The Day AI Had Two Frontiers

The One Thing: OpenAI doubled its API prices on the same day a Chinese lab released an open-source model that matches last month's frontier for free. The AI pricing war just became a philosophical argument about who gets to build on top of intelligence.

If You Only Read One Thing: Simon Willison's analysis of DeepSeek V4 cuts through the benchmark theater to explain what actually matters about a 1.6-trillion-parameter open-source model landing on the same day as GPT-5.5.

TL;DR: OpenAI and DeepSeek launched frontier models within hours of each other, crystallizing the open-vs-closed divide into a single news cycle. Intel's Q1 earnings crushed expectations with AI driving 60% of revenue, though its external foundry business remains negligible at $174 million. Meanwhile, Meta and Microsoft are shedding a combined 16,000+ positions to fund the AI pivot that is making everyone else's stock go up.

Intel's Sixth Straight Beat Validates the CPU-AI Bet. The Foundry Story Is Murkier.

Intel delivered the kind of quarter that makes analysts rewrite their models. Revenue hit $13.6 billion against a $12.42 billion consensus, EPS landed at $0.29 versus the penny Wall Street expected, and shares jumped 20% in after-hours trading.

As CNBC reported, this is Intel's sixth consecutive beat, and the margin of outperformance is widening. The Data Center and AI segment generated $5.1 billion against $4.41 billion expected. Client Computing hit $7.7 billion. AI-related products now account for 60% of total sales. When we covered Intel's foundry strategy on April 15, the question was whether Lip-Bu Tan could turn manufacturing credibility into external revenue. The earnings answer that question in two directions at once.

Why it matters (Value Chain Shift): The Intel story is really two stories wearing one stock ticker. The CPU story is genuinely impressive. Tan has executed a product roadmap that positions Intel's processors as the plumbing underneath AI inference workloads, complementing Nvidia's GPUs rather than competing with them. Diamond Rapids, the next server chip with multithreading, extends this positioning. The Q2 guidance of $13.8 to $14.8 billion suggests this is not a one-quarter aberration.

The foundry story demands more skepticism. Total foundry revenue was $5.4 billion, up 16% year-over-year. But external foundry revenue was $174 million. Intel is still overwhelmingly making chips for itself. The AWS deal on 18A is real, but Intel still routes its own higher-volume Nova Lake desktop chips to TSMC. When your pitch is "trust us to make your most critical chips" and your own highest-volume chips go to the competitor, prospective customers notice. The 18A node has improved yields, but analysts have noted those yields remain below commercially profitable levels.

Room for disagreement: The market clearly disagrees with the skeptics. Intel's stock is near all-time highs, and the argument is that external foundry revenue is a lagging indicator. The AWS partnership, CHIPS Act subsidies, and geopolitical demand for non-TSMC fabrication capacity create structural tailwinds that $174 million does not yet capture. The 14A roadmap may attract commitments that have not hit the revenue line.

What to watch: External foundry revenue trajectory over the next two quarters. If it does not reach $300 million or more by Q3, the "Intel as TSMC alternative" thesis is aspirational, not operational. (Prediction update: I predicted external foundry revenue below $500M, which was correct at $174M, but also predicted a 10%+ stock pullback. Dead wrong on the stock. The market is pricing the CPU turnaround, not the foundry gap.)

The 24-Hour AI Arms Race: OpenAI Prices Up, DeepSeek Opens Up

On April 23, OpenAI released GPT-5.5, its first fully retrained base model since GPT-4.5. Within hours, China's DeepSeek dropped V4 Pro and V4 Flash in preview. The timing was deliberate.

OpenAI's announcement positions GPT-5.5 as "a new class of intelligence" optimized for agentic workflows. On Terminal-Bench 2.0, it scores 82.7%, clearing GPT-5.4 by 7.6 points and well ahead of Claude Opus 4.7 at 69.4% and Gemini 3.1 Pro at 68.5%. The model is natively omnimodal, processing text, images, audio, and video in a unified system designed for multi-tool orchestration. The price doubled: $5 per million input tokens and $30 per million output, up from $2.50 and $15.

