Infrastructure

Anthropic Wants Its Own Chip, Not Just Nvidia's

The Information reports Anthropic is in early talks with Samsung to build a custom AI chip. The last big lab that only bought Nvidia is done doing that - here is what it means for the price of compute.

InfrastructureBy Servola Tech Desk2026-07-052 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Anthropic Wants Its Own Chip, Not Just Nvidia's

Every frontier lab now wants its own silicon

The Information reported on 2 July 2026 that Anthropic is in early talks with Samsung to manufacture a custom AI chip, reportedly eyeing Samsung's 2nm foundry process and advanced packaging. The talks are exploratory - Anthropic has not decided what the chip will do, how powerful it will be, or how it fits into a server - and Samsung already took part in Anthropic's Series H funding round as a strategic infrastructure partner.

The move completes a pattern rather than starting one. Google has run its own TPUs for years, Amazon has Trainium, and in late June OpenAI unveiled a Broadcom-designed inference chip called Jalapeno. Reuters first reported Anthropic weighing custom silicon back in April. With this, the last frontier lab that relied on merchant Nvidia hardware is now designing around it.

Why a model company builds a chip

The reasons are supply and control. Anthropic has cited chip shortages, and custom parts tuned for one workload can beat general-purpose hardware on performance-per-watt, which is the metric that decides a data centre's power bill. OpenAI made exactly that pitch for Jalapeno. Owning the design also loosens Nvidia's grip on both allocation and price.

This is addition, not replacement. Anthropic's own statement is that a diversified hardware stack including chips from Google, Amazon and Nvidia will remain pivotal to its compute strategy. A custom chip is a hedge against a single supplier, not a switch away from one, and the first working silicon is years out.

What it means for European operators

The near-term reading is boring and important: nothing gets cheaper soon. Custom chips take two to three years to reach production, so Nvidia's pricing power over your compute bill holds well into 2028. Anyone modelling AI costs on the hope that vertical integration relieves them this year is modelling the wrong year.

The strategic reading is the one to file. Compute is fragmenting into vendor-owned stacks where the company that makes the model also makes the chip, which quietly turns model choice into a silicon commitment. Europe, notably, has no lab in this silicon race at all - a reminder that the continent's compute dependence runs deeper than which cloud region you pick.

Frequently asked questions

What did Anthropic actually announce?

Nothing official. The Information reported that Anthropic is in early talks with Samsung to build a custom AI chip. Anthropic declined to comment on Samsung specifically and said a diversified stack of Google, Amazon and Nvidia chips stays pivotal to its strategy.

Does this cut AI compute costs soon?

No. Custom silicon takes roughly two to three years to reach production, so it does nothing to this year's or next year's bill. The safe planning assumption is that Nvidia keeps its pricing power through about 2028.

Should a European business change anything today?

There is no urgent action. The point to absorb is that model vendors are becoming chip designers, so a multi-model strategy now carries a hardware lock-in dimension worth factoring into longer-term compute planning.

The labs that trained on Nvidia are now designing around it. That frees no one from Nvidia this year, but it tells you the direction of travel: compute is becoming something the model vendors own end to end, and the buyers who plan for that now will not be the ones repricing their whole stack in 2028.

InfrastructureAnthropicSamsungCustom SiliconNvidiaCompute

Three Firms Control 90% of Memory. Now a Court Will Ask Why It Costs 700% More.

A US class action accuses Samsung, SK Hynix and Micron of coordinating the shift to AI memory while starving DDR4 and DDR5, driving prices up 700%. What it means for buyers.

4 min read4 views

Infrastructure

The 518 Billion Dollar Memory Fix Lands in 2028

South Korea unveiled a chip plan worth over 576 billion dollars, anchored by Samsung and SK Hynix fabs in Gwangju. For everyone buying memory, the key number is a date: no price relief before 2028.

3 min read5 views

Digital Sovereignty

Microsoft Raises 365 Prices as France and German States Walk Away

Microsoft 365 list prices rise on 1 July 2026 while France and German states move off Microsoft entirely. Why your renewal is now a sovereignty decision, and how to price the exit option.

3 min read14 views

← All articles