Anthropic has reportedly signed a '7-year, $11.6 billion infrastructure service provision contract' with Akamai. According to the report, Akamai is expected to invest $5.5 billion for this purpose, of which $1.16 billion will be allocated to CPUs or DRAM.
If we arbitrarily interpret the context hidden behind this article, it's clear that Anthropic and OpenAI are not only concerned about external scrutiny regarding the cost of learning infrastructure investment but also feel the internal burden of raising funds.
As mentioned briefly last week, Anthropic seems to be addressing this issue by converting 'part' of its existing training infrastructure into inference (service) infrastructure and expanding facilities in anticipation of increasing traffic (revenue).
In other words, behind Anthropic's previous statement that 'it may need to slow down AI development for 1-2 years for security reasons,' there seems to be a stance suggesting the need for 'surplus cash flow and investment capital.
However, articles predicting stagnation in HBM demand or price declines due to these hyperscaler moves have not yet been confirmed. Instead, there are more predictions that the exclusive sales of GPUs and HBMs will spread to server CPUs and DRAM.
Of course, there are limits to the investment assets of hyperscalers, so as the service infrastructure actually expands, the revenue on one side will inevitably decrease.
▶ Original source: https://seekingalpha.com/news/4647055-akamais-deal-with-anthropic-should-boost-cpu-demand-gf-says