DeepSeek's V4 Pro is a 1.6-trillion-parameter mixture-of-experts model with 49 billion active parameters, trained on 33 trillion tokens. It ships under Apache 2.0 with a 1-million-token context window. On LiveCodeBench, V4 Pro leads at 93.5, ahead of Gemini at 91.7 and Claude at 88.8. Its Codeforces rating of 3,206 edges GPT-5.4's 3,168. The price on DeepSeek's API: a fraction of OpenAI's.

Why it matters (Platform Economics): This is the open-versus-closed platform war playing out in real time. OpenAI is making the Microsoft bet: premium pricing, integrated ecosystem, enterprise lock-in. Build the super app, own the customer relationship, charge more for better tools. DeepSeek is making the Android bet: release the weights, commoditize the model layer, let the ecosystem build on top. Both strategies have historical precedent for winning.

The same-day timing is strategic. DeepSeek needed a launch window where "open-source frontier model at negligible cost" would not be buried under GPT-5.5 coverage. Instead, they split the news cycle and turned the comparison into the story itself.

The geopolitical dimension is sharpening. The same week, the Trump administration unveiled measures to prevent Chinese developers from using American AI models to build rival chatbots (first reported by Bloomberg [paywalled]). DeepSeek's response is effectively: we do not need your models, and we will give ours away.

Room for disagreement: GPT-5.5's performance on agentic tasks is genuinely superior. Enterprise customers paying for reliability, support, and integrated workflows may not care that a cheaper alternative exists. OpenAI's moat may be the ecosystem, not the model. And DeepSeek's V4 Pro trails GPT-5.4 by "3 to 6 months" by its own admission, putting it further behind GPT-5.5.

What to watch: Enterprise API adoption over the next 30 days. If developers build production systems on DeepSeek V4 at scale, OpenAI's pricing power erodes. If they stick with OpenAI despite the premium, the super-app strategy is working. (For technical architecture details on both models, see today's AI Intelligence.)

The Contrarian Take

Everyone says: Intel's earnings prove the turnaround is real. Shares surged 20%, it is the sixth straight beat, and AI now drives 60% of revenue. The "new Intel" narrative is cemented.

Here's why that's incomplete: Intel's CPU business is thriving, but the stock is priced for a future foundry giant, not just a chip designer. External foundry revenue of $174 million in a quarter is roughly what TSMC generates every 90 minutes. Intel uses 18A for low-volume Panther Lake laptop chips but outsources its high-volume Nova Lake desktop processors to the very competitor it claims to rival. The AWS deal is meaningful but narrow. Until Intel routes its own highest-stakes silicon through its own fabs, the foundry pitch carries a credibility gap that $174 million cannot bridge. The CPU turnaround is real. The TSMC-rival story is still a PowerPoint.

Under the Radar

$143 million in suspicious Polymarket profits. The soldier story is dramatic, but a Harvard Law paper published in March found over 200,000 suspicious bets totaling $143 million in insider-informed profits on Polymarket between February 2024 and February 2026. The soldier is the first prosecution, not the first case.
BCG says AI generated 25% of its 2025 revenue. Boston Consulting Group's AI advisory practice hit a quarter of total firm revenue last year (first reported by Bloomberg [paywalled]). When the consultants are making this much money telling everyone else how to use AI, the adoption curve is past the early-adopter phase and into the "everyone needs a strategy" phase.
Microsoft is investing $18 billion in Australian AI infrastructure by 2029. The largest single foreign tech investment in Australian history, focused on cloud and AI compute. The geographic diversification of AI infrastructure beyond US-only data centers is accelerating, driven by data sovereignty requirements and energy access.

