Kalshi launched Compute Forward Curves on July 14, a pricing benchmark for the future cost of renting Nvidia B200, H200, and A100 capacity, built from its own prediction-market trading data. Many believe compute is the new oil, so it will need a real derivatives market stood up around it. It’s the third market-structure entrant racing to price GPU capacity this year, after CME’s compute futures with Silicon Data in May and ICE’s tie-up with Ornn on cash-settled contracts, and it marks GPU scarcity shifting from an internal planning input toward a tradeable commodity class. Watch whether real contract volume follows the launches, and whether hyperscalers trade as counterparties or just watch the price discovery from outside. Check out this episode of Kalshi CEO Tarek Mansour on how prediction markets price the future to hear how Kalshi thinks about this model.
Private Companies
Vast
Partnership
Cloudera and VAST Data partnered to launch a “unified AI factory,” pairing Cloudera’s containerized data-lakehouse services with VAST’s disaggregated storage, database, and global-namespace layer, plus Nvidia’s cuVS vector search. The target is what Cloudera calls GPU starvation — accelerators idling behind data pipelines that can’t feed them fast enough. The deal lands five months after VAST’s Series F closed at a $30B valuation with Nvidia as an investor, a proof point for VAST’s platform bet as GPU utilization becomes the top question investors ask about every neocloud buildout.
Public Markets
ASML raised its full-year 2026 guidance for the second time this year, to €43–45B from €36–40B, after a Q2 beat; Intel disclosed the same day it’s now running production on ASML’s most advanced High-NA EUV tool, a sign both logic and memory demand keep outrunning forecasts.
Emerging
TYLsemi exited stealth with an oversubscribed $43M early-stage round led by Matter Venture Partners, alongside Viola Ventures, GHOVC, and Egis Technology. Founders Mohit Gupta and Sunil Bhardwaj, both Alphawave and Cadence veterans, are building a chiplet platform spanning IO connectivity (PCIe, UALink, co-packaged optics), integrated voltage regulation, and a planned memory chiplet, packaged as a custom-silicon integration service called TYL.Forge. First samples ship in 2027 through a TSMC manufacturing partnership. The bet is that mid-tier AI infrastructure buyers, not just hyperscalers, will want custom silicon, and owning IO, power, and integration end-to-end can cut design timelines in half.
Watch This Week
Thursday, July 16
TSMC reports Q2 2026 earnings, its first print since ASML’s guidance raise and June’s record monthly revenue; consensus expects profit up 57% YoY, with guidance the read on whether logic-side AI demand offsets memory-side volatility.
Wednesday, July 22
SK Hynix reports Q2 2026 earnings, the first full look at whether the softer-than-consensus operating-profit preview that triggered Monday’s selloff proves out, and whether ASML’s demand read extends to the memory side.