Highlighted Companies
KEY COMPANIESAMD is buying its way into Claude’s compute stack rather than just selling into it — tying a multibillion-dollar investment to a 2-gigawatt MI450 order and a parallel ROCm-optimization pact turns Anthropic into both customer and reference account. For Anthropic, it’s a second major non-Nvidia supply line alongside its Google TPU commitments, exactly the diversification playbook the leading labs are now running to avoid single-vendor pricing power.
Read More →Helios bundles MI455X GPUs, next-generation EPYC “Venice” CPUs, Pensando networking, and ROCm into a single rack — Microsoft adopting it wholesale for inference workloads is the clearest signal yet that Azure is building genuine multi-sourcing at the rack level, not just the chip level. That’s leverage Microsoft can use against Nvidia allocation and pricing just as its own AI capex guidance keeps climbing.
Read More →Going GA with Shopper, Buyer, and Merchant agents just ahead of peak season is the tell — Salesforce is racing to make Agentforce the default checkout layer before OpenAI or Google’s own commerce plays get there first. Native ChatGPT catalog sync, with no extra tooling required, is a bet that being the easiest agent to plug into someone else’s assistant beats trying to own the assistant itself.
Read More →AI & Semiconductors
AI & CHIPSThis is Nvidia reaching past chip sales into the memory layer that’s actually constraining the industry — co-developing HBM4 with SK hynix while backing a 2-gigawatt Vera Rubin data center for SK Telecom addresses the bottleneck directly instead of building around it. Locking in a $500B-scale partner for both compute and memory supply is Nvidia insuring its own roadmap against the shortage everyone else is complaining about.
Read More →With data centers now absorbing an estimated 70% of global memory output, the AI buildout isn’t just competing with consumer electronics for chips anymore — it’s crowding it out, which is why ordinary server and PC memory has doubled and tripled in price alongside HBM. A shortage the industry’s own leading supplier says could last to 2030 turns memory into the multi-year capacity constraint that GPUs were in 2023–24.
Read More →Markets & Tech Stocks
S&P 500 · NASDAQ · MARKETSRaising 2026 capex guidance to $195–205B while cloud revenue jumps 82% is Alphabet telling investors the AI spend is finally showing up as growth, not just cost — the negative $5.9B free-cash-flow print is the price of believing that story. Whether the market keeps tolerating the cash burn depends entirely on cloud margins catching up before patience for “trust the capex” runs out.
Read More →A 142% jump in capex funding AI infrastructure, robotics, and Optimus pushed Tesla into its first cash deficit since 2024, and the 12% stock drop shows investors drawing a hard line between Alphabet’s AI spending (backed by an 82%-growth cloud business) and Tesla’s (backed mostly by a story about the future). Capex without a matching revenue line is starting to get punished differently than capex with one.
Read More →A semiconductor-led pullback right before Alphabet and Tesla reported was the market hedging against exactly the capex-versus-cash-flow tension those earnings ended up confirming. Expect this pattern — chips wobble, megacaps report, the tape re-prices around whoever’s AI spending looks best-justified — to keep repeating through the rest of earnings season.
Read More →Supply Chain & Commodities
CHIPS · MATERIALS · FREIGHTWith Amazon, Alphabet, Meta, Microsoft, and Oracle collectively approaching $690B in 2026 capital spending, the conversation among hyperscalers has shifted from whether to spend to whether the grid can keep up — power, not chips or even memory, is now the binding constraint on how fast new capacity comes online. That reframes the AI infrastructure race as an energy and permitting problem as much as a semiconductor one.
Read More →Advanced packaging — stacking compute and memory dies into a single package — has quietly become as tight a bottleneck as the chips themselves, and Nvidia prepaying to expand Amkor’s Arizona capacity is Nvidia insuring its own supply chain the same way it’s doing with memory via SK hynix. Coming a month after Amkor’s 10-year TSMC packaging pact, this is the “invisible” layer of the AI buildout getting the same capital urgency as fabs and GPUs.
Read More →Meta building capacity faster than its own model roadmap can absorb it — and now shopping the surplus to outside customers — is the first real evidence that at least one hyperscaler has overbuilt relative to near-term demand, exactly the scenario the market has been pricing out. CoreWeave’s 13% drop on the news shows how quickly “compute shortage” sentiment can flip to “oversupply” fear for anyone whose business model depends on the gap Meta just said it can fill.
Read More →