Quick Takes

Meta will cut 10% of its workforce, about 8,000 jobs starting May 20, and scrap 6,000 open roles. Capital expenditure is projected to surge from $72.2 billion in 2025 to $115 billion in 2026. The pattern: fewer humans, more GPUs. CNBC reports further cuts are being discussed for later this year. Microsoft is running the same playbook more gently, offering voluntary buyouts to up to 7% of US staff. (TechCrunch)

Bitwarden's CLI was backdoored via the ongoing Checkmarx supply chain campaign. A malicious npm package published as @bitwarden/cli@2026.4.0 exfiltrated GitHub tokens, SSH keys, .env files, and cloud secrets through a domain impersonating Checkmarx. Socket.dev identified a self-propagating npm worm in the package. Bitwarden contained the breach within 90 minutes, but the Checkmarx campaign has now compromised multiple major open-source projects. (The Hacker News)

A U.S. Special Forces sergeant faces the first-ever criminal charges for prediction market insider trading. Master Sergeant Gannon Ken Van Dyke used classified knowledge of the Maduro capture operation to place $33,000 in Polymarket bets that returned $410,000. The DOJ and CFTC filed coordinated charges, establishing that prediction markets are subject to insider trading law. Over 40 lawmakers are now pushing regulation. (CNBC)

Oil crossed $96.50 as the Strait of Hormuz standoff tightens. Jet fuel prices have nearly doubled since the Iran conflict began, forcing American Airlines to slash its full-year outlook and Lufthansa to cancel 20,000 short-haul flights through October. The IEA warned several European countries may face jet fuel shortages within six weeks. (PBS NewsHour)

Stories We're Watching

The Iran Energy Crisis: Military Blockade vs. Global Oil Supply (Week 8) — Trump extended the US-Iran ceasefire, citing Tehran's government being "seriously fractured," but oil continues climbing past $96. The IEA's six-week European jet fuel shortage warning adds a hard deadline to negotiations.
Prediction Market Regulation: DOJ/CFTC vs. Polymarket (Day 1) — The Van Dyke prosecution establishes that insider trading law applies to prediction markets. With bipartisan congressional bills pending and $143 million in documented suspicious trading, Polymarket's regulatory window is closing fast.
Tech Workforce Rebalancing: Headcount Down, Capex Up (Week 3) — Meta's 8,000 cuts and Microsoft's 7% buyout offer join a wave that has eliminated over 96,000 tech jobs in 2026, even as AI capital spending accelerates to record levels.

The Thread

Thursday's news told one story from two directions. OpenAI doubled API prices while DeepSeek gave comparable capabilities away for free. Intel proved you can build a $13.6 billion quarter on CPUs that enable AI inference rather than training. These are all answers to the same question: where does value accumulate in the AI stack?

OpenAI bets it accumulates at the model layer. DeepSeek bets it accumulates at the application layer, above the model. Intel bets it accumulates at the silicon layer, below it. Meta's answer is the most revealing: value accumulates in the infrastructure itself, which is why you fire 8,000 people and spend $115 billion on data centers in the same breath.

The soldier who turned $33,000 into $410,000 on Polymarket is a footnote to this larger story, but it is also a preview. When powerful systems generate asymmetric information advantages, the people closest to the systems will exploit them. That is as true for prediction markets as it is for AI.

Predictions

New predictions:

I predict: DeepSeek V4-Pro will be adopted as a primary model by at least two Fortune 500 companies for production workloads within 60 days, driven by cost arbitrage against GPT-5.5. (Confidence: medium; Check by: 2026-06-24)
I predict: The CFTC will file a civil enforcement action against Polymarket's offshore operations within 90 days, using the Van Dyke case as predicate. (Confidence: high; Check by: 2026-07-24)

Prediction update:

pred-2026-04-15-03 (Intel Q1): Predicted external foundry revenue below $500M — correct ($174M). Predicted 10%+ stock pullback — wrong (stock surged 20%). Resolution: partially correct. The foundry gap is real, but the CPU turnaround overwhelmed it in the market's calculus.

Generated: April 24, 2026, 5:45 AM ET