In partnership with
Sunday, April 19, 2026 | Private Markets Intelligence
Good morning. OpenAI just committed $20B to chips it won't own. Snap blamed AI for firing 1,000 people. And Maine became the first state to ban the thing powering everything: data centers.
This week wasn't about who raised what. It was about the infrastructure war underneath—who controls compute, who pays for it, and who's left out when the bill comes due.
Founderscrowd
The intel:
OpenAI's $20B Cerebras bet reveals the real AI bottleneck (it's not models)
Snap cuts 16% of workforce, says AI replaced the humans (stock up 11%)
Sequoia's $7B fund shows late-stage venture economics have fundamentally changed
Maine bans large data centers nationwide, first state to pump the brakes
Nvidia's $20B Groq buy vs. OpenAI's $20B Cerebras deal = the inference war is here
The Biggest Names in Food Are Testing This New Tech
What do the world’s three biggest foodservice providers, the biggest food manufacturer, and some of the industry’s most beloved brands have in common?
They’re all partnering with Automated Retail Technologies (ART) to serve food without kitchens or delivery.
With ART’s Just Baked™ automated robotic kiosks, serving peoples’ favorite meals 24/7 is affordable and easy. That’s why the foodservice giants Aramark, Sodexo, Sysco, and Compass Group joined Nestlé and White Castle as ART partners.
And ART’s 800 units deployed currently only scratches the surface of their growth plans. They have 400 more units ready for immediate deployment, 340,000+ additional targeted locations, and a leadership team that knows exactly what it takes to scale. In other words, ART is perfectly positioned.
This is a paid advertisement for Automated Retail Technologies Regulation CF offering. Please read the offering circular at https://invest.automatedrt.com/
💰 OPENAI'S $20B CEREBRAS DEAL EXPOSES THE REAL AI BOTTLENECK
What happened: OpenAI agreed to spend $20B+ over three years on Cerebras chips—double their previous $10B January commitment. The deal includes equity warrants that could give OpenAI up to 10% of Cerebras. Cerebras filed for IPO the same day, targeting $35B valuation. This is one of the largest infrastructure deals in AI history.
The numbers:
New commitment: $20B over 3 years
Previous deal (Jan 2026): $10B for 750 megawatts
Total OpenAI-Cerebras commitment: $30B+
Cerebras IPO target: $35B valuation
OpenAI equity stake: Up to 10% via warrants
Nvidia's Groq acquisition (Dec 2025): $20B
Timeline: Same amount, same month, opposite strategies
Why it matters: This isn't about chips. It's about control. OpenAI is spending $20B to escape Nvidia's grip on AI inference (running models, not training them). Cerebras builds wafer-scale chips—one chip the size of an entire silicon wafer—optimized for low-latency inference. OpenAI needs this because inference is becoming 2/3 of AI compute spending.
The bigger picture: Two $20B deals in 5 months tell you everything about the AI infrastructure war:
Nvidia's $20B (Dec 2025): Acquired Groq (inference chip startup) defensively—plugging a gap Nvidia couldn't fill
OpenAI's $20B (April 2026): Buying Cerebras capacity offensively—building an inference highway that doesn't depend on Nvidia
Both moves happened because AI is shifting from training to inference. Training a model once costs billions. Running it billions of times (inference) costs more. The bottleneck moved.
What this means for private markets: Compute is the new currency. When the world's largest AI company writes a $20B check for chips and gets equity in return, it's not a purchase—it's a strategic alliance. Cerebras gets funding + the world's highest-profile customer. OpenAI gets inference capacity + optionality on Nvidia alternatives + potential upside if Cerebras IPO works.
For investors: This validates the "picks and shovels" thesis, but with a twist. The real money isn't in generic infrastructure—it's in specialized infrastructure that solves specific bottlenecks (inference vs. training, edge vs. cloud, low-latency vs. high-throughput).
Investor takeaway: When model builders spend more on infrastructure than some countries spend on defense, the bottleneck is no longer ideas—it's execution capacity. If you're investing in AI, ask: is this company a model (commoditizing fast) or infrastructure (capturing value as models scale)?
📉 SNAP FIRES 1,000 PEOPLE, BLAMES AI, STOCK JUMPS 11%
What happened: Snap cut 1,000 employees (16% of workforce) and closed 300 open roles, citing "rapid advancements in artificial intelligence" that "enable teams to reduce repetitive work." CEO Evan Spiegel said small teams using AI tools are driving progress across Snapchat+, ads, and infrastructure. Stock rose 11% on the news.
