TMTPOST --   Broadcom Inc. is further   challenging Nvidia   Corporation   in artificial intelligence ( AI )   arms race.

AI Generated Image

Broadcom   launch   a new Nvidia rivaling networking chip   called Thor Ultra, enables computing infrastructure operators to deploy far more chips than they otherwise could.   Networking chips help data center operators move information around inside a facility.

Thor Ultra   is the industry ’ s first 800G AI Ethernet   Network Interface Card ( NIC ) , capable of interconnecting hundreds of thousands of XPUs to drive trillion-parameter AI workloads. By adopting the open Ultra Ethernet Consortium ( UEC ) specification, Thor Ultra gives customers the ability to scale AI workloads with unparalleled performance, and efficiency in an open ecosystem, said   Broadcom.   XPUs are custom-designed   high-performance processing units   for AI datacenters.

"Thor Ultra delivers on the vision of   Ultra Ethernet Consortium   for modernizing RDMA   for large AI clusters," said   Ram Velaga, senior vice president and general manager of the   Core Switching Group   at Broadcom. "Designed from the ground up,   Thor Ultra   is the industry ’ s first 800G Ethernet NIC and is fully feature compliant with UEC specification."

Thor Ultra supports data speeds of 800 gigabits per second, allowing   computing infrastructure operators to build and run the large models used to power AI applications such as ChatGPT.

Thor   Ultra outperforms   traditional RDMA, or remote   direct memory access,   as it introduced   a groundbreaking suite of UEC-compliant, advanced RDMA innovations including:   packet-level multipathing for efficient load balancing,   out-of-order packet delivery directly to XPU memory for maximizing fabric utilization, selective retransmission for efficient data transfer,   and programmable Receiver-based and Sender-based congestion control algorithms.

The new chip   came   a day after Broadcom   renewed the market speculation   about its   big customer.

OpenAI is not the mystery $10 billion customer that it announced during its earnings call last month,   Charlie Kawwas, president of the semiconductor solutions group at Broadcom, said in a CNBC interview on Monday.

Broadcom and OpenAI on Monday announced they   agreed   on a collaboration for   10 gigawatts ( GWs )   of custom AI chips   and computing systems over the next four years. OpenAI will design the its own chips,    which   will allow it to embed   what it has learned from developing   AI   models directly   into the hardware that underpins future systems. Broadcom will help develop    and deploy   the systems, which are built on its Ethernet   and other connectivity solutions.  

Kawwas   in   the interview   discussed with   OpenAI ’ s president, Greg Brockman   about   two companies ’   new deal, and revealed OpenAI   is   not the customer which brought   Broadcom   a massive order worth of $10 billion.    "I would love to take a $10 billion [ purchase order ] from my good friend Greg," Kawwas said. "He has not given me that PO yet."  

Kawwas didn ’ t   specify who is scooping up $10   billion worth of Broadcom ’ s   products. The chipmaker   didn ’ t comment   on Kawwas ’   remarks.

Broadcom   had secured $10 billion in orders for its custom AI chips, which it calls XPUs, from a fourth customer, the CEO Hock Tan said at an   earnings   conference   on   September 4.   "One of these prospects released production orders to Broadcom, and we have accordingly characterized them as a qualified customer for XPUs," Tan said.   He added   the orders will ship next   year.

Given the major order   from the   new customer   who   has   "immediate and pretty substantial demand"   , Broadcom   upgraded   its forecast for AI revenue next year,   according to Tan.   He told   investors   the chip   designer ’ s outlook for fiscal year 2026   will improve   "significantly"   from   what   it   had   indicated last quarter,   and   a growth rate will accelerate   in a "fairly material"way. Tan   previously had said 2026 would   witness   the AI   revenue grow at the same rate   as the current year, namely, a   rate   between 50% and 60%.   

While Tan   didn ’ t namethe new customer, the Financial   Times   following Broadcom ’ s financial results   reported   that OpenAI was   the latest customer. The client   was identified as   OpenAI, Bloomberg echoed, citing   people with knowledge of the matter.   According to   the source, Broadcom ’ s   custom   chip program is intended   to ease supply constraints on   Nvidia   Corporatioin ’ s AI chips, and   to allows OpenAI to develop   processors tailored for   its models.