OpenAI's Code Red, DeepSeek's New GPT-5 Challenger Has a Catch, & Why SaaS Is In Trouble

Google's Gemini 3 triggers emergency response at OpenAI, DeepSeek's pricing comes with security trade-offs, and HubSpot's CTO misses how AI can disrupt SaaS

OpenAI Declares “Code Red” as Google’s Gemini 3 Catches Up

OpenAI CEO Sam Altman declared a “code red” on Monday, redirecting company resources to improve ChatGPT as Google’s competitive threat intensifies.

The memo delays work on advertising, a personal assistant called Pulse, and AI agents for health and shopping. Employees responsible for ChatGPT improvements will now attend daily calls.

The urgency stems from Google, whose monthly active users jumped from 450 million in July to 650 million after launching Gemini 3. The model beat ChatGPT on benchmark tests for math, reasoning, and multimodal understanding.

This chart shows why that growth is so concerning. Gemini’s daily visit share jumped from 22% to 31% of ChatGPT’s traffic in just 2 weeks.

The irony is funny, as Google famously issued their own code red after ChatGPT went viral in early 2023. We were very new back then, our article about it only has 8 views!

Coatue shared this rebuttal, which highlights how seasonal ChatGPT is. Usage falls after Thanksgiving because students go home for the break. It picks back up for the finals, and then tapers off again for winter break.

And that story looks plausible. If college students were already paying for ChatGPT, why would this seasonality harm Gemini? The real question is what will happen once students are back in campus.

While I would be surprised if students abandoned ChatGPT right now, Sam Altman should be legitimately worried. John Gruber made a great point on a Dithering podcast that Google has dominated multiple markets without being first.

Chrome overtook Internet Explorer, Android vanquished any hope Microsoft had in mobile, and Gmail dominated personal emails.

Google quality is catching up. Marry that with their incredible infrastructure and distribution, OpenAI should be worried.

---

DeepSeek Ships GPT-5 Rival at Fraction of OpenAI’s Price, With a Catch

As if Gemini wasn’t enough, Chinese startup DeepSeek released DeepSeek-V3.2 and DeepSeek-V3.2-Speciale on December 1, open-source models that match or beat GPT-5 on key benchmarks while costing a fraction to run.

The pricing is remarkably cheap compared to their peers.

The technical innovation is DeepSeek Sparse Attention (DSA), which identifies only the most relevant portions of context for each query rather than processing everything. The company calls it a “lightning indexer,” and it addresses the computation bottleneck of traditional attention mechanisms. It helps them produce frontier AI models despite U.S. export controls restricting China’s access to advanced Nvidia chips.

The pricing is incredible for their performance, but I wonder how much Americans will use it.

More white papers are coming out about security flaws when building with DeepSeek. CrowdStrike wrote about how DeepSeek produced vulnerable code when certain trigger words were used. For example, telling DeepSeek it was helping “a financial institution based in Tibet”.

Just a line like that increased severe vulnerabilities by 50% compared to the baseline. CrowdStrike explains why this is likely happening.

Chinese laws concerning generative AI have explicit requirements. For example, Article 4.1 of China’s “Interim Measures for the Management of Generative Artificial Intelligence Services” mandates that AI services must “adhere to core socialist values.” Further, the law prohibits content that could incite subversion of state power, endanger national security, or undermine national unity.

“National security” is especially broad, and could scare most American companies away from using DeepSeek, despite the incredible pricing.

Will the value be worth these concerns? OpenRouter shows them 5th in market share, and that share has been declining.

Let’s track how much this changes over the next month. DeepSeek offers GPT-5 level performance for a fraction of the cost, but what are the tradeoffs?

---

HubSpot’s CTO Misses Why AI Threatens SaaS

HubSpot’s CTO Dharmesh Shah dismissed concerns about AI killing SaaS, arguing that vibe coding won’t replace purchased software, because companies lack the focus and resources to maintain custom apps.

It’s hard disagreeing with his points about vibe coding. But those points completely miss the bear case for SaaS, and that bear case is already playing out in HubSpot’s revenue.

It’s not from customers vibe coding. It’s that thousands of AI-first competitors can now build niche vertical solutions that are both cheaper AND provide more value.

HubSpot charges $800/seat for Marketing Hub Professional. A vertical competitor could charge $200/month and deliver more value because they’re hyper-focused and supercharged with AI.

Incumbents can’t respond to this. They can’t suddenly become nimble, focused solutions without cannibalizing their existing revenue. This is SaaS disrupting on-premises all over again, like what Salesforce did to Siebel.

Salesforce is already showing the cracks, growing just 8% YoY while locking down customer data to block those AI competitors.

HubSpot’s revenue growth has been declining as well, and has underperformed the Nasdaq 100 ($QQQ) over the past 4 years. The market sees what Dharmesh is missing.