The numbers:
Employees cut: 1,000 (16% of 5,261 full-time staff)
Open roles eliminated: 300+
Cost savings: $500M+ annualized by H2 2026
Severance costs: $95M-$130M (mostly Q2 2026)
Stock reaction: +11% (opened at $6.23)
U.S. severance: 4 months + healthcare + equity vesting
AI-generated code: 65%+ of new code at Snap
Why it matters: This is the first major tech layoff where the CEO explicitly said "AI replaced these jobs" instead of blaming macroeconomic conditions, overhiring, or restructuring. Spiegel's memo was clear: AI tools let smaller teams do more, faster. Translation: we don't need these people anymore.
The market's reaction tells you everything: Stock up 11% means Wall Street believes AI-driven cost cuts are real. Activist investor Irenic Capital (2.5% stake) had been pushing for exactly this—eliminate 1,000 roles, let AI replace them, save $500M. Snap delivered.
What Snap protected: Despite cuts, Snap is still investing $3.5B in AR glasses (Specs), expected to launch this year. Irenic wants Specs shut down or spun off. Spiegel is betting Specs is the future, Snapchat is the cash cow to fund it. The layoffs are surgical—protect hardware moonshot, cut the legacy app overhead.
Why this is a turning point: Snap is the canary in the coal mine. If a struggling social app (946M MAU, still losing $460M/year) can cut 16% of staff and have Wall Street cheer, expect this playbook everywhere:
Deploy AI tools across product/ads/infra
Measure productivity gains
Cut headcount proportionally
Market it as "efficiency," not layoffs
Watch stock rise
Investor takeaway: AI isn't just a product category—it's a cost structure shift. Companies that adopt AI tools and resize accordingly will be rewarded. Companies that adopt AI but keep legacy headcount will be punished. For startup investors: ask founders how AI changes their unit economics. If the answer is "we're adding AI features," dig deeper. The real question is: how many fewer people do you need to scale?
WANT IN ON DEALS LIKE THESE?
Most of these opportunities disappear before they hit the news.
Founderscrowd Premium members get first look at exclusive pre-IPO investments, the kind of deals institutional investors see before they go mainstream. We're talking early-stage access to companies that could be the next exits in satellite, EVs, fintech, and AI.
Premium includes:
✅ Weekly deal memos on live investment opportunities
✅ Private markets intel before it's public
✅ Direct access to carry on winning deals
✅ Community of 1,000+ active private markets investors
The difference? When OpenAI acquires a company, Premium members already knew the founders. When Amazon drops $11.57B, Premium members already owned the company. When Tesla hits 10M subscribers, Premium members will have already ridden the private equity wave.
Become a Premium member →
Lock in $40/month before it goes to $100+ (last week)
💸 SEQUOIA'S $7B FUND PROVES AI CHANGED LATE-STAGE ECONOMICS
What happened: Sequoia Capital raised ~$7B for its expansion strategy fund (late-stage US/Europe), nearly doubling its 2022 fund ($3.4B). First major fundraise under new co-stewards Alfred Lin and Pat Grady, who took over after Roelof Botha was removed in November 2025. The fund will back companies like OpenAI, Anthropic, Physical Intelligence, and Factory.
The numbers:
New fund: ~$7B
Previous fund (2022): $3.4B
Growth: 105% increase
New leadership: Alfred Lin + Pat Grady (co-stewards since Nov 2025)
Focus: Late-stage US/Europe (Series C+)
Key portfolio: OpenAI, Anthropic, Physical Intelligence, Factory
Total Sequoia AUM: $56B globally
Both OpenAI and Anthropic eyeing: 2026 IPOs
Why it matters: When the best VC firm on the planet doubles its late-stage fund, it's not speculation—it's a structural shift. Pre-AI, late-stage meant writing $50M-$100M checks to help companies polish operations before IPO. Post-AI, late-stage means writing $500M-$1B checks into companies burning billions on compute to achieve model dominance.
What changed: AI companies scale faster and need more capital than any previous tech wave. OpenAI went from $0 to $852B valuation in 8 years. Anthropic: $0 to $330B in 4 years. Both need billions for compute, not millions for salespeople.
Sequoia's bet: This isn't a bubble. It's a new baseline. Companies building foundational AI infrastructure require more capital, grow faster, and produce bigger outcomes than internet, mobile, or cloud ever did.
The leadership change context: Alfred Lin (Airbnb, DoorDash early investor) and Pat Grady (ServiceNow, OpenAI, Harvey) taking over from Botha is significant. This fundraise is their opening statement: Sequoia is doubling down on AI, not retreating. They're breaking VC taboos (backing both OpenAI and Anthropic—direct competitors) because they believe both can win in a market this big.
