NVIDIA (NVDA) added approximately $116 billion in market value on Wednesday as expanding AI budgets from hyperscalers revitalized chip demand, while rival AMD secured a landmark agreement with AI developer Anthropic for up to 2 gigawatts of its next-generation MI450 accelerators. The pair of developments underscores the surging appetite for AI infrastructure—but also raises questions about the independence of hardware demand when vendors help finance their customers.
Alphabet raised its 2026 capital expenditure forecast to a range of $195 billion–$205 billion, an 8.1% increase over its previous midpoint. CFO Anat Ashkenazi told investors that “the demand still outpaces that investment.” Google Cloud revenue jumped 82% to $24.8 billion, though Alphabet posted a negative free cash flow of $5.9 billion after heavy infrastructure spending, sending its shares down about 3% in after-hours trading.
The renewed confidence in AI spending helped NVIDIA outperform the semiconductor index by 1.8 percentage points and the Nasdaq Composite by 2.9 points. Bullish analysts like @StockOptionCole flagged price targets of $240, $260, and $300, citing a technical recovery near the 100-day moving average.
AMD’s deal with Anthropic arrived shortly after. The agreement covers up to 2 gigawatts of AMD Instinct MI450 graphics processors deployed through the Helios rack-scale platform, with the first gigawatt scheduled to begin in the first half of 2027. Alongside the hardware commitment, AMD will invest up to $5 billion in Anthropic, tying payments to deployment milestones. The partnership also includes a multiyear engineering collaboration on hardware, software, and the use of Anthropic’s Claude models inside AMD’s development processes.
The structure rekindles a debate already seen with NVIDIA: when a chipmaker simultaneously finances a customer, how should investors interpret the demand? Tom Brown, Anthropic’s chief compute officer, described access to computing capacity as essential to keeping Claude competitive. For AMD, a successful rollout could validate its platform for other hyperscalers seeking alternatives to NVIDIA’s dominant CUDA ecosystem.
Yet the deal does not deliver immediate revenue. The first systems won’t deploy until 2027, and AMD must still demonstrate that MI450 meets performance targets in production environments. Wistron’s new $700 million Fort Worth factory, now producing NVIDIA GB300 systems, illustrates the capacity race on the other side. Alphabet, too, disclosed direct TPU chip sales for the first time, adding another competitive layer.
NVIDIA’s scale dwarfs all rivals. Fiscal first-quarter revenue surged 85% to $81.6 billion, with data-center revenue up 92% to $75.2 billion. The company guided for $91 billion in second-quarter sales, though it excludes China data-center compute. The expanding AI budgets continue to fuel chip demand, but competition is steadily broadening across the accelerator market.
