Elon Musk Defends Tesla’s AI Strategy After Short Seller Jim Chanos Questions FSD Progress: 'Building Cortex 2’

• Jim Chanos questioned why Tesla’s FSD remains at Level 2 while rivals have advanced.
• Gavin Baker said Tesla’s data scale and training architecture give it cost advantages in AI.
• Musk recently said Tesla is “a few months away” from offering unsupervised FSD.

Short seller Jim Chanos, the founder of Kynikos Associates, who gained prominence after predicting Enron's collapse, questioned Tesla’s progress on Full Self-Driving (FSD) and its investment in AI training. 

Chanos argued in posts on X that despite beginning work on FSD in 2014, Tesla remains at Level 2 supervised autonomy, which he noted is the same classification used by GM’s Super Cruise and Ford’s BlueCruise. He also pointed out that Mercedes’ Drive Pilot has obtained limited Level 3 approval, raising the question of why Tesla’s system has not progressed beyond Level 2 after more than a decade of development.

Gavin Baker Argues Tesla’s AI Economics Are Misunderstood

Atreides Management’s Gavin Baker challenged Chanos’s interpretation, saying Tesla’s relatively low capital expenditure does not signal a lack of seriousness in AI or robotics. Baker argued that Tesla’s architecture enables customers to subsidize inference compute by running inference inside the vehicle, reducing the need for costly synthetic data generation and large datacenter workloads.

He said Tesla’s access to real-world driving video that scales continuously gives the company an important data advantage for training Chinchilla-optimal FSD models. Baker added that Tesla’s coherent GPU clusters, alongside those of xAI, offer significant capital efficiency and remain unmatched by competitors in coherence, speed, and cost structure.

Baker Highlights Tesla’s Capital Efficiency And Data Advantage

Baker described Tesla as one of the lowest-cost producers of AI “tokens” relevant to FSD and robotics due to its fleet-generated data, customer-subsidized compute, and reduced reliance on synthetic data. He argued that rivals with less efficient training and inference pipelines, including those that rely heavily on synthetic or third-party datasets, face higher capital requirements and slower scaling dynamics.

Musk Confirms ‘Cortex 2’ Training Buildout

Elon Musk entered the exchange to endorse Baker’s assessment. “Exactly,” he wrote on X, responding to Baker’s argument that training had not been the limiting factor for Tesla’s FSD development so far. 

He added that this is now changing with the company’s humanoid robot program, Optimus. “It will be that for Optimus, so we’re building Cortex 2 for training,” Musk said, revealing that Tesla is developing a new AI training system tailored for robotics workloads rather than FSD.

'A Few Months Away' From Unsupervised FSD

Chanos’s criticism follows Musk’s comments at Tesla’s annual shareholder meeting, where shareholders approved his $1 trillion compensation package. Musk reiterated Tesla was only “a few months away” from delivering unsupervised FSD, effectively shifting his latest autonomy timeline into 2026.

Musk also said Tesla plans to allow drivers to “text and drive” while using supervised FSD “in a month or two,” without addressing how the company would comply with laws prohibiting phone use behind the wheel.

The remarks suggest Tesla may be preparing to classify future versions of FSD as Level 3 or Level 4 autonomous systems, which would require Tesla, not the driver, to assume responsibility when the system is active. Today, FSD remains at Level 2, and Tesla’s hands-off trials have been limited to internal robotaxi pilots in controlled environments.

Stocktwits Shows Muted Retail Mood

On Stocktwits, retail sentiment for Tesla was ‘neutral’ amid ‘normal’ message volume.

Tesla’s stock has risen 0.1% so far in 2025, among the worst performers on the S&P 500 and the “Magnificent Seven” group of stocks.

For updates and corrections, email newsroom[at]stocktwits[dot]com.

Share

·

Add us on