What this signals about AI durability: Sequoia doesn't gamble. They have 54 years of data on tech cycles. If they're doubling fund sizes for AI, they're seeing something structural that justifies the risk. The fund math only works if:
AI companies continue scaling faster than predecessors
Capital intensity remains high (compute costs don't collapse)
Exit multiples stay elevated (IPOs or M&A at premium valuations)
All three are happening. OpenAI and Anthropic are both rumored to IPO in 2026. If both go public at current valuations, Sequoia's returns from these two alone could justify the entire $7B fund.
Investor takeaway: Follow the smart money. When top-tier VCs increase fund sizes, they're responding to market structure, not hype. For angel investors who can't write billion-dollar checks: invest in the layers around the giants. Sequoia backs OpenAI/Anthropic (models). Where's the opportunity? Deployment infrastructure (Vercel), specialized compute (Cerebras), data labeling (Mercor), security (applied AI for cyber), and vertical applications (AI for legal, healthcare, finance).
🚫 MAINE BANS LARGE DATA CENTERS—FIRST STATE TO HIT THE BRAKES
What happened: Maine became the first U.S. state to pass a statewide moratorium on large data centers (20+ megawatts), lasting until November 2027. The bill creates a Data Center Coordination Council to study energy, water, and economic impacts before allowing future projects. Governor Mills requested an exemption for a $550M project in Jay (former paper mill site) but lawmakers rejected it.
The numbers:
Ban threshold: 20+ megawatts
Moratorium length: 18 months (until Nov 2027)
Coordination Council budget: $95,000
Rejected project (Jay): $550M proposed investment
Local opposition: Bangor passed 6-month ban, Lewiston protests
National trend: 230+ organizations called for nationwide moratorium (Oct 2025)
Senator Bernie Sanders: Introduced federal moratorium bill (March 2026)
Why it matters: Maine just became the first state to say "no" to AI infrastructure. Not "regulate it" or "tax it"—just stop. The reasoning: data centers spike electricity costs, drain water supplies, and provide limited local benefit. States that welcomed data centers (Virginia, Texas, Arizona) saw energy prices surge and communities revolt.
What Maine is worried about:
Energy: Data centers can consume more power than entire towns. Maine's grid can't handle sudden spikes without raising rates for everyone.
Water: Cooling systems use millions of gallons. Maine has water, but paper mills (like the closed Androscoggin mill) used to employ thousands. A data center employs dozens.
Jobs vs. infrastructure: The Jay project would create ~100 permanent jobs but require massive grid upgrades. Former mill employed 1,000+. Locals asked: is this progress?
The political fight: Governor Mills (Democrat, running for U.S. Senate) wanted an exemption for Jay because the town desperately needs jobs and tax revenue post-mill closure. Lawmakers said no—exemptions defeat the point. If Maine studies data centers and decides they're bad, the Jay project is bad too. If they're good, lift the moratorium for everyone in 2027.
Why this is bigger than Maine:
Trend: Bangor (6-month ban), Lewiston (protests), Vermont (considering similar), Pennsylvania (local opposition). Communities are rejecting data centers.
Federal attention: Bernie Sanders introduced a nationwide moratorium bill. 230 organizations signed a letter supporting it.
AI's dirty secret: Training GPT-5 uses as much energy as a small country for a year. Inference (running models) uses even more. Someone has to pay for that power, and it's usually local ratepayers.
What this means for AI infrastructure: Tech companies assumed they could build data centers anywhere with cheap land and grid access. Maine just proved that's wrong. Communities are pushing back on:
Energy rate increases (data centers spike demand)
Water scarcity (cooling systems compete with agriculture/residents)
Job creation myths (100 tech jobs vs. 1,000 mill jobs)
Tax revenue vs. infrastructure costs (roads, grid, water upgrades)
Investor takeaway: The AI boom has a NIMBY problem. Data centers are becoming as controversial as power plants. If you're investing in AI infrastructure, ask: where does this physically live, who pays for the power, and what's the local political risk? Maine won't be the last state to hit pause. Smart money is flowing to:
Distributed compute (edge inference, not centralized training)
Energy-efficient chips (Cerebras, Groq focus on inference efficiency)
Nuclear partnerships (Microsoft, Google, Amazon buying future power)
International (building where NIMBYism is weaker—UAE, Singapore, Iceland)
⚔️ NVIDIA VS. OPENAI: THE $20B INFERENCE WARS
What happened: Two $20B deals, five months apart, same amount, opposite strategies. Nvidia bought Groq (inference chip startup) for $20B in December 2025. OpenAI committed $20B to Cerebras chips in April 2026. Both moves are about the same thing: who controls AI inference (running models), not training (building models).
