AdvertiseFee Deal Desk

AMD

AMD Chips Take AI Inference Share as TensorWave ARR Hits $210 Million

AMD and Google AI chips are taking inference workloads from NVIDIA. OpenAI, Anthropic and Meta adopted AMD, pushing TensorWave's ARR toward $210 million.

Be a creator
November 18, 2026, 02:00 PM UTC4 min read
AI SummaryAI
  • AMD and Alphabet are taking AI inference workloads from NVIDIA, cloud operators report
  • OpenAI said on November 16 it will rent TensorWave's AMD-based cloud compute
  • Google disclosed in July up to $44 billion in guaranteed data center leases
  • NVIDIA's latest quarterly revenue exceeds AMD's by more than eight times

TensorWave Projects $210 Million ARR

AI inference workloads, the computing that runs trained models rather than builds them, are shifting away from NVIDIA (NVDA), and the beneficiaries are Advanced Micro Devices (AMD) and Alphabet (GOOGL). Cloud operators running this hardware describe the change in The Information's report on the chip market, and it now shows up in order books and revenue. Price is AMD's opening argument. Inference providers earn a margin on every job they serve, and cheaper accelerators widen it, a gap that matters because models run around the clock and hardware cost passes straight into the operator's economics.

TensorWave, a cloud startup that runs nothing but AMD silicon, shows how far demand has traveled. Chief operating officer Piotr Tomasik recalls that persuading customers during the company's first two years felt “like chewing glass”. External testing published last year changed that, showing AMD hardware matching or beating NVIDIA on several tasks. OpenAI, Anthropic and Meta each began adopting AMD chips from late 2025, and OpenAI said on November 16 that it will rent TensorWave's cloud compute, putting AMD-based capacity behind one of the industry's busiest workloads. TensorWave projects annual recurring revenue of $210 million by the end of this year, up from $125 million a year earlier. Cirrascale, a cloud provider operating 25 data centers, sees the same movement from the supply side. Chief executive Driggers said customers began asking last year whether workloads could move to AMD, with the questions clustered in inference, where computing demand is growing fastest. Earlier, he said, those same customers voiced skepticism about AMD rather than interest.

TPU Money Meets NVIDIA's Capacity Edge

Alphabet is courting many of the same buyers with its Tensor Processing Units, its own line of AI accelerators. The company disclosed in July that it has agreed to guarantee as much as $44 billion of data center leases, absorbing the payments if tenants default. Cirrascale recently became one of the first providers to offer TPU capacity outside Google Cloud, and Driggers said customers are preparing to spend hundreds of millions of dollars renting TPU v8 compute. Autolane, a delivery automation platform, moved part of its evaluations from GPUs to the cheaper TPUs starting in September, co-founder Chad Agate said. Adoption remains shallow so far: Vadim Solovey, chief executive of cloud software firm DoiT, counts only 16 of roughly 5,000 customers running TPUs at scale, and calls the chips “a second supplier, not a replacement”.

NVIDIA's lead still rests on supply and scale. Its latest quarterly revenue runs more than eight times AMD's for the same period, and when capacity tightens it gets first call on TSMC production lines. One startup founder who tried to rent AMD's newest chips early this year was quoted a delivery date near year-end and settled for NVIDIA hardware that shipped sooner. Michael Reid, chief executive of connectivity provider Megaport, watched a customer who inquired about AMD during NVIDIA shortages drop the idea when AMD quoted a six-month lead time against 12 to 14 weeks for NVIDIA. Daniel Svonava, chief executive of inference provider Superlinked, decided to deploy AMD after months of testing, but partly for a reason outside the silicon: since Jensen Huang's exit at NVIDIA, he doubts the company will expand capacity aggressively, on the view that finance-led leadership spends cautiously. Reid gives NVIDIA credit its rivals cannot match for its sales operation, saying it is the only vendor he has seen connect customers with financiers and data center resources. Driggers still places NVIDIA orders, and his summary is blunt: NVIDIA remains dominant, but its market share is eroding.

AMD shares last traded at $513.77, down 0.69% over 24 hours, and COINOTAG's composite scoring puts the strongest support at $504.47, rated 100/100, with the heaviest resistance at $529.66, also 100/100, leaving the stock boxed between two well-defended levels while the daily trend reads sideways. GOOGL closed the same window at $284.07, up 0.30%, inside a daily downtrend with an RSI of 39.3, and resistance at $291.62 not far above. Stock perpetual open interest stands at $27.08 million on AMD and $176.76 million on GOOGL, positions that would need to swell before derivatives flows, rather than chip orders, drive either tape.

COINOTAG's editorial and research desk.

AI-Assisted

AI-generated, AI-reviewed, under COINOTAG editorial oversight.