Enterprises spent $401B on GPUs and are using 5% of them. Meanwhile Ramp wants $40B, DeepSeek may hit $45B, and a 7B model just outperformed every frontier model by routing tasks between them. It is a wild Friday. Here is what moved today.
Funding
$40B+ Ramp in talks to raise $750M at $40B+ — six months after its $32B round
The corporate spend management platform is in discussions to raise $750 million at a pre-money valuation north of $40 billion — up from the $32 billion valuation it achieved just six months ago.
At a rumored 40× revenue multiple, investors are betting that AI-powered spend management is the next category-defining enterprise platform. The market cannot get enough of Ramp.
$22B Kalshi doubles to $22B — $1B Series F, Wall Street now all-in on prediction markets
The $1B Series F was led by Coatue with Sequoia, a16z, Paradigm, Morgan Stanley, and ARK Invest participating. Valuation doubled from $11 billion in just five months. Kalshi now hosts 90% of U.S. prediction market activity.
Prediction markets have gone from niche crypto experiment to mainstream financial infrastructure in under two years. When Morgan Stanley and ARK are both in the same round, the institutional validation is complete.
$80B China's Moonshot AI raises $2B — nearly 5× valuation jump in five months
The lab behind the Kimi series of open-weight LLMs has raised $3.9 billion over six months. The latest round was led by Meituan's venture arm with Tsinghua Capital, China Mobile, and CPE Yuanfeng. Valuation jumped from $4.3 billion at year-end 2025 to $20 billion today.
Chinese open-source models are catching up fast. Investors are placing massive bets on cheap inference and open-weight architectures. Western labs own the headlines; Chinese labs are closing the capability gap.
$45B DeepSeek eyeing $45B valuation — first external funding round after years of refusing outside capital
The Chinese AI startup is in talks to raise $3–4 billion led by China's National Integrated Circuit Industry Investment Fund. Valuation could reach $45 billion — up from an internal estimate of roughly $20 billion just weeks ago.
DeepSeek has rejected outside capital for years. This reversal signals that even the most disciplined players need a war chest to stay competitive in the frontier model race.
$1.4B Blitzy raises $200M — 5× engineering velocity for Global 2000 enterprises
The Cambridge-based autonomous software development platform counts State Street and QAD among its customers. Northzone led the round with Battery Ventures, PSG, Jump Capital, Liberty Mutual, and Erie Strategic Ventures participating.
The AI coding assistant market is white-hot. Cursor is valued at over $50B. Blitzy is carving out the enterprise segment — and the competition is only getting started.
$1.3B Corgi hits $1.3B — fastest YC insurtech unicorn, four months after its Series A
The AI-native insurance carrier raised a $160M Series B led by TCV, just four months after its $108M Series A. Valuation doubled from $630 million to $1.3 billion.
AI is eating insurance. Corgi's trajectory suggests the industry is ready for full-stack automation — not just incremental efficiency tools.
Launches
$16M Pit — a16z backs Voi founders' new enterprise AI startup out of Stockholm
The Swedish startup raised a $16M seed led by a16z. Pit builds enterprise AI products that learn how clients' businesses operate and then create custom software to automate their processes. CEO Adam Jafer, formerly of Voi, has staffed the team with ex-iZettle and Klarna engineers.
European enterprise AI is a16z's new hunting ground. Stockholm is quietly becoming a hub for serious AI engineering talent, and Pit is the latest signal of that shift.
AI & Tech
AI Sakana's 7B model outperforms every frontier model by routing tasks between GPT-5, Claude, and Gemini
Sakana AI's "RL Conductor" is a 7 billion parameter model trained with reinforcement learning to automatically orchestrate GPT-5, Claude Sonnet 4, and Gemini 2.5 Pro. The Conductor analyzes inputs, assigns tasks to the best available model, and coordinates multi-agent workflows. It outperformed every individual frontier model on reasoning and coding benchmarks.
If a 7B model can route tasks better than a human engineer, the orchestration layer of enterprise AI just became a commodity. This is a quiet shift with large consequences.
AI ZAYA1-8B — open-weight MoE model trained entirely on AMD, beats much larger models on math and code
Zyphra's ZAYA1-8B is an open-weight Mixture-of-Experts model trained on 1,024 AMD Instinct MI300X GPUs with AMD Pensando Pollara networking. It uses fewer than one billion active parameters per forward pass but matches or exceeds much larger open-weight models on mathematics and coding benchmarks. Released under the Apache 2.0 license.
Nvidia's dominance in AI training is not unassailable. AMD now has a production-grade reference stack — and the open-source community is starting to adopt it.
Market Pulse
Milestone Gusto hits $1B in revenue — both Gusto and Rippling now at $1B+, still private
The HR and payroll platform crossed $1 billion in actual revenue (not ARR). Rippling hit the same milestone last month. Gusto was last valued at $9.3 billion. CEO Josh Reeves has downplayed IPO timing — but the milestone puts pressure on rivals and private market comps.
The HR software market is a two-horse race. Both Gusto and Rippling at $1B+ revenue, both still private. The eventual IPO sets the ceiling for the entire category.
Earnings B2B reacceleration is real but uneven — Twilio, Atlassian, and Palantir proving it
Twilio went from 4% growth in Q2 2024 to 20% in Q1 2026 — its fastest rate in three years. Atlassian revenue jumped 32% to $1.79B. Palantir continues to beat expectations. HubSpot and Shopify still have work to do.
The B2B recovery is not a rising tide lifting all ships. If your company is not growing at 20%+ right now, you are losing share to competitors who have figured out AI go-to-market.
Trend April funding hits $56B — up 100% year over year, third-highest month in a year
Global venture funding reached $56 billion in April, up 100% year over year from $26 billion. The surge was driven by large rounds to Anthropic and Jeff Bezos's Project Prometheus, an AI manufacturing venture.
The top-tier funding spree is not slowing. The majority of the $56 billion is concentrated in AI, defense, and infrastructure — and the concentration is tightening.
Today's Takeaway
Insight $401B in AI infrastructure spending — and enterprises are using 5% of it
Gartner estimates AI infrastructure is adding $401 billion in new spending this year. But real-world audits tell a different story: average GPU utilization in the enterprise is stuck at 5%. Enterprises panic-bought GPU capacity for two years. Now the fleet sits idle — because teams hoard the scarce resource rather than risk losing access.
Releasing idle capacity would improve utilization. But the same shortage that drove panic buying is exactly why no team will give capacity back. So the fleet idles at 5%, and the bill keeps climbing.
Three things this means for you:
Wasted compute is the new shadow IT. CFOs are starting to ask why the GPU line item is exploding while model performance is flat. That conversation is coming to every boardroom.
The fix makes the problem worse. Central pools and schedulers help, but they do not solve the hoarding incentive. The only real fix is cultural: treat GPU capacity as a shared utility, not a personal stockpile.
High utilization is a competitive advantage. Startups that build lean inference pipelines can undercut incumbents on price. The $401B enterprise waste is their subsidy.
Enterprises panic-bought GPU capacity for two years. Real-world audits show average utilization stuck at 5%. $401B in new infrastructure spending — and 95% of it is sitting idle.
Join 5,000+ founders & operators
Get the daily startup pulse, delivered before 8am.
Know a founder still panic-buying GPUs? Forward them the 5% stat. The waste is staggering — and avoidable.