The numbers:
Nvidia's $20B (Dec 2025): Acquired Groq outright (inference specialist)
OpenAI's $20B (April 2026): 3-year Cerebras chip commitment + equity warrants
Market shift: Inference will be 2/3 of AI compute spending by 2026
Cerebras IPO: Filed April 17, 2026, targeting $35B valuation
Groq before acquisition: Valued at $2.8B (Nvidia paid 7x premium)
Why this is a war:
Nvidia's problem: Their GPUs dominate training (building models) but are inefficient for inference (running models). Inference is becoming the bigger market.
Groq's solution: Built chips optimized for inference—low latency, high throughput, deterministic performance. Exactly what Nvidia doesn't have.
Nvidia's move: Buy the competitor rather than build it internally. $20B is defensive—plug the gap before customers defect.
OpenAI's counter-move:
OpenAI's problem: Nvidia has pricing power. As long as OpenAI depends on Nvidia GPUs, Nvidia captures margin.
Cerebras's solution: Wafer-scale chips (one chip = one silicon wafer) optimized for inference. Faster than GPUs for specific workloads.
OpenAI's move: Commit $20B over 3 years, get compute capacity + equity warrants + leverage against Nvidia.
What this reveals about the AI stack: The real money isn't in models (commoditizing) or apps (competitive). It's in infrastructure control:
Nvidia: Owns training (GPUs for model building)
Groq (now Nvidia): Owns inference efficiency
Cerebras: Challenging both with wafer-scale architecture
OpenAI: Trying to own the full stack (models + compute)
Why inference matters more than training:
Training GPT-5: Happens once, costs billions
Running GPT-5 for 1 billion users: Happens continuously, costs more
Inference = recurring revenue engine
Training = one-time capex
As AI moves from "build the model" to "run the model at scale," inference becomes the bottleneck. Whoever controls inference chips controls AI economics.
What Cerebras's IPO timing tells you: Filing the same day as the $20B OpenAI announcement isn't coincidence. Cerebras is saying: "We're IPOing at $35B because we have the world's largest AI company locked in for $20B+ over 3 years." That's a revenue guarantee that justifies the valuation.
Investor takeaway: The AI infrastructure war is real, and it's expensive. Nvidia spent $20B defensively (buy Groq before customers leave). OpenAI spent $20B offensively (fund Cerebras to escape Nvidia). Both moves validate the same thesis: inference is the new battleground, and specialized chips win.
For investors: Don't bet on general-purpose AI. Bet on specialized infrastructure solving specific bottlenecks. Inference chips (Groq, Cerebras) are outcompeting GPUs for production workloads. Edge inference (running models on devices, not cloud) is next. The winners will be companies that make AI cheaper to run, not fancier to build.
QUICK HITS
🔥 Other macro moves this week:
Amazon/Globalstar $11.57B — Buying satellites to meet FCC deadline, not Starlink competition (regulatory shortcuts = premium prices)
Reed Hastings exits Netflix — Founder stepping down after 29 years, company thriving (325M subscribers, live sports launched)
Vercel IPO signals — ARR $100M → $340M (240% growth), 30% of apps now AI-generated, CEO says "ready every day"
Cybertruck inter-company sales — SpaceX bought 18% of Q4 sales, Musk companies spent $100M+ propping up demand
Nvidia acquires Groq — $20B for inference chip startup (same month as OpenAI's $20B Cerebras commitment)
BY THE NUMBERS
This week's macro moves:
$20B — OpenAI spending on Cerebras chips (3 years)
$20B — Nvidia paid for Groq (Dec 2025)
$7B — Sequoia's new AI late-stage fund
$500M — Snap's cost savings from AI-driven cuts
$95K — Maine's budget for data center study council
$35B — Cerebras IPO target valuation
16% — Snap workforce reduction
2/3 — Share of AI compute spending shifting to inference
30% — Vercel apps created by AI agents
65% — Snap's new code generated by AI
That's this week in private markets.
OpenAI is betting $20B on chips it won't own, Snap cutting people and calling it progress, Sequoia doubling down while Maine pumps the brakes. The AI boom is real, but the bills are coming due—in power costs, headcount cuts, and infrastructure battles.
The question isn't whether AI works. It's who controls it, who pays for it, and who gets left behind.
⭐⭐⭐⭐⭐ This is the signal
⭐⭐⭐ Average
⭐ Needs work
See you next week,
Founderscrowd team
P.S. Tomorrow, we're breaking down why carry (performance fees on private market investments) is where most investors make real money, not the fund size. The math is simple, but most people get it backwards.