors price the semiconductor capex cycle.

Nothing on this site constitutes investment advice. All coverage is independent and informational only.